EPA-600/R-97-010
February 1997
FUZZY LOGIC BASED INTELLIGENT CONTROL
OF A VARIABLE SPEED CAGE MACHINE
WIND GENERATION SYSTEM
FINAL REPORT
Bimal K. Bose
Marcelo Godoy Simdes
The University of Tennessee, Knoxville
Department of Electrical Engineering
Knoxville. Tennessee 37996
EPA Cooperative Agreement CR8205S5
EPA Project Manager: Ronald J. Spiegel
National Risk Management Research Laboratory
Research Triangle Park, North Carolina 27711
Prepared for:
U.S. Environmental Protection Agency
Office of Research and Development
Washington , DC 20460
-------
4. TITLE AND SUBTITLE
Fuzzy Logic Based Intelligent Control of a Variable
Speed Cage Machine Wind Generation Machine
TECHNICAL REPORT DATA
(Plane raid Immtctions on the reverse before comph
1. REPORT NO.
EPA-600/R-97-01Q
5- REPORT DATE
February 1997
«. PERFORMING ORGANIZATION CODE
in i urn mill i mi
PB97-144851
7. AUTHOR(S)
BimalK. Bose and Marcelo G. Simoes
B. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND AOORESS
The University of Tennessee, Knoxville
Department of Chemical Engineering
Knoxville, Tennessee 37996
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
CR 820555
12. SPONSORING AGENCY NAME AND ADDRESS
EPA, Office of Research and Development
Air Pollution Prevention and Control Division
Research Triangle Park, NC 27711
13. TYPE OF REPORT ANO PERIOO COVERED
Final; 9/93 - 5/96
14. SPONSORING AGENCY COOE
EPA/6Q0/13
is. supplementary notes project officer is Ronald J. Spiegel, Mail Drop 63, 919/
541-7542.
is. abstract rj-he rep0rt gives results of a demonstration of the successful application of
fuzzy logic to enhance the performance and control of a variable-speed wind genera-
tion system. A squirrel cage induction generator feeds the power to a double-sided
pulse-width modulation converter system which pumps power to either a utility grid
or an autonomous system. Maximum power point tracker control is performed with
three fuzzy controllers, without wind velocity measurement. One fuzzy logic con-
troller (FLC-i) searches the generator speed on-line to optimize the aerodynamic
efficiency of the wind turbine. A second fuzzy controller (FLC-2) programs the
machine flux by on-line search so as to optimize the machine-converter system effi-
ciency. A third fuzzy controller (FLC-3) performs robust speed control against tur-
bine oscillatory torque and wind vortex. Detailed analysis and simulation studies
were performed for development of the control strategy and fuzzy algorithms, and
DSP TMS320C30-based hardware with C control software was built for the perfor-
mance evaluation of a laboratory experimental setup. The theoretical development
was fully validated, and the system is ready to be reproduced in a higher power in-
stallation.
17.
KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS
b.IDENTIF1ERS/OPEN ENDED TERMS
c. cosati Field/Group
Pollution
Artificial Intelligence
Electric Controllers
Energy
Electric Generators
Wind Power Generation
Pollution Prevention
Stationary Sources
Fuzzy Logic
13B
Q6A
09E, 14B
14 G
10B
IDA
IB. DISTRIBUTION STATEMENT
Release to Public
19. SECURITY CLASS (ThisReport/
Unclassified
21. NO. OF PAGES
170
20.SECURITY CLASS (Thispage)
Unclassified
22. PRICE
EPA Form 2220 1 (9-73)
-------
NOTICE
This document has been reviewed in accordance with U.S.
Environmental Protection Agency policy and approved for publication.
Mention of trade names or commercial products does not constitute
endorsement or recommendation for use.
ii
-------
FOREWORD
The 13. S. Environmental Protection Agency is charged by Congress with pro-
tecting the Nation's land, air, and water resources. Under a mandate of national
environmental laws, the Agency strives to formulate and implement actions lead-
ing to a compatible balance between human activities and toe ability of natural
systems to support and nurture life. To meet this mandate, EPA's research
program is providing data and technical support for solving environmental pro-
blems today and building a science knowledge base necessary to manage our eco-
logical resources wisely, understand how pollutants affect our health, and pre-
vent or reduce environmental risks in the future.
The National Risk Management Research Laboratory is the Agency's center for
investigation of technological and management approaches for reducing risks
from threats to human health and the environment. The focus of the Laboratory's
research program is on methods for the prevention and control of pollution to air,
land, water, and subsurface resources; protection of water quality in public water
systems; remediation of contaminated sites and groundwater; and prevention and
control of indoor air pollution. The goal of this research effort is to catalyze
development and implementation of innovative, cost-effective environmental
technologies; develop scientific and engineering information needed by EPA to
support regulatory and policy decisions; and provide technical support and infor-
mation transfer to ensure effective implementation of environmental regulations
and strategies.
This publication has been produced as part of the Laboratory's strategic long-
term research plan. It is published and made available by EPA's Office of Re-
search and Development to assist the user community and to link researchers
with their clients.
E. Timothy Oppelt, Director
National Risk Management Research Laboratory
i i i
-------
ABSTRACT
Wind energy is one of the most cost-competitive renewable energy technologies for the
bulk power market, with electricity costs in the range of 5 to 7 cents/kWh. Production costs
for wind energy are estimated to drop to around 4 cents/kWh by the year 2000, Since fossil
fuels are the primary energy source for the production of electricity in the U.S. and worldwide,
a reduction in power production from these sources can result in a concomitant reduction of
emissions of gases and pollutants which are considered to be acid rain precursors and
contributors to the greenhouse effect (global warming). The potential for enhanced
environmental quality can be realized if energy produced by environmentally clean technologies,
such as wind turbines, is used to offset energy produced by fossil fuels such as coal.
This work demonstrates the successful application of fuzzy logic to enhance the
performance and control of a variable speed wind generation system. A maximum power
point tracker control is performed with three fuzzy controllers, without wind velocity
measurement, and robust to wind vortex and turbine torque ripple.
A squirrel cage induction generator feeds the power to a double-sided pulse width
modulation converter system which pumps power to a utility grid or to an autonomous
system. One fuzzy logic controller (FLC-1) searches the generator speed on-line so that
the aerodynamic efficiency of the wind turbine is optimized. A second fuzzy controller
(FLC-2) programs the machine flux by on-line search so as to optimize the
machine-converter system efficiency. A third fuzzy controller (FLC-3) performs robust
speed control against turbine oscillatory torque and wind vortex.
Detailed analysis and simulation studies were performed for development of the
control 'strategy and fuzzy algorithms, and DSP TMS32QC30 based hardware with C
control software was built for the performance evaluation of a laboratory experimental
setup. The theoretical development was fully validated, and the system is ready to be
reproduced in a higher power installation.
iv
-------
TABLE OF CONTENTS
1. INTRODUCTION.... I
2. WIND GENERATION SYSTEM DESCRIPTION 2
2.1 Converter System...... 2
2.2 Turbine Modeling and Characteristics 4
2.3 Power Circuit Description 11
2.4 Control Strategy............ 15
3. FUZZY LOGIC CONTROL 22
3.1. Generator Speed Tracking Control (FLC-1) 22
3.2 Generator Flux Programming Control (FLC-2) 26
3.3 Closed Loop Generator Speed Control (FLC-3) ...,30
3.4 Control Coordination.. 33
4. SIMULATION STUDY .....35
5. HARDWARE DESCRIPTION... 49
5.1 Hanning PWM 54
5.2 Inverter Interface 56
5.3 IGBT Power Inverter 59
5.4 Speed Sensor Interface.... 59
5.5 Peripheral Circuits 61
6. SOFTWARE DESCRIPTION ...62
6.1 Control Software in the Processor Board #1 75
6.2 Control Software in the Processor Board #2 84
6.3 Host Communication Program.. 88
7. EXPERIMENTAL EVALUATION 90
v
-------
8. CONCLUSION AND IMPLEMENTATION ASPECTS
FOR A 200 kW SYSTEM 147
REFERENCES : 149
APPENDICES . 152
A. Inverter interface board layout 153
B. Fuzzy scaling for p.u. (per-unit) operation 154
vi
-------
LIST OF TABLES
1 Set of rules for FLC-1 ' 1 25
2 Fuzzy computation for scaling gains 26
3 Set of rules for FLC-2 30
4 Set of rules for FLC-3 ; 31
5 Induction machine and turbine parameters 35
6 Interrupt names, section names and vector locations 71
7 Memory mapped hardware addressees for DSP #1 76
8 Memory mapped hardware addressees for DSP #2 85
1
9 Wind energy system design parameters 90
10 Power enhancement due to fuzzy logic control 144
vi i
-------
LIST OF FIGURES
1 A voltage-fed double PWM converter wind generation system 3
2 Parcel of air flowing at speed Vw in the x direction 5
3 Polynomial function curve fitting of turbine power coefficient (Cp) 8
4 Plane of turbine rotor moving in respect to the air flow 9
©
5 Model of wind turbine with oscillatory torque 9
6 A typical family of torque/speed curves for a fixed pitch wind turbine 10
7 Fuzzy logic based control block diagram of wind generation system 13
8 Start-up sequencing of wind generation system 14
9 Indirect vector control (JVC) signal processing for machine-side inverter 16
10 Phasor diagram for line-side DVC, (a) 3-phase line phasors, (b) 2-phase line phasors,
(c) Signal voltage and current waves, (d) Signals in ds-qs and de-qe frames 17
11 Direct vector control (DVC) signal processing for line-side inverter 19
12 Stationary frame model for decoupling analysis .20
13 Fuzzy control FLC-1 and FLC-2 operation showing maximization of line power 22
14 Block diagram of fuzzy control FLC-1 23
15 Membership functions for fuzzy controller FLC-1 24
16 Membership function of speed for scale factor computation 25
17 Search method of efficiency optimization control of machine by flux programming... 27
18 Block diagram of fuzzy control FLC-2...... 28
19 Membership functions for fuzzy controller FLC-2 ; 29
20 Block diagram of fuzzy control FLC-3 31
21 Membership functions for fuzzy controller FLC-3.. 32
22 Control coordination diagram 34
23 Coefficient of performance of wind turbine 36
24 Turbine torque variation with tip-speed ratio : 37
25 Oscillatory train for wind turbine 38
26 Utility grid model... 40
VI .1 1
-------
27 Turbine and system model simulation curves (without fuzzy control): (a) Turbine
developed torque, (b) Turbine developed power, (c) Line side generated power 42
28 Speed optimization with wind velocity step-up 43
29 Speed optimization with wind velocity step-down 44
30 Flux excitation optimization by FLC-2 after FLC-1 has finished 45
31 Comparison of speed controller response by PI and FLC-3 for wind vortex 46
o
32 Step-up variation in the wind velocity, (a) Wind velocity, (b) Generator speed,
(c) Flux current, (d) Output power 47
33 Time domain operation of fuzzy controls FLC-1 and FLC-2 (FLC-3 is also working):
(a) Wind velocity, (b) Generator speed, (c) Flux current,
(d) Output power 48
34 Steady-state line side power boost with FLC-1 and FLC-2 control 50
35 Steady-state line side power boost for several wind velocities 50
36 Hardware block diagram for wind generation system , 52
37 Hanning PWM integrated circuit : .....53
38 DSPLINK Interface 55
39 Hanning PWM interface with cable driver,, 57
40 Inverter interface ! 58
41IGBT Power Inverter 60
42 Speed sensor circuit 61
43 Current sensor interface,. 63
44 Three phase transformation circuit with filtering 64
45 Dc-link voltage interface circuit 65
46 Dc-link current interface circuit 66
47 Dynamometer control interface .....67
48 Diagram for turbine torque emulation by dynamometer 68
49 Flow diagram for the multitasking real-time DSP code 73
50 Tasks coordination chart 74
51 Example of indexing for C program membership function evaluation 79
52 Host communication program screen interface 89
ix
-------
53 Wind energy system experimental set-up 91
54 Evaluation of Harming PWM signals with low-pass-filter at frequency of 50Hz. 96
55 Evaluation of Manning PWM signals with low-pass-filter and stepped frequency
between 25 Hz and 12.5 Hz 98
56 Dead-time on the upper and lower signals for the IGBT phase-leg device 99
57 Collector-to-emitter voltage switching of IGBT (top) and
phase-current (bottom) 99
58 Pulse-width-modulation of an IGBT phase-leg device, upper device (top) and
lower device (bottom) 100
59 Phase-voltage (top) and phase-current (bottom) for RL load in V/Hz 100
60 Three-phase currents in V/Hz showing the natural machine imbalance 101
61 Torque control loop, speed (top) at 1000 RPM (400 RPM/div) and torque
current (bottom) at 8A (6 A/div) with Te" = 4 Nm.,., 102
62 Three phase currents i„ ib, and i* (10 A/div) for torque loop control 102
63 Phase voltage va (100 V/div), arid phase current i8 (10 A/div)
for torque control loop ........103
64 Speed response of a PI controller, (top) speed response (200 RPM/div),
(bottom) control effect of the PI loop 103
65 Speed fuzzy control (FLC-3) with preliminary gains, (top) speed response
(200 RPM/div), (bottom) control effect of the FLC loop 105
66 Speed fuzzy control (FLC-3) with final gains, (top) speed response
(200 RPM/ div), (bottom) control effect of the FLC loop 105
67 Robustness of speed FLC with square-wave load torque, (top) speed at 400 RPM
(200 RPM/div), (bottom) load torque from 3.75 Nm to 6.25 Nm with 5 Nm average.. 106
68 Robustness of speed FLC with sinusoidal disturbance torque, (top) speed at
900 RPM (200 RPM/div), (bottom) load torque from 4.35 Nm to 5.65 Nm
with 5 Nm average 106
69 Robustness of speed FLC with square-wave load torque intentionally corrupted by
noise, (top) speed at 800 RPM (200 RPM/div), (bottom) load torque
from 3.75 Nm to 6 25 Nm with 5 Nm average 107
x
-------
70 Unit vector for line-side DVC inverter, sine and cosine waveforms 108
71 Sine waveform and reference angle for line side DVC inverter 108
72 SPWM line-side inverter phase-a voltage waveform voltage (100 V/div),
and line-side phase-a voltage (200 V/div) 110
73 Current control loop (1^*) for line-side inverter, (top) line-side voltage va
(150 V/div), (bottom) line-side current ia (5 A/div)... 110
74 Positive power command, line-side va voltage (60 V/div)
and current ia (3 A/div) 111
75 Negative power command, line-side va voltage (60 V/div)
and current i, (3 A/div) „• 111
76 Power reference P0* signal (bottom) at 400 W (15 W/div) and iqs' reference signal
(top) at 4A (0.5 A/div) 112
77 Raw phase-a inverter voltage on top (100 V/div), and raw phase-a current
on bottom (5 A/div) 112
78 Dc-link voltage transient response for a square wave command
from 240 V to 265 V 113
79 Dc-link voltage ripple of 1 V peak-to-peak, the average voltage is 300 V 113
80 Induction generator voltage (100 V/div) and current (10 A/div), raw signals, for
double-pwm converter operation with speed set to to, = 550 RPM and regenerative torque
constant at 3,5 Nm 115
81 Wind turbine static characteristics, (a) Turbine Power, (b) Turbine torque,
and (c) Generated Power 116
82 Transient in the wind with dc-link voltage response, (top) dc-link voltage at 300 V
(30 V/div), (bottom) wind velocity stepping up from 4.25 m/sec to 7.5 m/sec 118
83 Transient in the wind with dc-link voltage response, (top) dc-link voltage at 300 V
(30 V/div), (bottom) wind velocity stepping down from 9 m/sec to 7 m/sec 118
84 Fuzzy logic steady state performance enhancement control of wind turbine at
several operating points 119
85 Performance data for step-up wind velocity, (a) Wind velocity (m/sec), (b) Excitation
current ids (A), (c) Generator speed (RPM), (d) Generated power (W),
xi
-------
(e) Torque current % (A) 120
86 Power changing as generator speed goes up for wind velocity stepping-up
from 4.75 m/see to 7.25 m/sec 121
87 Performance data for step-down wind velocity, (a) Wind velocity (m/sec),
(b) Excitation current U* (A), (c) Generator speed (RPM), (d) Generated power (W),
(e) Torque current (A) 122
88 Power changing as generator speed goes down for wind velocity stepping-down
from 9 m/sec to 7 m/sec 123
89 First step of speed optimization by FLC-1 for wind velocity varying from
4 m/sec to 7 m/sec, speed reference is 500 RPM, generated power is 155 W,
(a) Line-side voltage v2 (120 V/div) and line-side current ia (0.5 A/div),
(b) Machine current ia (10 A/div) ' 125
90 Second step of speed optimization by FLC-1 for wind velocity varying from
4 m/'sec to 7 m/sec, speed reference is 625 RPM, generated power is 220 W,
(a) Line-side voltage va (120 V/div) and line-side current i, (0 5 A/div),
(b) Machine current ia(10 A/div) - - - 126
91 Third step of speed optimization by FLC-1 for wind velocity varying from
4 m/sec to 7 m/sec, speed reference is 750 RPM, generated power is 270 W,
(a) Line-side voltage va (120 V/div) and line-side current ia (0 5 A/div),
(b). Machine current ia (10 A/div) 127
92 First step of speed optimization by FLC-1 for wind velocity varying from
6 m/sec to 9 m/sec, speed reference is 700 RPM, generated power is 370 W,
(a) Line-side voltage va (120 V/div) and line-side current ia (2 A/div),
(b) Machine current ia (20 A/div) 128
93 Second step of speed optimization by FLC-1 for wind velocity varying from
6 m/sec to 9 m/sec, speed reference is 875 RPM, generated power is 575 W,
(a) Line-side voltage va (120 V/div) and line-side current i„ (2 A/div),
(b) Machine current ia(20 A/div) 129
94 Third step of speed optimization by FLC-1 for wind velocity varying from
6 m/sec to 9 m/sec, speed reference is 1050 RPM, generated power is 840 W,
xf i
-------
(a) Line-side voltage va (120 V/div) and line-side current ia (2 A/div),
(b) Machine current ia(20 A/div) 130
95 Fuzzy logic flux enhancement optimization of induction generator,
(a) Wind velocity (m/sec). (b) Generator Speed (RPM), (c) Generated Power (W),
(d) Excitation Current ids (A), (e) Torque current % (A) 132
96 First step of flux optimization by FLC-2 for wind velocity at 6 m/sec,
generator speed (cv = 623 RPM, (a) Machine voltage va (150 V/div) and
current i* (10 A/div), (b) Phasor diagram for such operating conditions 133
97 Second step of flux optimization by FLC-2 for wind velocity at 6 m/sec,
generator speed ov = 623 RPM, (a) Machine voltage va (150 V/div) and
current i, (10 A/div), (b) Phasor diagram for such operating conditions 134
98 Third step of flux optimization by FLC-2 for wind velocity at 6 m/sec,
generator speed Or = 623 RPM, (a) Machine voltage v, (150 V/div) and
current ia (10 A/div), (b) Phasor diagram for such operating conditions 135
99 Wind transient step-up from 4 75 m/sec to 7.25 m/sec, (a) Wind velocity (m/sec),
(b) Generator speed (RPM), (c) Line-side generated power (W),
(d) De-link voltage (V) .136
100 Wind transient step-down from 9 m/sec to 7 m/sec, (a) Wind velocity (m/sec),
(b) Generator speed (RPM), (c) Line-side generated power (W),
(d) De-link voltage (V) 138
101 Wind transient ramp-up profile from 6.125 m/sec to 9 125 m/sec, (a) Wind velocity
(m/sec), (b) Generator speed (RPM), (c) Line-side generated power (W),
(d) De-link voltage (V)...- 139
102 Flux excitation current programming by FLC-2, (a) Wind velocity at 5 m/sec,
(b) Generator speed at 520 RPM, (c) Excitation current i* (A),
(d) Line-side generated power (W) 140
103 Flux excitation current programming by FLC-2, (a) Wind velocity at 6 5 m/sec,
(b) Generator speed at 675 RPM, (c) Excitation current i^ (A),
(d) Line-side generated power (W) 141
104 Time domain sequencing of FLC-1 and FLC-2, (a) Wind velocity sinusoidally
xi i i
-------
going down from 9.35 m/sec to 6.35 m/sec with wind vortex,(b) Generator(RPM),
(c) Excitation current i
-------
1. Introduction
Wind energy is one of the most cosl-competitive renewable energy technologies for the
bulk power market, with electricity costs in the range of 5 to 7 cents/kWh. Since fossil fuels
are the primary energy source for the production of electricity in the U.S. and worldwide, a
reduction in power production from these sources can result in a concomitant reduction of
emissions of gases and pollutants which are considered to be acid rain precursors and
contributors to the greenhouse effect (global warming). The potential for enhanced
environmental quality can be realized if energy produced by environmentally clean technologies,
such as wind turbines, is used to offset energy produced by fossil fuels such as coal.
The world has enormous resources of wind power. It has been estimated that, even if 10%
of raw wind potential could be put to use, all the electricity needs of the world would be met
[1]. There are currently over 1700 MW of wind generators installed worldwide with
generation of 6 billion kWh of energy annually. It has been estimated the generation will grow
to 60 billion kWh by the year 2000, with production costs to drop to around 4 cents/kWh.
Traditionally, wind generation systems used variable pitch constant speed wind
turbines (horizontal or vertical axis) that were coupled to squirrel cage induction
generators or wound-field synchronous generators and fed power to utility grids or
autonomous loads. The recent evolution of power semiconductors and variable frequency
drives technology has aided the acceptance of variable speed generation systems. In spite
of the additional cost of power electronics and control, the total energy capture in a
variable speed wind turbine (VSWT) system is larger and, therefore, the life-cycle cost is
lower The following generator-converter systems have been popularly used [2] [3] [4];
• 'Doubly fed induction generator with cascaded converter slip power recovery.
* Doubly fed induction generator with cycloconverter slip power recovery.
1
-------
• Synchronous generator with line-commutated and load-commutated thynstor
converters.
In addition to the above schemes, squirrel cage generators with shunt passive or
active VAR (volt ampere reactive) generators [5] [6] have been proposed which generate
constant voltage constant frequency power through a diode rectifier and line-commutated
thyristor inverter Recently, a variable reluctance machine [7] and doubly stator-fed
induction machine [8] have also been proposed in wind generation systems. The major
problems in traditional power conversion schemes are the poor line power factor and
harmonic distortion in line and machine currents. The recent IEEE Standard 519 [9]
severely restricts line harmonic injection. Therefore, to satisfy the stringent harmonic
standard and poor power factor problem, active type VAR and harmonic compensators
can be installed with large additional cost, Again, the conventional control principles used
in these systems make the response sluggish and give non-optimum performance. Very
recently, a double-sided pulse width modulated (PWM) converter system has been
proposed to overcome some of the above problems.
This work describes a VSWT system with squirrel cage induction generator and
double-sided PWM converter where fuzzy logic control has been used extensively to
maximize the power output and enhance system performance. All the control algorithms
have been validated by simulation study, and system performance has been evaluated in
detail and an experimental study with a 3 .5 kW laboratory drive system was constructed to
evaluate the performance.
2, Wind Generation System Description
2.1 Converter System
The voltage-fed converter scheme used in this system is shown in Fig. 1. A vertical
(or horizontal) wind turbine is coupled to the shaft of a squirrel cage induction generator
through a speed-up gear ratio The variable frequency variable voltage power
2
-------
WIND
TURBINE
SQUIRREL CAGE
INDUCTION —
GENERATOR
UTILITY
BUS
IGBT IGBT
FWM RECTIFIER PWM INVERTER
Fig. 1 A voltage-fed double PWM converter wind generation system.
from the generator is rectified by a PWM IGBT (insulated gate bipolar transistor) rectifier.
The rectifier also supplies the excitation need of the machine. The inverter topology is
identical to that of the rectifier, and it supplies the generated power at 60 Hz to the utility
grid. Salient advantages of the converter system include:
• Line side power factor is unity with no harmonic current injection (satisfies IEEE
519).
• The cage type induction machine is extremely rugged, reliable, economical, and
universally popular.
• Machine current is sinusoidal — no harmonic copper loss.
• Rectifier can generate programmable excitation for the machine.
• Continuous power generation from zero to highest turbine speed is possible.,
3
-------
• Power can flow in either direction permitting the generator to run as a motor for
start-up (required for vertical turbine). Similarly, regenerative braking can quickly stop
the turbine
• Autonomous operation of the system is possible with a start-up capacitor charging the
battery.
• Fast transient response is possible,
• Multiple generators or multiple systems can be operated in parallel.
Considering all the above advantages, and with the present trend of decreasing
converter and control cost, this type of conversion system has the potential to be
universally accepted in the future. Of course, in recent years, soft-switched resonant link
and resonant pole topologies have been proposed, but additional R & D are needed to
bring them to the market-place.
2.2 Turbine Modeling and Characteristics
Both horizontal and vertical axis wind turbines are used in wind generation
systems The vertical axis wind turbine (VAWT) type Darrieus, popularly called by egg
beater due the blades configuration, has the advantages that it is located on the ground,
can accept wind from any direction without any special yaw mechanism and, therefore, is
preferred for high power output. The disadvantages are that the turbine is not self-starting
and there is a large pulsating torque which depends on wind velocity, turbine speed, and
other factors related to the design of the turbine. This type of turbine has fixed pitch
blades, but in case of very high wind velocity, the pitch angle can be adjusted for safety.
A model for the turbine was developed for simulation for the whole wind
generation system in order to evaluate the fuzzy control features. Two peculiarities were
explored in such model: (1) the VAWT model should have efficiency relatively to the
torque and speed region of operation, and (2) the VAWT should have intrinsic pulsating
torque.
4
-------
pig. 2 represents a parcel of air with mass m, flowing at speed Vw in the x
direction.
zt/ Air
1 Flow
Fig, 2 Parcel of air flowing at speed Vw in the x direction.
The kinetic energy is given by the following equation:
Et=im(Vw)2=ipAx(Vw)2 (1)
where A = cross sectional area of the air parcel moving in the x direction
p = air density
The power in the wind (Pw) is the time derivative of the kinetic energy as
Pw=xE*4pa3 <2>
at 2
The presence of a wind turbine modifies the local air speed and pressure. The
speed of the air decreases as the turbine is approached causing the tube of air to enlarge to
the turbine diameter. The air pressure rises to maximum just in front of the turbine and
drops below atmospheric pressure behind the turbine. It can be shown that the mechanical
power extracted in such conditions is [10]:
P,urt,,ne,,d«J=5p(-^)A(Vw)3 (3)
5
-------
Therefore, an actual turbine cannot extract more than 59.3% of the power in an
undisturbed tube of air of the same are as the turbine. In practice, the fraction of power is
always less because of mechanical characteristics. Such fraction of power is indicated by
the symbol Cp, standing for coefficient of performance. Using this notation the actual
mechanical power output can be written as:
Pm = CpPw (4)
The coefficient of performance (Cp) is not constant, but varies with the wind
speed, rotational speed of the turbine and turbine blade parameters such as angle of attack
and pitch angle. Cp is defined as a non-linear function of the tip-speed-ratio (X) as
(5)
where. X - twC°w
V,
w
The torque at the output shaft is
Tm=^L = -7-(cpWPm)
«in
(6)
The turbine is coupled to the induction generator (IG) through a step-up gear box
(1:t|gear) so that the IG runs at a higher rotational speed (Oa despite the low speed to* of
the wind turbine. Therefore, the aerodynamic torque of a vertical turbine is given by the
equation
Tm=Cp(X).
0.5
pnR&
"Hgear.
¦v!
w
(7)
6
-------
where
Cp = power coefficient
X = tip speed ratio (TSR) (
V w
p = air density
Rw = turbine radius
t1GEAr= speed-up gear ratio
Vw = wind velocity
ow = turbine angular speed
The power coefficient for this project was taken from the characteristics of a
Sandia-17 m turbine [3], A polynomial curve was fitted as indicated in Fig. 3.
The static torque as given by (7) is the average torque production which must be
added to the intrinsic torque pulsation. An actual turbine is quite complex to model since
the moving surface of the turbine slips in respect to the wind and there are also extra
components of wind in the x, y and z directions. The blade which moves with the wind
during one half of its revolution, moves against the wind during the other half revolution
and some turbulence occurs behind the moving surface. There are various inertias in the
system as rotor blades, hub, gearbox, shaft and generator and damping caused by the wind
itself and the oil in the gearbox The most important source of torque pulsation taken in
account for the present model is that of the rotor blades passing by the tower. Fig. 4
shows how the plane of the turbine rotor moves in respect to the wind. If the angle 0 =
90° the plane of the rotor is parallel to the wind and there is a peak of torque. If the
turbine had just one
7
-------
y ¦ a+bs.+ad+dx* + tx* + ft?++hx?
8=0.001001018 IH3.0017591S34 c=0.0030428053 (1=0.0041214335
c=0.0025649299 £=-0.0011473798 g*0.00013883836 b=^5.4415739&06
0.40
^ 0.35
U
H 0.30
0.25
£ 0.20
a
0 0.15
El o.to
O 0.05
0.00
TIP SPEED RATIO (X)
Fig. 3 Polynomial function curve fitting of turbine power coefficient (Cy.
blade, there would be one pulsation per revolution. A rotor with two identical blades has
the lowest pulsation frequency equal to two pulses per revolution. In practice, the two
blades are never identical so the fundamental pulsation also appears.
The oscillatory torque of the turbine is more dominant at the first, second,
and fourth harmonics of fundamental turbine angular velocity (cow ) and is given by the
expression
TOK=Tm-[Acos({ow)+Bcos{2cow)+Ccos(4cow)] (8)
where A, B, and C are constants.
Fig. 5 shows the block diagram of the turbine model with oscillatory torque. A
typical family of turbine torque/speed curves at different wind velocities is shown in Fig. 6.
Superimposed on the family of curves is a set of constant power lines indicating
8
-------
Fig. 4 Plane of turbine rotor moving in respect to the air flow.
TURBINE TORQUE
WIND
POLES
V
Taie=T.[Acos(wwt) + Bcos(2w,t) + Ccos(4w.f
WIND
VELOCITY
Fig. S Model of wind turbine with oscillatory torque.
9
-------
the region of maximum power delivery for each wind speed. This means that, for a
particular wind speed, the turbine speed (or the TSR) is to be varied to get the maximum
power output, and this point deviates from the maximum torque point, as indicated. Since
the torque/speed characteristics of the wind generation system are analogous to those of a
motor-blower system (except the turbine runs in reverse direction), the torque follows the
square-law characteristics (Te=K(i)J) and the output power follows the cube-law
(p0=KtoJ), as indicated in Fig. 6. This means that, at reduced speed light load steady state
conditions, generator efficiency can be improved by programming the flux [11] which will
be discussed later.
UJ
3
LOCUS OF MAXIMUM
POWER DELIVERY
o
cc
o
i_
UJ
z
£0
DC
z>
I—
TURBINE TORQUE/SPEED
CURVES FOR INCREASING
WIND SPEEO •
TURBINE ROTATIONAL SPEEO
Fig, 6 A typical family of torque/speed curves for a fixed pitch wind turbine.
10
-------
2.3 Power Circuit Description
? .
Fig. 7 shows the control block diagram of the system that uses the power circuit of
Fig. 1. The machine and inverter output currents are sinusoidal, as shown. The machine
absorbs lagging reactive current, but the reactive current is always zero on the line side,
i.e., the line power factor is unity The rectifier uses indirect vector control in the inner
current control loop, whereas the direct vector control method is used for the inverter
current controller. Vector control permits fast transient response of the system. The fuzzy
controllers are described in the Section 3. For a particular wind velocity (Vw), there is an
optimum setting of generator speed (©/). The speed loop control generates the torque
component of machine current so as to balance the developed torque with the load torque.
The variable voltage variable frequency power from the supersynchronous induction
generator is rectified and pumped to the dc link. The dc link voltage controller regulates
the line power P0 (i e, the line active current) so that the link voltage always remains
constant A feedforward power signal from the machine output to the dc voltage loop
prevents transient fluctuation of link voltage. There is a local inductance L* connected
between the line-side inverter output and the three-phase utility, bus. Such inductance is
very important for stable operation of the synchronous current controller [12] and it is
selected in such a way that the maximum modulation index at which the line-side inverter
operates is as close to one as is permitted by the minimum pulse or notch widht capability
of the device. Under these conditions the control varies over a wide range making the
control less sensitive to the errors in the controller gains and compensation. The
inductance value is such that the slope of the output of the PI is smaller than the slope of
the triangular carrier of the SPWM [12] in accordance to the equation (9). In this wind
energy system, the value of the inductance was fine tuned by trial-and-error in the
simulation study, and an inductance with alloy core was built for the experimental
evaluation with the value used in the simulation.
« VV + 0 5v,i .
2rfsiqs a < 2fsvm (9)
11
-------
where:
Ls = local inductance
vac = line-side input voltage
Vd = dc-link voltage
fs = frequency of the triangular carrier frequency
vm = amplitude of the triangular carrier frequency
iqS = active power current
The system can be satisfactorily controlled for start-up and regenerative braking
shut-down modes besides the usual generating mode of operation. The low-chart in Fig, 8
shows the procedure for the initial start-up of the wind turbine. Both inverters are initially
disabled during the charging of the dc-link capacitor, i.e. all the gate pulses are off. The
capacitor is charged from the diodes on the line-side inverter with the peak value of the
line voltage. To prevent the damage of the diodes there is a series resistance for the
capacitor initial charging (R*) in the dc-link. Such resistance is by-passed by an
electromagnetic relay when the bus voltage reaches 95 % of the line voltage. When the
capacitor has been charged, the dc-link voltage loop control is exercised (the pulse gates
are enabled, and the control loops are activated) to establish a smaller voltage than the
peak value in the bus (typically 75 % of the peak-value), therefore, successfully operating
the line-side inverter in PWM mode.
With the dc-link bus voltage fixed, the induction generator can be excited with ids,
so as the rated flux is established in the machine. Next, a speed reference is commanded to
rotate the turbine with the minimum required turbine speed to catch some power from
wind (the fuzzy speed controller FLC-3 is always working). The flow of the wind imposes
a regenerative torque in the machine, of course the power generation is not optimum yet,
but the slip frequency becomes negative and the power starts to flow from the turbine
towards the line-side. As the power starts to flow to the line-side, the dc-link voltage loop
control can be gradually commanded to a higher value than the peak value of the line-side
voltage (typically 75 % higher). After the dc-link voltage is established in the new higher
value, the system is ready to be controlled by the fuzzy controllers FLC-1 and FLC-2
12
-------
SPWM
MOD.
SIGNAL
SPWM
MOD.
SIGNAL
SYNCHRONOUS
CURRENT CONTROL WITH
DECOUPLER AND
VECTOR ROTATOR
SYNCHRONOUS
CURRENT CONTROL
AND
VECTOR ROTATOR
UV
UV
Po Calc,
FLC-2
FEEDFORWARD
POWER p
Aid,
FLC-1
FLC-3
Fig. 7 Fuzzy logic based control block diagram of wind generation system.
-------
ACTIVATE FLC-3
ACTIVATE MACHINE
CONTROL WITH RATED
FIELD
START TURBINE WITH
SPEED CONTROL
CLOSER, BYPASS
CONTACTOR
CIRCUIT BREAKER
CLOSE LINE
DC LINK CAPACITOR
CHARGES THROUGH R*
MOTORING AND THEN
REGENERATION
ACTIVATE Vd CONTROL
LOOP WHEN V„>Vd mi8
SYSTEM AT STEADY STATE
ACTIVATE FLC-1 WHEN
Fig. 8 Start-up sequencing of wind generation system
14
-------
2.4 Control Strategy
The machine-side inverter uses indirect-vector-control (IVC), With the control of
vector currents and the rotor flux is aligned with the d-current i^ [13], With vector
control there is no fear of instability problem and a four-quadrant operation, including zero
speed is possible. The drive has a dc machine like transient response; there is no direct
frequency control, the frequency is controlled through the unit vectors generated by the
addition of the machine speed (
-------
sin0,
COS0,
Pulses for
IGBT inverter
:*
sin/cos
table
VR
Hanning
PWM
0.
Fig. 9 Indirect vector control (IVC) signal processing for machine-side inverter
-------
a)
b)
sin9,
V*/ COS0
C)
d)
Fig. 10 Phasor diagram for line-side DVC, (a) 3-phase line phasors, (b) 2-phase line
phasors, (c) Signal voltage and current waves, (d) Signals in d'-qs and de-qe frames
17,
-------
The equations used in the line-side DVC strategy are given below:
vqs = va
Vds = ^|(vc-Vb)_
COS0e
sin9e =
'qs
vls
K) +(vi)
Po - (vqs iqs + vds 'ds)
ee = J
/ \
^vqs vds ~ ^vds Vqs
K)j+(vL)2
(12)
(13)
(14)
(15)
(16)
(17)
The vector control block diagram for the line side inverter is shown in Fig. 11.
There is a decoupling network connected to the PI controllers. It can be shown that series
inductance in the line creates coupling effect, i e , if i^ is changed, it creates voltage drop
(©Li^) on the q-axis, or if i^ is changed a drop is created on the d-axis. This cross-
coupling effect slows down the response of the inverter to establish line % (active
current). This coupling can be cancelled by the feedforward terms coLid, in the iq, loop and
©Liqs in the i^ loop, where K = ©L, the additional feedforward emf (vv) in the i<,s loop
cancels the counter emf effect and enhances the response of the % loop The v4s loop does
not need this compensation since the reactive current is kept to be zero.
The analysis of such decoupling network is made with the stationary frame model
given in Fig, 12. It is assumed here that the resistance is very small and can be neglected.
18
-------
Unit Vector
Calculation
VR-
cos 0,
Harming
PWM
Pulses for
IGBT inverter
Calculation
Fig. 11 Direct vector control (DVC) signal processing for line-side inverter
19
-------
The stationary frame model given in Fig. 12 shows the impressed inverter voltages,
the quadrature currents and emfs from the line-side.
qs
qs
o
vqs - ^s'qs+ sLs,qs + eqs
' dS
rm
"ds
0
¦'ds = + sLsi|s + eds
Fig. 12 Stationary frame model for decoupling analysis
After vector-rotation with the unit-vector the equations (18) and (19) follow.
* ^qs ,
Vqs= dt +G,e ^
vds = ^ ^e'J'qs
(18)
(19)
Since the derivative of the quadrature flux is the quadrature voltage, the following
(20) and (21) hold And the matricial form of the voltages, currents and emfs is presented
in (22). Since the power factor is unity, e^ = 0, and the resulting control law is given in
(23).
20
-------
= v = e +Lcd
'qs
dt
qs "qs » dt
d*ds +L d]ds
dt ® m 5 dt
(20)
(21)
vqs
vds
sL g g L
-coeLs
sL
s J
'qs
jds
+
cqs
_eds.
(22)
vqs*
Cqs
+
,Vds\
°
(A)
0
® e^slf *qs
-coeLs 0
t
(B)
+
Jds.
sLs 0 T'qs
0 sLs
t
(C)
(23)
sLs 0 "
vL
^-s ® T'^'qs
~PI(s)
0 "
'qs ~ 'qs
0 sLs^
Jds J
- ^ ^sjL^ds.
= 0
PI(s)
Jds ~ *ds.
(24)
The controller in Fig. 11 shows the implementation of the matricial equation (23),
The terms (A) and (B) are provided by feedforward connections, whereas the term (C) is
provided by the synchronous PI current controller, as indicated by equation (24).
21
-------
3. Fuzzy Logic Control
The system has three fuzzy logic controllers for performance enhancement and
control.
3,1 Generator Speed Tracking Control (FLC-1)
Since the power is given by the product of torque and speed, and turbine power
equals the line power (assuming steady state lossless system), the turbine torque/speed
curves of Fig, 6 can be translated to line power (P0) - generator speed (cCr) curves, as
shown in Fig. 13. For a particular value of wind velocity, the function of fuzzy controller
FLC-1 is to search the generator speed until the system settles down at the maximum
output power condition. For wind velocity of VW4 in Fig. 13, the output power will be at
A if the generator speed is 0\i- The FLC-1 will alter the speed in steps until it reaches
PB JUMP BY
V. CHANGE
FLC-2
FLC-2
~4
WIND
VELOCITY
wi
GENERATOR SPEED (©,)
Fig. 13 Fuzzy control FLC-1 and FLC-2 operation showing maximization of line power.
the speed o\2 where the output power is maximum at B. If the wind velocity increases to
Vw2. the output power will jump to D, and then FLC-1 will bring the operating point to E
22
-------
by searching the speed to ©*. The profile for decrease of wind velocity to Vwj is also
indicated. The principle of the fuzzy controller is explained in the block diagram of Fig.
14. With an incrementation (or decrementation) of speed, the corresponding
incrementation (or decrementation) of output power P0is estimated. If AP0 is positive with
KPO
Ms,
WJLE
BASE
FUZZ3FICATI0N
DEFUZZmCAHON
EVALUATION of
CONTROL RULES
FUZZY COMPUTATION
of scaling GAINS
Fig. 14 Block diagram of fuzzy control FLC-1.
last positive AtOr, the search is continued in the same direction. If, on the other hand, +&mr
causes -APo, the direction of search is reversed. The variables AP0 (variation of power),
Ac*. (variation of speed) and LA©r (last variation of speed) are described by membership
functions given in Fig. 15 and the inference is given by the set of rules in Table 1. In the
implementation of fuzzy control, the input variables are fiizzified, the valid control rules
are evaluated and combined, and finally the output is defuzzified to convert to the crispy
value. The output A©, is added to some amount of LA©,, in order to avoid local minima
due to wind vortex and torque ripple. The controller operates on a per-unit basis so that
the response is insensitive to system variables and the algorithm is universal to any system.
-------
M. F for
'rfUSI}
M. F for
i(outpui)
Fig. IS Membership functions for fuzzy controller FLC-1
24
-------
Table 1 Set of rules for FLC-1
apd
P
ZE
N
PVB
PVB
PVB
NVB
PBIG
PBIG
PVB
NBIG
PMED
PMED
PBIG
NMED
PSMA
PSMA
PMED
NSMA
ZE
ZE
ZE
ZE
NSMA »
NSMA
NMED
PSMA
NMED .•
NMED
NBIG
PMED
NBIG
NBIG
NVB
PBIG
NYB
NVB
NVB
PVB
The scale factors KPO and KWR, as shown in Fig 14, are function of generator
speed so that the control becomes somewhat insensitive to speed variation. The scale
factor is generated by fuzzy computation. The speed is first evaluated into seven fuzzy sets
as shown in Fig 16. The scale factors KPO and KWR are generated in accordance to the
rules in Table 2.
1.0
0.8
M. F for 0.6
Speed
0.4
0.2
0.0
0,0
w ^ m
to & on
N e- 0-
Uj
0-
£22
CL
wd 03
pa eta
cu cm
\ /
\ /
\ /
\ /
0.2
0.4
0.6
0.8
1.0
Fig. 16 Membership function of speed for scale factor computation
25
-------
Table 2 Fuzzy computation for scaling gains
1
KPO
KWR |
PSS
40
' 25
PSB
210
40
PMS
300
40
PMB
375
50
PBS
470
50
PBB
540
60
The advantages of fuzzy control are obvious. It provides adaptive step size in the
search that leads to fast convergence, and the controller can accept inaccurate and noisy
signals. The FLC-1 operation does not need any wind velocity information, and its real
time based search is insensitive to system parameter variation
3.2 Generator Flux Programming Control (FLC-2)
Since most of the time the generator is running at light load, the machine rotor flux
ids can be reduced from the rated value to reduce the core loss and thereby increase the
machine-converter system efficiency [11]. The principle of on-line search based flux
programming control by a second fuzzy controller FLC-2 is explained in Fig. 17, At a
certain wind velocity Vw and at the corresponding optimum speed established by
FLC-1 (which operates at rated flux the rotor flux vj/R is reduced by decreasing the
magnetizing current i*. This causes increasing torque current i^ by the speed loop for the
same developed torque As the flux is decreased, the machine iron loss decreases with the
attendant increase of copper loss. However, the total system (converters and machine)
loss decreases, resulting in an increase of total generated power P0. The search is
continued until the system settles down at the maximum power point, as indicated in Fig.
17. Any attempt to search beyond point A will force the controller to return to the
maximum power point The principle of fuzzy controller FLC-2 is somewhat similar to
that of FLC-1 and is explained in Fig 18. The system output power P0(k) is sampled and
26
-------
compared with the previous value to determine the increment AP0. III addition, the last
excitation current decrement (LAi*) is reviewed. The membership functions for variation
of power (AP0), last variation of i*. (LAi^) and change in y, (Ai^) are given in Fig. 19. On
these bases, the decrement step of i^ is generated from fuzzy rules given in table 3
through fuzzy inference and defuzzification, as indicated.
TORQUE
TURBINE SPEEO
ZONE OF >
OPERATION
OUTPUT POWER
STATOR CURRENT
AIRGAP FLUX
FLUX COMPONENT
—t , CURRENT
TIME
TOTAL LOSS
COPPER LOSS
CONVERTER
LOSS
IRON LOSS
TIME
EFFICIENCY OPTIMIZED
OPERATING POINT
Fig. 17 Search method of efficiency optimization control of machine by flux programming
27
-------
KIDS
KP
F„(k) APa(k)^ APB(k),
,(k-l)
SCALING
FACTORS
COMPUTATION
ITfT77V
* U m m* X
INFERENCE AND
DEFUZZIFICATION
z1
Fig. 18 Block diagram of fuzzy control FLC-2
The adjustable gains KP and KIDS which convert the actual variable to per unit
variables are given by the respective expressions
KP = acot + b (25)
KIDS = CiCflt - c*T e + cj (26)
where a, b, clf ci, and C3 are derived from simulation studies. The effect of controller
FLC-2 in boosting the power output was shown in Fig. 13. Hie FLC-2 controller
operation starts when FLC-1 has completed its search at the rated flux condition. If wind
velocity changes during or at the end of FLC-2, its operation is abandoned, the rated flux
is established, and FLC-1 control is activated.
28
-------
M. F for
^*OAST)
-0.01 0.0 0.01
M. F for
AP„
M. F for
AL
Fig. 19 Membership functions for fuzzy controller FLC-2
29
-------
Table 3 Set of rales for FLC-2
Ai^ast)
AP0
N
P
PB | MM
PM
PM
NS
PS
PS
NS
PS
NS
PS
NS
NM
PM
NM
NB
PB
NB
3 J Closed Loop Generator Speed Control (FLC-3)
The speed loop control is provided by fuzzy controller FLC-3, as indicated in the
block diagram of Fig. 7. As mentioned before, it basically provides robust speed control
against wind vortex and turbine oscillatory torque. The disturbance torque on the machine
shaft is inversely modulated with the developed torque to attenuate modulation of output
power and prevent any possible mechanical resonance effect. In addition, the speed
control loop provides a deadbeat type response when an increment of speed is commanded
by FLC-1. Fig 20 explains the proportional-integral (PI) type fuzzy control [15] used in the
system. The speed loop error (Ecor) and error change (AEcjr) signals are converted to per-unit
signals, processed through fuzzy control in accordance to the membership functions given in
Fig. 21 and the rule table 4. The output of the fuzzy controller FLC-1 is summed to produce
the generator torque reference Te\
30
-------
FUZZY
CONTROL
Fig. 20 Block diagram of fuzzy control FLC-3.
Table 4 Set of rules for FLC-3
E
CE
NVL
NL
NM
NS
ZE
PS
PM
PL
PVL
NVL j
' ;
NVL
NL
NM
NS
ZE
NL j
-
NL
NM
NS
ZE
PS
NM |
NL
NM
NS
ZE
PS
PM
MS j
NL
NM
NS
ZE
PS
PM
PL
1 ZE [
NL
NM
NS
ZE
PS
PM
PL
PS
1 NL
NM
NS
ZE
PS
PM
PL
:
j.:**;
PM
NM
NS
ZE
PS
PM
PL
mximwi
PL
NS
ZE
PS
PM
PL
asisgsssisi
PVL
ZE
PS
PM
PL
PVL
:
31
-------
M. F for
Error
mwm5
55 N fr. E
M. F for
Change in Error
M. F for
Change inTorque
Fig. 21 Membership functions for fuzzy controller FLC-3
32
-------
3.4 Control Coordination
While the speed fuzzy controller FLC-3 is always active during system operation,
the controllers FLC-1 and FLC-2 operate in sequence at steady (or small turbulence) wind
velocity. A start-up procedure is required for complete control activation and a shut-down
sequence is required in case of any fault. The control coordination diagram shown in Fig.
22 indicates the sequencing and the start-up/shut-down procedures. For the start-up the
line side circuit breaker is closed. The de-link capacitor charges to the peak value of the
line voltage through a series resistance which avoids the inrush charging current. After a
delay of 0.5 sec, the resistance is by-passed with a relay and the rated flux is imposed in
the induction machine. The turbine is started with speed control and as the power starts to
flow, the dc-link voltage rises. Hie dc-link voltage control is activated when the delink
voltage is above the limit value and FLC-1 starts to search the optimum speed reference
CD*". As ox* is altered the power generation goes up, until the FLC-1 settles down in the
steady-state condition, indicated by a small variation in |Acq*| with alternating polarity. At
this condition the system is transferred to FLC-2 in order to optimize the flux by
decreasing i^". During the speed search, any deviation from expected variation of power
indicates that the system is subject to large wind variation and the system is transferred
from FLC-1 to non-fuzzy operation, waiting for the transient to vanish. During the flux
optimization the excitation current (i**) is decreased adaptively by FLC-2. The control is
then transferred to the optimum operation when the variation |Au,"| is small with
alternating polarity. The search is finished and the optimum operation state keeps the
optimum a\" and iThe power and variation of power is recorded in order to sense if
there is any variation in the wind velocity to restart the search. Again, during the operation
of FLC-2, any load transient, indicated by the variation of the torque, transfers the system
to the non-fuzzy operation state, waiting for the transient to vanish in order to restart from
FLC-1.
The system has external fault indications from the turbine and inverters, which can
indicate a dangerous operation during too high wind velocity, or the wind velocity is too
low and the power cannot satisfactorily be generated. Hie inverters can indicate a tripping
-------
4*
•Search optimum a,'
T., - 1.24 sec
1.24 msec
\&®, I for 2 j*
Search optimum L
^ poUrity alters,
KeepWiM
•Enable FLC-2
Disable FLC-I
FLC-1
FLC-2
%
NON
FUZZY
OPERATION
SHUT
DOWN
START
OPTIMUM
Keep id,
•Keep l4' - lto"
•Record AP
START-IIP
OPERATION
¦Keep cd, andl*'
•Record P. and AP.
1) Close line circuit breaker
2) After 0.5 sec., close Rgbypass contactor
3) Activate machine control with rated field
4) Start turbine with speed control
5) Activate dc-link voltage control when Vd> Vd (m)n)
6) Activate FLC-1
•Any fault
RHirr.nnwN
1) Disable FLC-1 and FLC-2
2) Impose speed deceleration as
3) Yaw control to make wind pass through blades
4) Disconnect lino circuit breaker
JD.
Fig. 22 Control coordination diagram
-------
by short-circuit, overcurrent or over temperature. If the machine inverter trips, the line
side inverter can easily shut-down the system. On the other hand, if the line side inverter
trips, the dynamic break in the dc-link bus will keep the bus voltage in a safe range, while
the machine is decelerated to zero speed. Any fault that occurs during the search
procedure transfer the system to the shut-down procedure The fuzzy controllers are
disabled and a deceleration profile is imposed in the speed control mode.
4. Simulation Study
A 3.5 kW wind generation system, shown in Fig. 7, was simulated by PC-SIMNON [16]
to validate all the control strategies and then evaluate performance of the system. The machine
was modelled by a dynamic D-Q nonlinear model based on Park's transformation [13] , the
losses were incorporated in a simplified way, as will be explained later. Simnon allows a very
convenient simulation tool because the control algorithms (vector control and fuzzy
algorithms) are written in a similar C-language software implementation. The parameters of the
system under study are shown in Table 5.
Table 5 Induction machine and turbine parameters
Machine Parameters:
5 hp 230/460 V 13.4/6.7 A
4 poles 1S00RPM ' NEMA Class B
Rs = 0.370Q 1^ = 0.4360
Lis = 2.13 mH L:r = 2.13 mH 1^ = 62.77 mH
Turbine Parameters:
A = 0,015 B = 0.03 C = 0 015
tIgear^ 5.7
35
-------
In the beginning, the turbine was simulated with the model given in Fig. 5, and its
performance was verified with and without the oscillatory torques. Fig. 23 shows the
simulation of the coefficient of performance in respect to the tip-speed ratio simulation for
the turbine. Fig. 24 shows the variation of average turbine torque in respect to the tip-
speed ratio for several wind velocities. Fig. 25 shows the oscilatory train for wind
velocities from 6ra/s to 16 m/s.
0.40
0.35
0.30 h
w ^
£ z 0.25
2 2
U
E O 0.20
U_
LU d
O U-j
u 0.15
0.10
0.05
0.00
-
I
0
tr
tr
**
tr
_
\
tr
tr
7
tr
tr
ft
tr
-
XT
¦6
tir
tr
%
tr
k
t
r
j
&
tr
tr
«*
tr
U
tr
tr
ft
~
• tr
£
tr
tr
Jt
*
-t
*r
~ .tr
£
tr
tr
"O
r
«¦
*
III!
ill i
\ i t i
tr
i i ^ i
10
TIP - SPEED RATIO (X)
Fig, 23 Coefficient of performance of wind turbine
36
-------
V„= 16m/s
20 -
t
8
es
V„= 8 mis
V, = 6 m/s
9
2
3
4
5
6
7
8
I
0
TIP - SPEED RATIO (k)
Fig. 24 Turbine torque variation with tip-speed ratio
37
-------
30
25
20
15
10
V_ = 16 m/s
V„= 14 m/s
V.= 12 m/s
JV„=10 m/s
Vw = 8 m/s
V = 6 m/s
1 i I I I 1 1 1 I 1
0123456789 10
TIME (s)
Fig. 25 Oscillatory train for wind turbine
3$
-------
Although a detailed machine loss model is available in the literature [17], it takes too much
computational time to be simulated, because a very small step-size of integration is required.
Therefore, it was chosen to use a simplified loss computation to validate the control strategy
and fuzzy algorithms The stator and rotor losses are already embedded in the regular D-Q
model of the machine, as indicated by equations (27) and (28). The core loss was computed
as equations (29) and (30) indicate and the total machine power was computed as (31). The
generated power in the machine is negative, therefore, an addition as indicated decreases the
generated power by that amount. The inverter losses were also computed and taken into
consideration from the total line-side generated power
Stator copper loss = 3RsIj (27)
Rotor copper loss = 3RrI^ (28)
where:
Vg : Generator airgap voltage
Rm : Core loss resistance
fs : Stator frequency
s : Slip (p.u.)
Pm/c : Generator terminal power
(29)
(30)
(31)
39
-------
The utility voltage grid was considered to be balanced, but the inverter voltages
are not balanced during transient conditions. Therefore, Fig. 26 shows the circuit used for
derivation of the modeling equations (32) - (35). The inductance U was split in two
values, a local physical inductance (500 mH) and a grid distributed one (2 mH), and the
voltages were tapped for unit vector generation.
Vcz=>-/W\
R< U
• T? Lg
i i—r 111
Fig. 26 Utility grid model
=-p (vao - eao - - eno)
dt Ls
j (32)
~I7" = T (vbo -ebn -eno)
dt Ls
(33)
~jf" = 7 (vco -ecn — ^s*e -eno)
dt Ls
(34)
eno =^(vao + vbo + vco)
(35)
40
-------
Fig. 27 shows respectively the steady state turbine torque (Tt), turbine power (Pt),
and generated power (P0) as functions of wind velocity (Vw) and generator speed (%)
when none of the fuzzy controllers are in operation.
Fig 28 shows the operation of the fuzzy controller FLC-1 when the wind is
stepped up from 8 m/s to 12 m/s. The generator speed increases with the on-line fuzzy
search and the generated power boosts, the line power is negative in the simulation due
the following circuit convention: positive line power is for motoring and negative line
power is for regeneration. Every time that the fuzzy controller FLC-1 produces one step,
there is a ramp command, so as the power flow in the double pwm converter does not
reverse minimizing the power dips in the line side.
Similarly Fig 29 shows when the wind steps down from 12 m/s to 8 m/s. The
reference speed for the generator is decreased step-by-step and the generated line power is
again optimized.
Alter the speed optimization is finished, the fuzzy controller FLC-2 takes over and
decreases the excitation current command Ids" so as to boost a little more the generated
power. Fig. 30 shows the operation of fuzzy controller FLC-2 where the excitation current
Ids* is decreased, the power increases from 560 W to 640 W. The generator speed is kept
constant during this flux optimization cycle.
The fuzzy controller FLC-3 makes the speed controller robust against wind vortex
as indicated by Fig 31. In Fig. 31(a) it is shown an average wind velocity with an
oscillatory component to simulate wind vortex Fig. 31(b) shows the performance of a
regular PI controller and the fuzzy-PI responses. It can be seen that the fuzzy controller
keeps a better regulation of speed in the presence of wind turbulence.
The whole system with all the three fuzzy controllers was simulated to verify the
overall performance. The turbine was modeled with oscillatory torque and some
turbulence was added with the wind velocity to verify the robustness of controller FLC-3.
Fig 32 shows a step-up variation in the wind velocity with the correspondent speed
increase due to the action of the fuzzy controller FLC-1. Fig. 33 shows a more typical
slow
41
-------
16 m/s
4 m/s
20 -
12 m/s
10 m/s
10 -
m/s
m/s
4 m/s
(a)
16 m/s
4000
3000
.14 m/s
2000
2 m/s
Om/s
1000
8 m/s
J m/s
0
(b)
16 m/s
3000
14 m/s
2000
12 m/s
1000
10 m/s
8 m/s
m/s
0
500
0
1000
1500
(c)
Generator speed (RPM)
Fig. 27 Turbine and system model simulation curves (without fuzzy control): (a) Turbine
developed torque, (b) Turbine developed power, (e) Line side generated power.
42
-------
u
o
i—3 %n
S £
> o
12
10
2000
o
¦fi.
0
-2000
TURBINE POWER
*
LINE POWER
1100
a2 2:
51 ooo
1/5 W
10
15
_!
20
Time (sees.)
Fig. 28 Speed optimization with wind velocity step-up
I
25
43
-------
H 12
»—<1
w| 10
> £
cd
u
£ ?£
O w
dm
2000
0
-2000
TURBINE POWER
LINE POWER
1100
Q o
u 2
g 21000
m w
10
15
-r~
20
Time (sees)
Fig, 29 Speed optimization with wind velocity step-down
\
25
44
-------
C-
cc
960
950
940
U3
Cd
X
D
-J
(¦Li
-580
c*
jS ^ -620
O w
Cm
-660
12
14
i—
16
18
20
Time (sees,)
Fig. 30 Flux excitation optimization by FLC-2 after FLC-1 has finished.
45
-------
1040 1 1 1 1—
2.6 2.8 3.0 3.2
Time (sees.)
Fig. 31 Comparison of speed controller response by PI and FLC-3 for wind vortex (a)
Wind velocity with vortex, (b) FLC and PI responses
46
-------
8
10
—1—
20
30
10
Jju'i i'Uti'V -"" ""<*w***Wi*w***
-------
G
s
g,
nL<
0.7!
0.6
0.5
1.0
Q
0.8
W "a*
0.6
W cl
xfl
0.4
0.2
0.0
h
1.0
0.8
*1
0.6
0.4
§
0.2
J
0.0
U-,
1.0
p4
0.8
0.6
Q
CM
0.4
0.2
0.0
0
WMMMMMMMM
AP. due to FLC-2
1
10
20
30
40
HME(s)
FLC-2
FLC-1
FLC-2
FLC-1
FLC-2
Fig. 33 Time domain operation of fuzzy controls FLC-1 and FLC-2 (FLC-3 is also
working): (a) Wind velocity, (b) Generator speed, (c) Flux current, (d) Output power.
48
-------
average variation of the wind as given by the ramp profile. As the generator speed is
increased by FLC-1, the line output power gradually increases, but the line power
indicates some dips which require explanation. As generator speed command is
incremented by FLC-1, the machine accelerates to the desired speed with the power
extracted from the turbine output power. As a result, line power temporarily sags until
boosted by the turbine power at steady state. With a large increment of speed command,
the direction of Pc can even reverse. In order to prevent such conditions, the maximum
speed command increment was limited to a reasonably small value (75 RPM) and had a
ramp shape. Hie slope of the ramp can be adjusted to control the power dips. Note that
the speed command decrement will have an opposite effect; i.e., the generator tends to
decelerate, giving bumps in the output power. Fig. 34 shows the performance of the
system at variable wind speed when the three fuzzy controllers are in operation. If the
generator speed (o^) remains fixed and FLC-1 and FLC-2 are not working, line power
increases with increasing wind velocity. The operation of FLC-1 will give higher power
except at a wind velocity of 10 m/s where it is tangential because the generator speed is
optimum for that wind velocity. The incremental power gain due to FLC-2 is also
indicated in Fig. 34. This power gain gradually diminishes to zero as the wind velocity
increases. Fig. 35 shows the comparison of power gain with fuzzy controllers FLC-1 and
FLC-2 for fixed generator speed for wind velocities from 8 m/s to 14 m/s. In all modes of
system operation, the line current was verified to be sinusoidal at a unity power factor.
5. Hardware Description
The wind turbine system was implemented in a laboratory experimental setup to
validate the control algorithms and the overall design of the hardware and software. Such
apparatus is the basis for a 200 kW field installation, as required by EPA. Fig. 36 shows
the hardware block diagram for the experimental setup. There are two Texas Instruments
TMS320C30 digital signal processor boards and two analog I/O boards, manufactured by
Spectrum, constituting a multi-processor, multi-tasking DSP system. The remaining
circuitry was completely designed, built and debugged specially for this project and will be
49
-------
2000
Fixed Geoeralor
Speed (1050RFM)
FLC-1
FLC-2
12
10
14
8
WIND VELOCITY (m/s)
Fig. 34 Steady-state line side power boost with FLC-1 and FLC-2 control
2500
2000
1500
1000
FLC-1
500 — FLC-2
WIND VELOCITY (m/s)
Fig. 35 Steady-state line side power boost for several wind velocities
50
-------
described later. Special purpose microprocessors, such as the TMS320 family of
processors, have been used in digital signal processing since the early 1980s. The Intel
2920 appeared first, followed by NEC's uPD7720. In 1982, Texas Instruments Inc.,
introduced the first-generation TMS32010 digital signal processor, followed by the
second-generation TMS32020 in 1985 and the faster C-MOS version TMS320C25 in
1986. The third-generation TMS320C30 as the one used in this project, is a true 32-bit
processor containing integer and floating-point arithmetic units, 2048 x 32 bit words of
on-chip RAM, 4096 x 32 bit words of on-chip ROM, control unit and parallel/serial
interfaces. This microprocessor operates from a 33.3 MHz clock, achieving a performance
of 16 7 million instructions per second with features such as a 32-bit by 32-bit floating-
point multiply in one instruction cycle and special addressing modes for circular buffering
and bit reversal.
The board manufactured by Spectrum contains the TMS320C30 chip along with
special features for ease of software and hardware development. The board is suitable for
PC-AT compatibles as it uses the 16-bit ISA-bus. There is a dual port interface which
allows the PC to access the memory in the DSP board by using the Hold/Hold
Acknowledge feature of the TMS320C30, such peculiarity allows a host program in the
personal computer to communicate and change parameters on-the-fly in the software that
it is running the DSP board. The board occupies a single 16-bit slot within a PC-AT
expansion bus and the base address is specified by links (LK2) The Processor Board #1
was configured for the address 0x290 and the Processor Board #2 for the address 0x390.
The TMS320C30 provides two programmable serial ports which permit 8, 16, 24
or 32 bit transfers for connecting external CODEC interface or any other synchronous
serial circuitry at rate of 6 Mbits/sec. These ports have been buffered in the board and are
accessible from the ear endplate of the PC, they were used for transference of signals
between the Processor Board #1 and the Processor Board #2,
51
-------
6 Mbits/sec
2 CH 4 CH
2 CH 4 CH
ANALOG I/O
ANALOG I/O
Encoder
Interface
Inverter
Interface
Inverter
Interface
Hanning
Chip
Hanning
Chip
DSP Link
Prototyping
Module
DSP Link
Prototyping
Module
IGBT Power
Inverter
IGBT Power
Inverter
C30 Processor
Board #1
DSP Link
Serial Port
Serial Port
C30 Processor
Board #2
DSP Link
Fig. 36 Hardware block diagram for wind generation system
-------
A parallel expansion system is provided as a memory-mapped peripheral area. It
has a 16-bit width, and follows the proprietary DSPLINK bus. Transfers over this
expansion use 2 wait-states to achieve a 180 nsec transfer cycle which is suitable for
ribbon-cable connection to peripheral boards. The DSPLINK in each Processor Board is
connected to both the 4-channel I/O board and to the Hanning PWM integrated circuit.
There is a prototype interface module which consists of bidirectional buffers and address
decoders to help to interface the DSPLINK bus to the Hanning PWM. On the Processor
Board #1 there is an extra connection from the encoder interface to the DSPLINK. This
encoder interface counts the pulses coming from the optical encoder coupled to the
machine shaft to generate the speed signal.
, The Hanning PWM chip generates the PWM pulse pattern signals for the IGBT's
gate drives The inputs as indicated in Fig 37 are the d-q synchronous rotating reference
frame voltages (v'd/, v^*) and the correspondent angle (0e*).
VECTOR
ROTATION
I INVERTER
* INTERFACE
DEAD
TIME
e!
Fig. 37 Hanning PWM integrated circuit
The Hanning PWM integrated circuit has salient features as;
• 3-phase pulse width modulator (PWM) for 3-phase motors.
• Pulse pattern generation for a 3-phase supply at the required voltage, frequency and
phase angle.
• Switching frequencies up to 20 kHz and resolution of the switching signals up to 50
ns.
53
-------
• Presetable dead-band, turn-on and turn-off times
• Makes provisions for turn-on, turn-off and dead-band compensation
• Transforms input voltages from cartesian into polar form
• Optional 8/16 bit wide bus interface compatible with a range of 8 bit single-chip
microprocessors and signals processors.
The pulse patterns output from the Banning PWM integrated circuit are connected
to the inverter interface circuit which consists of a low impedance RS-422 cable driver, a
buffer that disables the gate pulses in case of fault, a latch-up to hold the signal fault and
optocouplers to isolate the gate pulses from the IGBT's gate drivers The power IGBT
inverters were built from Powerex Intellimod modules which have all the built-in gate
drivers, over current, overvoltage and temperature protection Each subsystem is discussed
in detail in the following sections*
5.1 Hanning PWM
Both Hanning integrated circuits for the two inverters are mounted in a PC
prototype vector board with wire-wrapped connections. This board is inserted in the PC
ISA slot. The interface with the C30 DSP board is achieved with DSPLINK prototype
boards which are also mounted in the PC vector board. The circuit of the DSPLINK
interface is given in Fig. 38. All control signals and data signals that make up the
DSPLINK interface with the TMS320C30 are physically carried on a 50-way conductor
ribbon cable. The header connectors have two rows of 25 pins, with 2 54 mm spacing
between all pins. The fully buffered data, port access strobes and control lines are brought
out as wirewrap pins, which are inserted through holes in the prototyping card There are
read (Rx) and write (Wx) signals for four circuits that can be conveniently addressed by
hardware links in the address decoder IC4.
54
-------
AO
1 8
TT
1 B
i fe"
1 4
13
if
1 1
1B
jE
1 3
ia
11
B1
A1
52
AS
B3
A3
04
A4
05
A5
86
AS
B7
A 7
B0
A8
G
DIR
74ALS245
81
A1
52
A2
83
A3
04
A4
B5
AS
B5
AS
B7
A 7
ee
A6
G
is_
OJR
g
*32
74ALS24S
l/w >-
Ap
Ott-
I
[/ I OE >-
VCC
,>
Y1 3rf"
Y2 3rf-
I «9 ¦*
-------
The Harming PWM interface is shown in Fig. 39. It has a 16 Mhz clock and the
16-bit data, address, and cpntrol signals are connected to the DSPLINK interface. The
interruption that drives the control software in the DSP is generated by the Banning chip
each 32 jis. This pulse is counted four times by the two flip-flops 74HC74 in order to
generate the fastest interruption service routine of 128 |is. Hie monoestable 74HC221
shapes the pulse into a 140 ns low level pulse to drive the interruption INT1 in the
TMS320C30, A low level of 140 ns is required to not trigger multiple interruptions with
longer pulses. The interruption #1 in the €30 attends an interruption service routine with
all software required for the fast software task, including the synchronous current
controller, vector rotation and writing into Harming chip. The switching gate pulses output
for the six devices (01, Ul, 02, U2, 03, and U3) are connected to the cable drivers
DS8922A. Such configuration was-used because the interface from the Hanning board to
the interface board in the inverter is somewhat far. A direct connection from the Hanning
output pins to the inverter would be very sensitive to noise due the high npedance in
those lines. The cable driver DS8922A is a RS482 compatible driver, able to transmit the
pulses from Hanning to the inverter interface with a low impedance and differential
connection, where the logic level "1" is the flow of the current in the cable in one
direction, whereas the logic level "0" is the flow of the current in the cable in the opposite
direction.
5.2 Inverter Interface
The inverter interface as shown in Fig. 40, receives the gate drive signals through
the cable driver DS8922A and applies into the 74HC365 buffer which can be disabled in
case of fault like overcurrent, overvoltage or overtemperature. The raonoestable
constituted by 74HC221 is used for power inverters that require a Longer pulse for
disabling the gate drives. However, the Powerex Intellimod has already a 1 ras disable
pulse for faults, and only the input GONl was connected. The input G0N2 was used for
56
-------
~D
QD
-MOM
vcc vcc
DO
CLK CLKO
01
M Upper 1
Lower 1
02 30 Upper 2
27 Upper 3
AO
Al
GND GND GND GND
74HC74
+1
VCC CLR PR
D Q
+VCC
,£S| . |
> CLK
74HC221
17 IINT
+VCC
100 oF
i 3
j
1 4
1
5
j
|
6
J
I
7
1
)
8
J 9
1 11
12
1
I
13
J
14
1
1
15
J
1
16
I
17
j
t-
18
-i GND
iGNDt
VLC
PR
D
Q
>CLK
0
Fig. 39 Hanning PWM interface with cable driver
57
-------
+5V
CT1
+5 V
>U+
>u-
*v+
74HC365
CI2
i f9 +
V—
*w-
CI3
iftn
rrfrt
CI6
SnF
74HC221
GONl (ACTIVE LOW)
5 K
GON2 (ACTIVE HIGH)
Fig. 40 Inverter interface
-------
user turn-on/turn-off with a switch. The 74HC365 outputs can sink up to 30 mA and are
easily interfaced with the optocouplers from the next stage.
5.3 IGBT Power Inverter
The IGBT power inverter was built with Intellimod Powerex modules which are
very convenient for laboratory experimentation. Such modules have built-in gate drives
and protection circuitry and just need to be interfaced with fast optocouplers and four
isolated 15 V power supplies as shown in Fig 41.
The six incoming PWM signals are connected to the HCPL4504 optcouplers. The
upper three IGBTs have three linear power supplies derived from a small transformer and
a 7815 linear regulator. The 4N26 optocouplers are driven in case of fault, the transistor
outputs are OR-hardwired to set the flip-flop 7474 that turns-off the gate signal. The three
bottom IGBTs have a common ground power supply, but the gate signals are also
isolated, The high frequency capacitor in parallel with the 300 V supply minimizes the de-
bus leakage inductance effect and the two 1N5388 zeners protect the module against
spikes more than 400 V. Of course there is a dynamic break set to 365 V in the system
that protects overvoltages due to machine regeneration. The inverter circuit was built in a
printed circuit board which is shown in the appendix A.
5.4 Speed Sensor Interface
The speed sensor is a Teledyne Gurley optical encoder, mounted on the machine
is
shaft, with 3600 pulses per revolution. The pulse outputs have a time width of 0.4 \xs and
are direction-sensed (CW pulses and CCW pulses are on different terminals). The output
device is a balanced differential line driver that complies with the RS-422 standard Since
the generator always works with a positive speed, the simple circuit shown in Fig. 42 was
used to count the incoming pulses. The transceiver 75173 receives the differential
pulses
59
-------
-|7»lSh
,20K
lOOn
10/4
HCPL 4504
u+:
PM50RSA060
v+:
-f7R15 h
i20K
lOOn
10/4
20K.
20K
10 K
lOOn
lOOn
lOOn
100/4
ft—~ GON
yis
' J""
HCPL 4504
v-:
u-:
47 74HC74
Fig. 41IGBT Power Inverter
-------
74HC02
RCK
560
CCK
_ 74LS590
CCLR
CCKEN
rcd
D15
SCO
DATA BUS
Fig. 42 Speed sensor circuit
and gives TTL compatible output pulses. They drive the 3-state 74LS590 counters which
are connected in the DSPLINK interface in the Processor Board #1. When the read signal
R2 goes low, the counter outputs are transferred to the data bus and read by the C30
processor as described later in the software section.
5.5 Peripheral Circuits
The analog 10 board interface with the required peripheral circuits for the input
voltages and currents and the output for the dyno control The Processor Board #1 needs
the machine currents for generation of i<,sK and W8, the dc-link voltage and current (vd and
id) as indicated in the main block diagram in Fig. 7. In addition to those inputs, the
Processor Board #1 also sends the command voltage for programmable wind turbine
torque in the dynamometer. On the other hand the Processor Board #2 needs both line
61
-------
voltages and phase currents to generate \>vs, v*', V and W. The currents are sensed by
Hall-effect sensor devices as given by the Fig. 43. These currents are the inputs for a
hardware circuit that carries out some filtering and the 3 to 2
-------
10 oF
LEMLA50-S +J5V
+VCC
1K2
-VCC
fT7T7
-15V
c*
10 oF
LEM LA 50-S +1SV
5K
+VCC
1K2
-VCC
-15V
rrrn
5V
Fig. 43 Current Sensor Interface
63
-------
irn t9<
P0T4
orrstT rtOJ
VA ¦
*K
POT 6
orrseT flOJ
,POTO
Fig. 44 Three phase transformation circuit with filtering
-------
+15V
AD202 vcc
lOQpF
4 x 470KD
470nF
® f—~
V V
rrm
Fig. 45 Dc-link voltage intetface circuit
65
r
-------
10 oF
+15V
LEMLA50-S
1K2
-15 V
+VCC
-vcc
Fig. 46 Dc-link current interface circuit
66
-------
+15V
AD202
COM
Dyno
Ground
Computer
Ground
Fig. 47 Dynamometer control interface
67
-------
Vw
FROM PC
DSP C30 CALCULATION
®t
FROM DSP
CALCULATE
CP(X)
SOLVE TURBINE
MODEL
AND LOAD DAC
i -V,
On
00
TORQUE PROGRAM PLUG IN J21
\
E1546 PC BOARD
> >
¥
Q O
\o
¦
+MAX -MAX
VOLTS VOLTS
0
0
202TB [9Mm2 1314 1516 17 1819
\zr—v a' ' —
FWD TORQUE
~
REV TORQUE
Fig. 48 Diagram for turbine torque emulation by dynamometer
-------
The assembler translates assembly-language source files into machine language
object files. Source files can contain instructions, assembler directives, and macro
directives. Assembler directives control various aspects of the assembly process such as
the source-listing format, symbol definition, and method of placing the source code into
sections. Macro directives permit a concise representation of groups of instructions that
occur frequently
The linker combines object files into one executable object module. As it creates
the executable module, the linker performs relocation operations and resolves external
references. The linker accepts relocatable COFF (Common Object File Format) object
files, created by the assembler, as input. It can also accept archive library members and
output modules created by a previous linker run. Linker directives allow the user to
combine object-file sections, bind sections or symbols to specific addresses or within
specific portion of the DSP memory, and define or redefine global symbols. An associated
archiver can create macro or object-file libraries.
The C-source-debugger is a very important tool for debugging programs. Its
interface consists of a screen broken into windows that display the internal registers, the
reverse-assembled program, it shows memory, breakpoints and a wealth of information.
There is a window where the C code is listed and can be stepped together with the
assembly code.
There are two DSP boards named Processor Board #1 and Processor Board #2
plugged into a host personal computer running basically three programs: the Processor
Board #1 runs the program "dspl out" which corresponds to the C source program
"dspl.c," the Processor Board #2 runs the program "dsp2.out" which corresponds to the
source code "dsp2 c," and the PC runs the DOS-executable program "host exe," which
corresponds to the C code "host c."
The resident program in the PC is compiled with Microsoft Quick-C. By including
the header TMS30.H in the source file all the library functions for using the DSP board are
declared with function prototypes. Such library routines are used to download
TMS320C30 object code to one or more boards, to start execution of that code (or stop if
69
-------
already running) and to pass data back and forth between the program running on each of
the TMS320C30 boards and on the PC.
The function SelectBoard is used to select a board, for the first time. Then all
functions used after a call to this function apply to the current board. To switch between
boards after each one has been initialized by SelectBoard the function WarmSelect is used
instead. Both functions need the hexadecimal I/O port base address, e.g. to select
Processor Board #1 the function should call SelectBoard(0x290), whereas to select
Processor Board #2 the function should call Se!ectBoard(0x3 90).
The best way to move data between the PC and DSP is to use the dual-access
memory on the DSP board. The board is normally supplied with 64K of memory in this
dual-access area, addressed as locations 0x030000 through 0x3FFFF, Although the PC
can read and write other memory areas on the board, it incurs more overhead because the
DSP chip must be "held" during PC accesses. The variables or arrays that are being passed
must be defined as global variables. This will ensure that they are placed in the dual-access
memory, in accordance to "bss" memory section defined in the mapfile
"LSICMAP.CMD " The linker automatically reserves the first sixteen words in the dual-
access memory to hold variables pointed from 0x030000 through 0x3000F. Those
locations can hold pointers to variables* or arrays of pointers, therefore, they are used to
communicate between the Processor Boards and the PC '
Before describing the overall control software it is important to understand how
the interruptions are handled in the C programs that run in the DSP boards. The system
comes pre-configured with "sections" for setting up interrupt vectors by assembler
directives that are automatically passed to the C30 assembler after the C Compiler runs.
The interrupt names, section names, and vector locations are given in the Table 6,
70
-------
Table 6 Interrupt names, section names and vector locations
Interrupt Names
"Section" Name
Absolute Location
RESET
.intOO
0x00
INTO
.intOl
0x01
INT1
,int02
0x02
INT2
int03
0x03
INT3
,int04
0x04
XTNTO
into 5
0x05
RTNTO
,int06
0x06
XINT1
int07
0x07
RINT1
int08
0x08
TINTO
int09
0x09
TINT1
.intlO
0x0a
DINT
Jntl 1
0x0b
In order to set up an interrupt vector (a pointer to the interrupt service routine,
placed in an absolute location in low memory) the following statements must be used at
the very beginning of the interruption routine:
asm(" .sect \".int02\"")>
asm(" .word _c_int02").
asm(" text");
The first statement tells the assembler to create , a program "section" named
" int02 " The assembler output is passed to the linker, which places this section in absolute
memory location 02. It is essential to direct the linker to use the "map file"
LSICMAP.CMD for this process, The second statement tells the assembler to form the
address of the interrupt service routine function, which gets placed into an absolute
memory location The underscore must be used because the C compiler puts an
underscore in front of all function names and variable names before handing them to the
71
-------
assembler. The third statement tells the assembler to go back to assembling things into the
"text" (code) section, instead of the absolute-located sections that were created for the
interrupt vectors.
In order to initialize and enable the interrupts, the following in-line assembly
language statements must be used as an example for INT I:
asm(" OR 2h, IE");
asm(" OR 2000h, ST"),
The first statement sets one of the bits in the C30's Interrupt Enable Register, by
OR'ing a value of 02h with the current contents of the register. The second statement sets
the Global Interrupt Enable bit (bit #13) in the C30's Status Register, This bit must be set
or the C30 will not respond to any interrupts, even if they are enabled in the IE registers.
The control software that runs in each of the Processor Board is a multitasking,
real-time program. Therefore, it is necessary to trigger the interruptions by software in
such a way that lower level interruption, are able to be interrupted by higher level ones
with automatic context save/restoring of variables. The following statement is used to
trigger the interruption #2 (INT2) by software, as if one hardware interruption has
occurred;
asrn(" OR 2h, IF");
The most critical time assignment task is the operation of writing in the Manning
chip the 16-bit words for the reference frame voltages (vAs", v^') and the angle reference
(0/ ), For this reason, the control software timing is tied to the interruption generated by
the Harming chip. The programs "dspl.c" and "dsp2.c" have a basic structure as indicated
by Fig. 49, where the background loop for host communication is interrupted by four
services: ADC interruption service routine (INTO), and the correspondent decreasing
priorities interruptions for task #1 (INT1), task #2 (INT2) and task #3 (INT3).
72
-------
INITIALIZE POINTERS
INITIALIZE VARIABLES
INITIALIZE SINE/COSINE TABLE WITH MALLOC()
INITIALIZE MANNING CHIP
INITIALIZE SERIAL PORTS
CALIBRATE ADC
BACKGROUND LOOP (HOST COMMUNICATION) i
ADC ISR WITH INTO ,,
I!
TASKl WITH INTl
TASK2 WITH INT2
TASK3 WITH INT3 .;
Fig. 49 Flow diagram for the multitasking real-time DSP code
Fig. 50 shows the tasks coordination chart. The four-channel ADC conversion is
triggered at the end of task 1 and although the ADC is INTO-based, the complete conversion
of the four channels takes 16 ^s, which is inside the 128 us timing for rNTl. The context
save/restore is automatically handled by the DSP, due to the built-in priorities for the
interruptions INTO, INTl, INT2, and INT3. The interruptions INT2 and INT3 are triggered
inside the INTl by a loop counter, so as to generate 1.028 msec and 1.024 sec interruptions
as explained later.
73
-------
Fig. 50 Tasks coordination chart
The interruption service routine that converts the four channels from the ADC
board is triggered at the end of task 1. The total conversion time is 16 (is. Therefore, the
conversions are over before the next sampling time (128 [is) for task 1, Since the software
code for conversion is common for both Processor Boards, the explanation of operation
follows.
The interruption service routine for the ADC conversion is embraced by c_int01().
The IO board is triggered to start the conversion of the first channel at the end of the task
1 by writing to the correspondent C pointer (identified by *timer). (My one analog to
digital converter is used to convert all four channels, therefore, each channel is converted
one at a time. An end of conversion interrupt (INTO) is generated after each channel is
finished being converted, producing four interrupts after a single ADC software trigger.
The variable adc_value reads the converted channel in the C pointer named ~ade_result.
Because the ADC has a resolution of 12 bits, the variable adc_value is shifted 20 positions
to the right, due to the fact the all DSPLINK communications are made on the upper 16
bit word of the TMS320C30. The variable isrO is initially at value I, indicating that the
74
-------
first channel needs to be converted, after each conversion, isrO is updated for the next
channel and the pointer * control triggers the 10 board again. At the fourth channel, isrO is
set to the value 1 and the 10 board conversions are held until next cycle
6.1 Control Software in the Processor Board #1
The interaction with Manning chip, 10 board, speed sensor, serial port, and host
communication are memory mapped. Therefore, the correspondent C pointers are
initialized in accordance to the Table 7,
The hardware interruption INTl indicates that Harming chip is ready to receive the
words for vds* represented by the variable UA, vqs* by UB, and 0e* by PHIL The interrupt
is initiated at the end of a Hanning processing cycle with the positive edge of the Harming
INT signal which is counted four times by the two flip-flops connected to the Hanning
INT The software for the interruption is coded inside the function c_int02(). Initially the
global interruptions are disabled and the control word is written by means of write
address (WAD = 0) into status register (A1 = 1) to write UA, and the following values
(UB and PH31) are separated by assembly ''NOP" commands in order to achieve the
required processing timing in the Hanning chip. The voltage words UA and UB are
normalized in 2's complement format from 0x8000 to 0x7FFFF as the synchronous
current controller outputs vas* and vqs* vary from -1 to +1. The angle word PHI1 is
normalized from 0x0000 to 0x6000 as the phase angle varies from 0 to 2n,
After the operation of writing in the Hanning chip, the stationary frame currents id/
and i<,ss are filtered by a digital low-pass-filter. The encoder is then read and the difference
of angle is generated. Since the encoder is precisely read each 128 |is the angle difference
is already proportional to the machine speed. The last five angle differences are stored to
make a speed moving average filter. The interruption service routine must perform the
necessary computations for the synchronous current controller, which are Pi's controllers
that generate vd5" and vqs* based on the error of currents, The synchronous current
controller requires inverse vector rotation of currents. The phase angle generation comes
from integration of the angular frequency coc
-------
Table 7 Memory mapped hardware addresses for DSP #1
Memory pointers
Hex address
*pwm data
0x0800000
~pwm status
0x0800001
~encoder read
0x0800002
~control
0x0800008
~timer
0x0800009
~adc result
0x080000a
~dacO output
0x080000a
~dacl output
0x080000b
~serial globalO
0x0808040
*xmit controIO
0x0808042
~rece controO
0x0808043
~ser tim controIO
0x0808044
*ser tim countO
0x0808045
~ser tim periodO
0x0808046
~data xmitO
0x0808048
~data receivedO
Ox080804C
~serial global 1
0x0808050
~xrrut control 1
0x0808052
~rece control
0x0808053
~ser tim control 1
0x0808054
~ser tim count 1
0x0808055
~ser tim period!
0x0808056
~data xmitl
0x0808058
~data received 1
0x080805C
~WRITE TO LOG
0x0030000
~LOGO
0x0030001
~LOG1
0x0030002
~LOG2
0x0030003
~LOG3
0x0030004
~CHANO
0x0030005
~CHAN1
0x0030006
~CHAN2
0x0030007
~SPDREF
0x0030008
~IDSREF
0x0030009
~IQSREF
0x003000A
~RPM
0x0030008
~START UP IS FINISHED
Ox003000C
~ENABLE FLC2
0x003000D
~CHAN3
0x003000E
~VW
Ox003000F
76
-------
The angular frequency tct is calculated by the addition of the machine shaft speed
o\ (in electrical rad/sec) with the slip frequency ay, The machine shaft speed is computed
in task 2. The slip frequency is the result of multiplication of the slip gain (K*) with ip\
which is the output of the inner torque feedback control in task 2. By having the numerical
value of the angular frequency tCfe, the integration is performed in accordance to the
Hanning chip scaling as
Kfe = 0.5* we;
angle_count = last_angle_count + Kfe;
if (angle_count > 24575)
angle_count = angle count - 24576;
}
last_angle_count = angle_count;
PHI1 = angle_count« 16;
where Kfe is the constant for integration that depends on the Hanning setup, the
angle_count is integrated over the last_angle_count and shifted down when the upper limit
for angle occurs (0x6000). The word for PHI1 is the variable angle_count left shifted 16
bits because the DSPLINK interface uses only the upper 2 bytes in the TMS320C30 data
The inverse vector rotation for currents uses the current value for sine and cosine
which is pointed by the numerical value of angle_count. A look-up table for sine and
cosine has been loaded in the initialization with the memory allocation C function mallocO-
The pointer for the look-up table is a multiple of 32 since the look-up table has 768 values
for sine and cosine. The retrievement is done with the following code segment:
step_table = 0.03125*angle_count;
look_up_sine = init_sine_add + step_ table;
look_up_cosine = init_cosine_add + step„table;
sinthe = *look_up_sine;
costhe = *look_up_cosine;
77
-------
The variable stcp_table is the 1/32 multiple of the angle_count» the pointers are
loaded with the initial address (init_sine_add and init_cosine_add) plus the current value
for the step_angle and finally the values for sine and cosine are taken by pointing to the
correspondent memory address.
The calculation for inverse vector rotation, PI controller for generation of
reference voltages, limitation of outputs and scaling comes next to produce the values for
UA, UB and PHI1 for next interruption.
The last assignment for this interruption service routine is to trigger the 10 board
for conversion and to trigger the tasks at exactly L024 msec and 1.024 sec timing. There
are two loop counters lpc2 and lpc3 which count 8 times and 10000 times respectively to
trigger the interruptions by OR'ing the interrupt flag (IF) as
if (lpc2 — 8)
{
lpc2 =1;
asm(" OR 00G4h, IF"); /* Trigger interruption # 2 */
}
else {++lpc2;}
if (lpc3 = 10000)
I
lpc3 =1;
asm(" OR 0008h, IF"); /* Trigger interruption # 3 */
}
else {++lpc3;}
At the end of task 1 the global CPU interrupt is enabled and the 10 board is
triggered to receive the four channels one at the time by the following code
asm(" OR 2000h, ST"); /* Enable global CPU interrupts */
* timer = adc_trigger;
The interruption INT2 corresponds to the control software that must be executed
each 1.024 msec and is triggered by software in the task 1 with the loop counter lpc2. This
interruption is identified by the software code embraced by c_int03Q. Initially the speed
78
-------
moving average filter prepares the speed signal wr. The speed control loop is a fuzzy-PI
that uses the actual speed wr and the reference speed wref (which comes from the serial
interface communication). The fuzzification, rale base evaluation and defuzzification are
performed as already discussed. Initially the control software identifies where the inputs
error-in-speed (E) and change-in-error (CE) belong. As an example, the indexing for the
universe of discourse for error (E) is arranged as Fig. 51 shows.
Fig. SI Example of indexing for C program membership function evaluation
The index jj indicates one interval where two straight line equations define the
degree of membership. Each fuzzy set is indexed by two values of the variable jj, e.g. if the
error in pu (Epu) is NL then jj = -4 or jj = -3, such fuzzification is indicated in the
following software segment;
else if (Epu <= -0.5)
{
. jj = -4;
uel = 2 + 2*Epu;
}.
/* Membership value of NL */
79
-------
else if (Epu <= -0.2)
{
jj = -3; f* Membership value of NL */
uel = -0.66666667 - 3.3333333*Epu;
}
The indexing of the intervals help to evaluate the rales as indicated by the
following software code, where the rule #1 is evaluated.
if (jj =-5 II jj =-4) /* E = NVL */
{
/* Rule #1 IF E=NVL AND CE=ZO THEN CTE=NVL */
if (kk ==-1 II kk = 1)
{
CTEpu += NVLCTE*MIN(ue2,uc2);
}
}
The variation of torque in pu (CTEpu) is reset at the beginning of every fuzzy
evaluation. After all the relevant rules are fired, the total variation of torque is summed
over in order to build the reference torque value as
CTE = CTEpu *KTE;
UK = last_Terf + CTE;
where CTEpu is multiplied by the scaling gain (KTE) and the controller output (UK) is
formed with the integration of CTE. Such output (UK) passes through a limiter in order to
make up the reference torque signal (Terf),
This reference torque is compared to the estimated torque Tl. The estimated
torque H comes from the dc-link power divided by the speed. The dc-link power is also
converted to a scaled integer value (pfint) to be sent via serial port communication to the
Processor Board #2 for feedforward power loop control. The inner torque loop is a
proportional controller that generates i^* indicated by the variable iqsref which is the input
of the current controller executed in the faster loop of 128 Us. The angular frequency at is
80
-------
calculated by the addition of the machine shaft speed (in electrical rad/sec) with the slip
frequency coy. The wind turbine programmable torque for the dynamometer is executed by
if (vw != 0)
{
L = 0.28333*wref/vw;
}
if (L > 9.25)
i
L = 9.25;
}
if (L<0)
{
L = 0;
}
Cp = at+bt*L+ct*L*L+dt*L*L*L+et*L*L*L*L+ft*L*L*L*L*L+
gt*L*L*L*L*L*L44it*L*L*L*L*L*L*L;
if (Cp<0)
{
Cp = 0;
Tro = 0.26604*Cp*vw*vw;
The tip-speed-ratio is given by L, the coefficient of performance is given by Cp and
the turbine torque Tm is calculated as above. Hie wind velocity can be either programmed
in the DSP by software or can be input with the host interface program. The turbine
torque is sent with the DAC channel 1 to the dynamometer with the isolation amplifier
interface. It is also possible to add the oscillatory torque to Tm, as provided in the code
(only the fundamental oscillatory torque was considered, in order to not overload the
control software timing).
The next assignment is the serial port communication. There are two signals to be
sent and two signals to be received. The software code below shows how this
communication is accomplished.
81
\
-------
/*** Test if serial port #0 is ready to transmit ***/
cal_status = (*serial_globalO & serial_is_ready),
if ( cal_status!- 0 )
{
*data_xmitO = pfint,
}
/**.* Test if serial port #0 is ready to be read ***/
calstatus = (*serial_globalO & new_data_is_in),
if ( cal status NO)
{
wrefint = *data_receivedO,
wref = 1.8626483 le-7*wrefmt»
I
/~** Test if serial port #1 is ready to transmit ***/
cal status = (*seriaI_globall & serial is ready),
if ( cal status 1= 0 )
{
*data_xmit 1 = vdint;
}
/*** Test if serial port #1 is ready to be read ***/'
cal status = (*serial_global 1 & new_data_is_in),
if ( cal status != 0 )
{
point = *data_receivedl,
po = 1 8626483 le-6*point;
}
The serial port # 0 sends the dc-Iink power pf and receives the reference speed
wref. The serial port # 1 sends the dc-link voltage vd and receives the generated line
power po. All such variables are scaled in 32 bits for transmission in the serial port.
The last piece of software executed by the task 2 is the signals log service The
reason for such software is for documentation and debugging purposes. It is possible to
write an ASCII file in the PC with five columns The first column is the time interval for
the correspondent logging The following four columns are four signals that can be chosen
at the time of compilation. Here it is demonstrated for the case of wind velocity (vw), line
side power (po), generator speed (wr) and machine excitation current (ids).
82
-------
if (logindex = 39) /* The logging will happen every 79.872 msec */
{
logindex = 0;
if (write_to_log == 0)
{
write_to_iog = 1;
*WRITE_TO_LOG = 1;
~LOGO = vw;
*LOGl = po;
*LOG2 = wr;
*LOG3 = ids;
}
else
f
write_to_log = 0;
* WRITE_TO_LOG = 0;
}
}
else {-H-logindex;}
The interruption INT3 corresponds to the control software that must be executed
each 1.024 sec and is triggered by software in the task 1 with the loop counter ipc3. This
interruption is identified by the code embraced by e_int04() and has the software code for
the fuzzy controller FLC-2 which performs the flux optimization by searching the
optimum ids*. The signal enable_flc2 commands when the fuzzy algorithm starts, just
after the speed optimization by FLC-1 has been finished. The fuzzy algorithm is
implemented in a similar way as described for FLC-3. The scaling gains for variation of
power and variation of current are calculated as
GPA = (kl*2.6525198939e-3)*wref + k2;
GIDA = (c 1*2.6525198939e-3)*wref + (c2/20.83)*abs(iqsref) + c3;
Such gains are then limited and the per-unit input values for the fuzzy algorithm
are calculated. The indexes mm and E mark where the input variables lie in the range of
the universe of discourse. For instance, the software code below shows the fuzzy inference
for rule #2.
83
-------
/~Rule #2 IF Di=N AND DP=PM THEN DIDSpu = PS */
if (mm =-2 !! mm =-1)
{
if (11 = 3 || II = 2)
{
DIDSpu += PSDI*MIN(uil,upl),
}
}
The variation of current in pu (DIDSpu) is reset at the beginning of every sampling
time, and after the inference engine is completed it is scaled to the actual value, limited to
ensure stable operation and summed over with last excitation current (last_Idsr) to build
the output (Idsr).
6.2 Control Software in the Processor Board #2
The Processor Board #2 has a similar software structure as the Processor Board
#1. The DSP board #2 also interacts with the line-side inverter with a Harming chip, it has
an 10 board for acquisition of voltage and currents, it communicates with the DSP board
#1 with the serial port, and with the PC with a host communication dual-port memory,
Table 8 shows the initialization of the required C pointers for hardware interface
As already explained, the hardware interruption INT1 indicates that Harming chip
is ready to receive the words for v^" represented by the variable UA, vqs* by UB, and 9e*
by PHI 1. After the writing operation in the Hanning chip the stationary frame currents ids"
and iqss and the stationary frame voltages vds* and v<,s5 are filtered by a digital low-pass-
filter Since the control for the line side inverter is a direct-vector-control system, the sine
and cosine waves are generated based on the voltages, and the phase angle generation
comes from integration of the angular frequency oe, which is computed from the unit
vector and conveniently reset when cosine is one and sine is zero. The stationary frame
currents are inverse-vector-rotated and the PI controllers are executed to generate the v*,*
and vqs" signals based on the error of currents. The decoupling network was not
implemented because the voltage loop control worked very satisfactorily for this drive
system, but it might be considered for the final 200 kW installation.
84
-------
Table 8 Memory mapped hardware addresses for DSP #2
¦ " VV, g' — '"V, .
I Memory pointers | Hex address |
*pwm data J 0x0800000
| *pwm status | 0x0800001
*encoder read
0x0800002
* control
0x0800008
•timer
0x0800009
*adc result
0x080000a
*dac0 output U Ox080000a
*dacl output
0x080000b
* serial globalO
0x0808040
*xmit controlO
0x0808042
*rece controO
0x0808043
*ser tim controlO
0x0808044
*ser tim countO
0x0808045
*ser tim periodO
0x0808046
*data xmitO
0x0808048
*data receivedO
0x080804C
* serial global!
0x0808050
*xmit control 1
0x0808052
*rece control
0x0808053
*ser tim control 1
0x0808054
*ser tim count 1
0x0808055
*ser tim period 1
0x0808056
*data xmitl
0x0808058
*data received!
Ox080805C
* WRITE TO LOG
0x0030000
~LOGO
0x0030001
*LOGl
0x0030002
*LOG2
0x0030003
*LOG3
0x0030004
*CHAN0
0x0030005
*CHAN1
0x0030006
*CHAN2
0x0030007
*SPDREF
0x0030008
* VDREF
0x003000A
*RPM
0x003000B
•START UP IS FINISHED
0x003000C
*VD
0x003000D
*CHAN3
0x003000E
85
-------
The instantaneous power is calculated and filtered and the tasks #2 and #3 are
triggered with loop counter variables Ipc2 and lpc3. The 10 board is also triggered for
acquisition at the end of the INT1 service. The interruption INT2 corresponds to the
control software that must be executed each 1,024 msec and is triggered by software in
the task 1 with the loop counter lpc2. This interruption is identified by the software code
embraced by c_int030- Initially the angular frequency passes into a moving average filter,
as well as the line side power signal. The power control loop receives the power reference
from the delink voltage loop, compares with the actual power in the line and adds the
feedforward generator power that comes through the serial port communication. Such
proportional control generates the signal that goes into the fast current controller
executed every 128 (i^. The delink voltage feedback loop receives the voltage information
from the serial port communication, compares with the reference value (300 V) and
generates the power reference with a PI controller. The INT2 service routine executes at
the end the serial port communication and the logging service, as already illustrated for the
Processor Board #1.
The interruption INT3 corresponds to the control software that must be executed
each 1.024 sec and is triggered by software in the task 1 with the loop counter lpc3. This
interruption is identified by the software code embraced by c_int04() and comprehends the
fuzzy controller FLC-1 which performs the speed optimization search. The fuzzy
controller FLC-1 is enabled by the signal enable_flcl. Such fuzzy controller comprises two
fuzzy algorithms, one for scheduling the gains for pu operation and the actual optimization
algorithm. Initially it is calculated the gains for variation in power (KPO) and variation in
speed (KWR), What dictates the required gains, is the current value of the generator
speed, as the example of rule #2 is shown:
/* Rule #2 IF wr=PSB THEN KPO=KPOSB; KWR=KWRSB */
if (ww == 3 11 ww = 4)
{
KPO += KPOSB*uw2;
KWR += KWRSB*uw2;
}
86
-------
The index ww indicates which membership function needs to be evaluated and
which roles are fired. The gains KPO and KWR are reset at the beginning of each cycle
and after the evaluation of the roles the inverse of KPO and KWR are computed to get the
per-unit values of variation of power and last variation of speed as
DPI = Delpo *1NV_KP0;
LDW = D WREF* INVJC WR;
where Delpo and DWREF are the actual variation of power and speed, DPI and LDW are
in pu. The variation of power DPI passes through a minimum step computation in order
to get the input for the fuzzy controller (DP).
The fuzzy optimization algorithm receives the inputs LDW and DP, the interval
where such inputs are on the universe of discourse are indicated respectively by the
indexes qq and pp, e.g. if the variation of power is positive small (PSMA), pp is either I or
2 and the membership function evaluation is given as
else if (DP <= 0.125)
I
pp = 1; /* Membership value of PSMA */
udpl = 8*DP;
}
else if (DP <= 0.3)
{
pp = 2; /* Membership value of PSMA */
udpl ='1.71428571429 - 5.71428571429*DP;
}
After the calculation of the membership functions for DP and LDW, the
correspondent rules are fired. The example below shows the computation of rule #11;
/* Rule #11 IF Dw=P AND DP=PBIG THEN Dwref = NBIG ~/
if (qq == 1 II qq = 2)
{
if (pp == 4 II pp == 3)
{
87
-------
Dwrefpu += DWRNBIG*MIN(udw2,udp 1),
}
}
where qq = 1 or qq = 2 indicates that the last variation of speed is positive and pp = 4 or
pp = 3 indicates that the variation of power is positive big The output variation of speed
in pu (Dwrefpu) is reset at the beginning of every sampling time and is added with every
rule that is fired, then it is multiplied by the scaling gain and the actual variation of speed is
limited to ensure stable operation. Such a step is divided into mini-steps and a ramp
generator builds the output of the speed reference. Such a speed reference is sent to the
Processor Board #1 via the serial port communication for controlling the machine with the
fuzzy-PI FLC-3
6.3 Host Communication Program
The host communication program down-loads the two DSP programs in the
correspondent DSP boards and waits for the user command There is a screen with several
options which can be accessed by typing the correspondent numbers, as indicated in Fig.
52.
Any time that "0" is pressed the menu screen is displayed on the screen. To switch
between DSP boards the selection "22" should be typed, and to quit the host program to
come back to the DOS the selection is "15."
When the system is started up, the host communication has a default value of 50 V
for the dc-link reference voltage which can be gradually increased up to 300 V by
choosing the DSP board #2 and using the selection "11." After the dc-link is established
the excitation current in the induction generator can be established with the selection "10"
and a minimum speed should be established in the induction generator by the selection "9."
The dynamometer impresses a regenerative torque, in accordance to the voltage command
coming from the wind turbine equation through the DAC output of the Processor Board
#1 The wind velocity is chosen with the selection "8." When this initial start-up is done
88
-------
Host Communication with DSP 1
0 - Print this screen
1 - Start acquisition
2 - indicate speed
3 - Start-up
4 - Indicate 4 channel readings
5 - VoltsHz Operation
6 - Wind transient
7 - Enable FLC2
8 - Enter wind velocity (nv's)
9 - Spdref
10 - Idsref
11 - Iqsref for DSP1 and Vdref for DSP2
12 - Halt DSP program
13-Reset DSP program
15 - Quit
22 - Switch DSP1 <==> DSP2
Enter Selection for DSP 1;
Fig. 52 Host communication program screen interface
the choice "3" should be selected and the reference speed co/ is switched from the user
interface to the fuzzy controller FLC-1. As the wind velocity is changed by the choice "8"
the fuzzy controller FLC-1 automatically searches the new reference speed m order to
optimize the generated power. The flux optimization by the fuzzy controller FLC-2 can be
either started by the selection "7" which freezes the last reference speed and disables FLC-
1, or can be implemented in the wind transient selection "6." Such a wind transient uses a
look-up table for a pre-defined wind profile and, after the power optimization is over, the
flux optimization takes place. Finally the choices "1" and "4" are used for debugging. The
selection "1" writes a file with the pre-defined variables of the DSP programs ("dspl .c"
and "dsp2,c"), and the selection "4" displays on the screen four variables which were
chosen during the program compilation.
89
-------
7. Experimental Evaluation
The wind energy system was evaluated in a laboratory set-up with the design
parameters shown in table 9. The line-side voltage is a standard 220 V three phase utility
system, connected to the line-side inverter through the indicated series inductance. The
induction machine is a 3.5 kW standard class B machine. Of course, the machine power
cannot be fully generated because the dc-link loop voltage limits the maximum
fundamental phase voltage at 0.353vd. The dc-link voltage was kept to 300 V, because the
maximum voltage rating for the IGBT Powerex modules was 600 V. For the final 200 kW
field installation, however, it is necessary to use 1200 V devices and the dc-link should be
set to 500 V.
Table 9 Wind energy system design parameters
SYSTEM PARAMETER
VALUE
Line-side voltage (Vac)
220 V
Series inductance (Ls)
500 (lH
Dc-link voltage (v
-------
Auxiliary F
Power Jr
Supply I
ISOLATION
r . TRANSFORMER
VARIAC ^
u. 220 220 Vu,
? =; | 300 V
-44^
120 V,
DSP CONTROL
AND HOST PC
DB
DB
MACHINE
SIDE
INVERTER
LINE
SIDE
INVERTER
Fig. 53 Wind energy system experimental set-up
-------
used for charging the capacitor and testing the dc-link loop voltage control (the power
supply can impose a higher voltage which is then stabilized with the dc-link loop control).
The reason for charging the capacitor with the auxiliary power supply is that the inverters
could not be electronically disabled, i.e. the inverter firing pulses could not be all kept in
the "off' conditions during the capacitor delay time charging. If the inverter firing pulses
are not kept "off," the Hanning chip imposes a 50% duty-cycle pattern in the devices,
which acts as a short-circuit for the line-side voltage, with corresponding damage in the
devices. For the final 200 kW installation, a simple flip-flop circuit addressed by a C-
pointer in the DSPLINK bus can electronically enable/disable the inverter firing pulses
For safety reasons, two dynamic brakes were connected to the dc-link bus, in case of mal-
function of the voltage loop. For the final installation the dynamic brakes can be integrated
in the Powerex module, since they have a seventh available IGBT which can be turned on
by an operational amplifier comparator circuit,
Each inverter was manually enabled by a mechanical switch and the series dc-link
resistor was also bypassed by a manual switch There was also a manual switch to enable
the dynamometer interface to apply regenerative torque on the machine shaft. Of course,
with the final integration of the protection/fault interface and the electronic enable/disable
of the inverters the following procedure can be automatic and the auxiliary power supply
can be eliminated. The following steps indicate the system start-up, as done in the
laboratory:
1. Turn on the PC.
2. Tum on the inverters and interfaces circuitry.
3 Turn on the dynamometer but keep it disabled with the manual switch.
4. Turn on the auxiliary power supply and charge 53 V at dc-link.
5. Set the variae initially at 10 V position
92
-------
6. Turn on the main circuit breaker. No current will flow because the dc-link has a
higher voltage.
7. Type "host" in the correct directory (this downloads the DSP programs and establishes
50 V reference voltage for the dc-link voltage loop).
8. Immediately after the two DSP programs are running, enable the line-side inverter by
turning on the mechanical switch of the inverter. (This will establish the dc-link voltage
at 50 V )
9. By switching to DSP2 (option #22) and typing the dc-link voltage reference (option
#11), repeatedly increase the auxiliary power supply voltage, then the dc-link voltage
reference and the variac position until the dc-link voltage is stabilized at 100 V and the
variac is at 50 V.
10. Hold the auxiliary power supply at 100 V, keep increasing the dc-link voltage
reference in option #11 up to 165 V and fix the variac voltage at 120 V (line-to-line).
11. When the line-side inverter is fully operating to keep the dc-link voltage at 165 V the
machine excitation current can be established. Switch to DSP1 (option #22). Type the
excitation current (option #10), initially with a small value of 3 A, and immediately
enable the machine-side inverter with the mechanical switch. Continuous current must
flow in the machine to establish some amount of flux The rated value can then be
established (9,5 A).
12. After the flux is established, some minimal speed must be impressed in the system.
Switch to DSP2 and type 350 RPM in the option #9. The machine has to rotate at that
speed.
13 Enable the dynamometer to impose a regenerative torque. The amount of the
regenerative torque is dictated by the wind turbine torque equation in the DSP. At this
point some small power starts to flow from the machine to the line-side.
93
-------
14. Impose at least 4.75 ra/sec as wind velocity in the option #8 (DSP1). The amount of
generated power should increase a little depending on the reference speed typed in
option #9 (DSP2). The dc-link voltage loop can now be established to a higher value
(300 V), by keying the reference in the option #11 (DSP2). The auxiliary power
supply can be switched off at this time.
15. When the system is fully operating, with the dc-link voltage set to a higher value, the
dynamometer imposing some regenerative torque, the wind velocity set around 5
m/sec and the generator speed set around 550 RPM, the system can be transferred to
FLC-1 control by typing the option #3 (DSP1), which will ask if the initial start-up is
over, transferring the command of the generator reference from the keyboard to the
FLC-1 control.
16. At this point the wind velocity (option #8) can be changed and FLC-1 will
automatically set the reference speed with the fuzzy algorithm search.
The shut-down steps, as done in the laboratory, followed the procedure below:
1. Impose small wind velocity (like 4.75 m/sec)
2. Turn-off the mechanical switches for the system exacdy in the following sequence, one
after the other:
3. Disable the dynamometer
4. Disable the machine-side inverter
5. Disable the line-side inverter
6. Disconnect main-circuit breaker
7. Type the "quit" option in the host communication program
8. After the dc-link voltage reaches zero turn off the inverters and interfaces circuitry
9. Turn the dynamometer off
10. Turn the PC off
94
-------
The following data present the experimental results taken for the system evaluation
and validation. Depending on the convenience the results were acquired in several ways:
some results were taken via oscilloscope GPIB, internal software variables were sent to
the DAC for visualization in the oscilloscope and, photographs were taken from the scope
screen For long time transients, the software variables were exported through the "log
service" for an ASCII file and later processed for plotting. In order to prepare the figures
for the steady-state charts and tables, the required variables were observed and noted in
the laboratory notebook.
The wind turbine system was carefully tested and systematically integrated step-by-
step In the beginning the DSP Processors were tested, and the analog IO boards were
checked together with the current and voltage interface circuits. The Hanning interface
board was tested and the main interruption from each Hanning chip was connected to «ach
of the DSP boards. The Hanning PWM IC signals were evaluated by sending the PWM
signals through a digital circuit that simulated the operation of a three-phase inverter. With
a low-pass-filter connected at the output of such digital circuit the fundamental waveform
was observed. Fig. 54 shows the three outputs of the Hanning chip at a constant frequency
fe = 50 Hz, The third harmonic can be observed that the Hanning chip intentionally
adds to the fundamental. Such third harmonic helps to improve the PWM efficiency and
does not circulate in the machine phase windings because the neutral is isolated.
95
-------
1 V/div
angle 0,
3 Phase
Voltages
. 5 ms/div
Fig. 54 Evaluation of Hanning PWM signals with low-pass-filter at frequency of 50 Hz
96
-------
Fig. 55 shows one output of the Haiming chip into the Low-pass-filter, but with the
frequency being stepped between 25 Hz and 12.5 Hz. Such tests were convenient not only
to test the Hanning chip, but also to get familiar with the interruption service routine
software and to implement a Host communication program that could control the
operation of the Hanning PWM. After checking the operation of the PWM Hanning circuit
for each board, the cable driver interface was tested and the Hanning chip outputs were
connected to the IGBT power inverter interface. The dead-time of 4 ^sec was observed as
indicated by the photo of the gate drive pulses for the upper and lower devices indicated in
Fig. 56. The operation of the IGBT modules was carefully observed. The collector-to-
eraitter voltage switching and phase-current in a RL load are indicated in Rg. 57. Fig. 58
shows the pulse-width-modulation of an IGBT phase-leg device for the upper and lower
devices in a V/Hz operation of the inverter, and Fig. 59 shows the phase-voltage and
phase current for the RL load (top) in V/Hz. Hie natural imbalance of the machine is
depicted in Fig. 60 in V/Hz operation. Initially the inverter for the machine side was put to
work. The V/Hz operation was very useful for debugging the current sensors, the 3
transformation and the inverse-vector rotation. The speed sensor was also debugged and
calibrated with the speed indication in the dynamometer. After straightening out such
points the current controller with i^* and i** feedback control loop was implemented and
the torque control loop was added. Fig. 61 shows the torque control response, where the
top signal is the speed, kept constant at 1000 RPM (400 RPM/div) with the dynamometer
in speed control mode, and the bottom signal is the torque current variable (i^* = 8A)
variable from the torque loop (Te* = 4 Nm). Fig. 62 shows the three phase machine
currents i», it,, and ic (10 A/div) for the above conditions, and Fig. 63 depicts the phase
voltage v, (100 V/div) and current i« (10 A/div). It was also observed that the de-link
voltage rises with negative torque command (regeneration mode). After the torque control
loop was integrated, the speed control loop was added. Initially a regular PI was
implemented in order to debug the system in speed control. Of course, the dynamometer
was changed to torque control mode for such operation.
Fig. 64 shows the speed response of a PI with the machine shaft
97
-------
1 V/div
: j/\
T
|
-T" —
A j
r !
r — i
1 Jr i
-
r~\
\
1
i
j
1
i I
i /
i /
i /
i /
1 i \ /
/ ' \ Is>
j i \ I \
i 1 i j j
/ i \ / J
/ 11 1 1
\ \
/ 1 1 / '
J i V... J \
0" 1 -rn-ixT j
i i
i i
i i
. — i-
V... J
i
!
i
,
angle 0t
Phase Voltage
i 25 ms/div,
r »1
Fig. 55 Evaluation of Hanning PWM signals with low-pass-filter and stepped frequency
between 25 Hz and 12.5 Hz
98
-------
Fig. 56 Dead-time on the upper and lower signals for the IGBT phase-leg device
Fig. 57 Collector-to-emitter voltage switching of IGBT (top) and phase-current (bottom)
99
-------
Fig. 58 Pulse-width-modulation of an IGBT phase-leg device, upper device (top) and
tower device (bottom)
Fig. 59 Phase-voltage (top) and phase-current (bottom) for RL load in V/Hz
100
-------
2A /div
3 Phase
Currents
10 ms/div
Fig, 60 Three-phase currents in V/Hz showing the natural machine imbalance
101
-------
. 61 Torque control loop, speed (top) at 1000 RPM (400 RPM/div) and torque current
iy* (bottom) at 8A (6 A/div) with Te* = 4 Nra.
Fig. 62 Three phase currents ia, ib> and i* (10 A/div) for torque loop control
102
-------
1
Fig. 63 Phase voltage va (100 V/div), and phase current ia (10 A/div)
for torque control loop
liiiiaiBRH
mwwSm
< •
Fig. 64 Speed response of a PI controller, (top) speed response (200 RPM/div), (bottom)
control effect of the PI loop
103
-------
loaded with 5 Nm. The speed profile is a square wave from 650 RPM to 850 RPM The
signal on the top is the speed response (200 RPM/div) and the signal on the bottom is the
control effect of the PI controller.
The speed fuzzy controller (FLC-3) was implemented and some fine-tuning was
necessary for the gains KE, KCE and KTE. Fig. 65 shows a somewhat underdamped
response from the fazzy controller for a preliminary set of gains and Fig. 66 shows a better
response with the final gains. Even though the response has a very small overshoot it was
decided to keep such gains because in the final system it is necessary to have a fast and
robust response for the speed controller against torque disturbance.
In order to test the robustness of the speed fuzzy control (FLC-3), some
disturbance had to be added to the dynamometer torque command. Therefore, the
dynamometer control interface as described in section 5 was implemented and the torque
command could then be programmed as well. Fig. 67 shows the speed trace on the top at
400 RPM (200 RPM/div) and the load torque trace on the bottom at an average of 5 Nm
added with a square wave of 2 5 Nm peak-to-peak. The speed response is nicely robust,
despite the disturbed shaft torque. Fig. 68 shows the speed trace on the top at 900 RPM
(200 RPM/div) and a sinusoidal torque profile that goes from 4,35 Nm to 5.65 Nm with 5
Nm in the average and the Fig. 69 shows the speed response at 800 RPM with a square
wave intentionally corrupted by random noise added to an average torque of 5 Nm, In all
such situations the fuzzy controller operated very well, keeping the speed robust and
maintaining a stable drive operation.
The above evaluation steps completed the hardware and software validation for the
machine side inverter. The next task was to operate the line-side inverter , and a similar
procedure was followed. The line-side inverter is a DVC based control. Therefore, all the
connections, voltage and current sensors and line-side phase voltage sequence were
checked to ensure a correct unit vector production. Fig. 70 shows the sine and cosine
waveforms calculated from the line-side voltages used for inverse-vector-rotation. Fig. 71
shows the sine wave and the reference angle (9e") for vector-rotation. It seems that the
sine wave is inverted, this was necessary to make the line-side generated power a positive
variable, which is more convenient for visualization of system's operation
104
-------
Fig, 65 Speed fuzzy control (FLC-3) with preliminary gains, (top) speed response
(200 RPM/div), (bottom) control effect of the FLC loop
Fig. 66 Speed fuzzy control (FLC-3) with Final gains, (top) speed response
(200 RPM/div), (bottom) control effect of the FLC loop
105
-------
Fig. 67 Robustness of speed FLC with square-wave load torque, (top) speed at 400 RPM
(200 RPM/div), (bottom) load torque from 3,75 Nra to 6,25 Nm with 5 Nm average
Fig. 68 Robustness of speed FLC with sinusoidal disturbance torque, (top) speed at 900
RPM (200 RPM/div), (bottom) load torque from 4.35 Nm to 5.65 Nm with 5 Nm average
106
-------
Fig. 69 Robustness of speed PLC wiih square-wave load torque intentionally corrupted by
noise, (top) speed at 800 RPM (200 RPM/div), (bottom) load lorque from 3.75 Nm to
6,25 Nm with 5 Nm average
107
-------
Fig. 70 Unit vector for line-side DVC inverter, sine arid cosine waveforms
Fig. 71 Sine waveform and reference angle for line side DVC inverter
108
-------
The voltage outputs of the line-side inverter must instantaneously track the line-
side phase voltages due the i^" and i*" current control loop. Fig. 72 shows the match of
the SPWM inverter phase-a voltage (100 V/div) and the corresponding line-side phase-a
voltage (200 V/div). Fig. 73 shows the agreement of the line-side voltage v, (150 V/div)
and line-side current i, (5 A/div) for i^* loop control, demonstrating that the system is
operating under unity power factor. It was also observed that for positive command
the voltages and currents were in-phase as Fig. 73, and for negative i
-------
L
Fig. 72 SPWM line-side inverter phase-a voltage waveform voltage (100 V/div),
and line-side phase-a voltage (200 V/div)
Fig. 73 Current control loop (1^*) for line-side inverter, (top) line-side voltage
va (150 V/div), (bottom) line-side current i„ (5 A/div)
110
-------
Fig. 74 Positive power command, line-side vt voltage (60 V/div)
and current ia (3 A/div)
ilfiHH
—tMPHWi—M—Mil
.
Fig, 75 Negative power command, line-side va voltage (60 V/div)
and current i» (3 A/div)
111
-------
ig. 76 Power reference Pn* signal (bottom) at 400 W (15 W/div)
and reference signal (top) at 4A (0.5 A/div)
Fig. 77 Raw phase-a inverter voltage on top (100 V/div),
and raw phase-a current on bottom (5 A/div)
112
-------
Fig, 78 Dc-link voltage transient response for
a square wave command from 240 V to 265 V
Fig. 79 Dc-link voltage ripple of 1 V peak-to-peak, the average voltage is 300 V
113
-------
The wind turbine programmable torque equation was checked to correctly control
the dynamometer and the feedforward power from the machine-side inverter was added to
the line-side power control loop (such, signal is transferred with the serial port
communication as described in sections 5.1 and 5.2). For the start-up of the double pwm
converter the dc-link capacitor was initially charged from the peak of the line-side utility
voltage through a series resistance, which was later by-passed by a mechanical switch The
dc-link voltage was initially established with a lower dc-link voltage reference command
With the dc-link voltage established the excitation current (i^) in the induction generator
could be established and a minimum speed could then be set up for generation with the
dynamometer impressing a regenerative torque on the shaft. Since the system started to
generate some power to the utility side the dc-link voltage could be commanded to rise
up to 300 V, All those above procedures constitute the initial start-up of the system which
can be transferred for the fuzzy optimization of speed and flux as described in the control
coordination transition in section 2.4. Fig. 80 shows the raw machine voltage and current
with the double-pwm converter. The dc-link voltage is at nominal voltage (300 V), the
dynamometer is set to a constant regenerative torque of 3.5 Nm, and the speed is kept
constant by the speed fuzzy controller at or* = 550 RPM One phasor diagram for Fig. 80
can easily show that the machine is in regenerative mode with negative slip frequency.
With the double pwm converter fully operating, and the wind turbine
programmable torque commanding the dynamometer to work in the regenerative torque
mode, it was possible to trace the static characteristic curves of the system as shown in
Fig. 81. Such chart contains a family of curves for turbine power, turbine torque and
generated power for varying wind velocity and several fixed generator reference speeds.
As the wind velocity increases for a fixed generator speed the corresponding power or
torque increases tending to a saturation level. The rate of increasing torque or power is
sharper as the generator speed increases and the saturation level gets flat as the generator
speed decreases. For a fixed wind velocity the corresponding power or torque increases,
reaching a maximum and then decreases in accordance to the turbine power coefficient,
114
-------
Fig. 80 Induction generator voltage (100 V/div) and current (10 A/div),
raw signals, for double-pwm converter operation with speed set
to = 550 RPM and regenerative torque constant at 3.5 Nro.
-------
9
wa
2000
1670
g
1340
(4
s
1010
o
PL,
680
350
20
20,0
16,7
1
13,4
£
g
10.1
a
fVj
p
6.8
H
3.5
2.0
950
815
$
680
Pi
545
O
410
Ch
275
140
5
©;=i850.
a' =
700 RP3
ffl," = 5;
I
0 RPM
¦
03," = 41
'0 RPM
¦ 1 '
5 6 7 8 9
WIND VELOCITY (m/sec)
10
Fig. 81 Wind turbine static characteristics, (a) Turbine Power,
(b) Turbine torque, and (c) Generated Power
116
-------
Any transient in the wind velocity reflects in changing the generated power in the
utility side. The dc-link is a kind of elastic coupling, any difference in the instantaneous
power flow in the double-pwm converter reflects in some change in the bus voltage. Of
course, the feedforward power connection minimizes the dc-link voltage variation. Fig. 82
shows the dc-link voltage response when the wind velocity jumped from 4.25 m/sec to 7.5
m/sec. The transient in the dc-link voltage went from 300 V to 306 V, settling down in
accordance to the picture. Fig. 83 shows the dc-link voltage response when the wind
velocity stepped down from 9 m/sec to 7 m/sec. The transient in the dc-link voltage went
from 300 V to 298 V, settling down in accordance to the picture.
The host program permits that after the initial start-up of the double pwm
converter the reference speed be commanded from the fuzzy controller FLC-1. Therefore,
by changing the wind velocity the reference speed automatically changes to the new
operating point. The host communication program also allows the flux optimization search
with the fuzzy controller FLC-3 by freezing the last reference speed from FLC-1 and
decreasing the excitation current to boost a little more the generated power. Fig. 84 shows
the steady state performance enhancement control of wind turbine -at several operating
points. After the initial start-up the system was operating at A. As the speed reference
command switched from the host communication interface to the fuzzy controller FLC-1
the reference speed increased, and the optimized point at B (for wind velocity of 5 m/sec)
was reached. The flux search further enhanced the power generation by boosting to C.
The figure indicates the power in D when the wind velocity stepped from 5 m/sec to 7
m/sec. The rated flux was established and the fuzzy controller FLC-1 was enabled to
search the optimum point at E and again the flux optimization boosted the generated
power to F. The path G-H-I indicates the control operation when the wind velocity was
stepped from 7 m/sec to 6 m/sec.
Fig. 85 shows system performance data during the on-line search of fuzzy
controller FLC-1, when the wind velocity is stepped from 4.75 m/sec to 7.25 m/sec. The
variation of power in terms of speed for such conditions is given in Fig. 86. Fig. 87 shows
the system performance when the wind velocity is stepped down from 9 m/sec to 7 m/sec,
and the variation of power in terms of speed for such conditions is given in Fig. 88.
117
-------
Fig. 82 Transient in the wind with dc-link voltage response, (top) dc-link voltage at 300 V
(30 V/div), (bottom) wind velocity stepping up from 4.25 m/sec to 7.5 m/sec
Fig. 83 Transient in the wind with de-link voltage response, (top) dc-link voltage at 300 V
(30 V/div), (bottom) wind velocity stepping down from 9 m/sec to 7 m/sec
118
-------
155
100
Vw = 5 ra/s
»«..»;
45
J I I L-J I 1 I . I- 1 ¦ I - I . I .-I -I . J -1 .. I I I III ,,1 ml ml I ,1
450 500 550 600 650 700 750
GENERATOR SPEED (RPM)
Fig. 84 Fuzzy logic steady state performance enhancement control of wind turbine at
several operating points
119
-------
7.25 m/sec
4.75 m/sec
Wind Velocity
a)
9.5 A
4.25 A
Excitation Current (i,J
b)
d)
683 RPM 679 RPM 677 RPM
640 RPM^ ^ m _
607 RPM <*.-W ~ 9 ^
,npPw 664RPM 678RPM 675 RPM 676RPM
540 RPM, 9 623 rpm *
475 RPM^ Generator Speed c)
411
247.54 W 252.12 W 251.65 W 252.27 W 252.33 W
232.6W ¦- —¦ — *-
221JWp--"Ml.l8W 249.33 W 248.55 W 250.40 W 250.17 W
/
/ x Generated Power
/
86.5 W M
-3.52 A J
I
| /Torque Current (y
I /
I -7.38 A -7.1^5 A y T ^3_ T e)
1; 7 55 A,'" ^-7.35 A •6 M A "6'89 A -«-67 -6.86
¦» /
-8.51 A
Fig. 85 Performance data for step-up wind velocity, (a) Wind velocity (m/sec), (b)
Excitation current (A), (c) Generator speed (RPM), (d) Generated power (W), (e)
Torque current i^ (A)
120
-------
o
300
250
200
150
100
50
-
-
1
- 1
i
. „
, L ,
1 —
400
450 500 550 600 650 700
SPEED (RPM)
Fig. 86 Power changing as generator speed goes up for wind velocity stepping-up from
4,75 m/sec to 7.25 m/sec
121
-------
9 m/sec
7,15 A
Wind Velocity
gf 7 m/sec
9.5 A
Excitation Current (ij
a)
b)
f 1195 RPV1
\
\
\
Generator Speed
*866 RPM
-* 828 RPM 8K1 RpM „„„„„
„„DDy* *—^ 794 RPM 789 RPM
5D2 KxlVl 007 DTJTVi" "w*— - -»• _ JBfc—. _
837 RPM 819 RPM 801 RPM*" -• -^pM
786 RPM 785 RPM
c)
^ 411.12 W
V
V
V
V
\
\
Generated Power
/
4
d>
230.31 W
229.36 W 235.71 W 238£5 VV
233 51 W 237-2SW 242.78 W 242.37 W 242.18 W
227.5 W
-5.37 A
-4.86 A
/ ^ -5.12 A
j -5.01 A "*" *~
/
-5 71 A 1 -5.72 A
Torque Current (i^)
¦5.22 A
-5.55 A
-5.41 A
-5.65 A
-5.69 A
5.79 A -5.81 A
Fig. 87 Performance data for step-down wind velocity, (a) Wind velocity (m/sec), (b)
Excitation current i^ (A), (c) Generator speed (RPM), (d) Generated power (W), (e)
Torque current i,* (A)
122
-------
(4
O
tu
450
400
350
300
250
200
1200 1100 1000 900 800 700
SPEED (RPM)
Fig. 88 Power changing as generator speed goes down for wind velocity stepping-down
from 9 m/sec to 7 ra/sec
123
-------
The vector control on the line side makes the system operate under unity power
factor all the time, since ids" = 0. As the fuzzy controller FLC-1 searches the'optimum
reference speed the controller pumps more power to the utility side by increasing the
active current i^ Fig. 89 shows one step during a search transient when the wind velocity
varied from 4 m/sec to 7 m/sec. For such conditions the reference speed is Or* = 500 RPM
and the line side generated power is 155 W. There are two snapshots on the same picture,
the top one shows the line side voltage v„(l20 V/div) and line side current i, (0 5 A/div)
whereas the bottom one is the machine current ia (10 A/div) for the same conditions. Fig
90 shows the second step where the reference speed is 0/ = 625 RPM and the line side
generated power is 220 W. The top snapshot shows the line side voltage va and line side
current ia, by comparing to the previous figure the increase in power is obvious. The
bottom snapshot is the machine current ia, showing the increase in the machine speed
Finally Fig. 91 shows a third step where the reference speed is cix = 750 RPM and the line
side generated power increased to 270 W which can be observed by the higher line-side
current. The bottom snapshot shows the machine current i, with the corresponding higher
operating speed.
Another run of power optimization was made for the wind velocity changing from
6 m/sec to 9 m/sec. The snapshots are recorded in the pictures given in Fig. 92, Fig. 93,
and Fig. 94. The first step shown in Fig. 92 displays the line-side voltage (120 V/div) and
current (2 A/div) as well as the machine current (20 A/div). The reference speed is Or =
700 RPM and the generated power is 370 W. As the machine speed changes to ©,* = 875
RPM the power increases to 575 W as indicated by Fig. 93 and as the machine speed
changes to 0/ = 1050 RPM the generated power increases to 840 W as shown in Fig. 94.
124
-------
». 89 First step of speed optimization by FLC-1 for wind velocity varying from 4 m/sec
to 7 m/sec, speed reference is 500 RPM, generated power is 155 W,
(a) Line-side voltage v, (120 V/div) and line-side current ia (0.5 A/div),
(b) Machine current i, (10 A/div)
125
-------
Fig, 90 Second step of speed optimization by FLC-1 for wind velocity varying from 4
m/sec to 7 m/see, speed reference is 625 RPM, generated power is 220 W,
(a) Line-side voltage va (120 V/div) and line-side current ia (0.5 A/div),
(b) Machine current ia(10 A/div)
126
-------
IgtfBBKffiBBBKft
iramiSBBisiiB
fllBBBBBBBBB
Fig. 91 Third step of speed optimization by FLC-1 for wind velocity varying from 4 m/sec
to 7 m/sec, speed reference is 750 RPM, generated power is 270 W,
(a) Line-side voltage va (120 V/div) and line-side current i, (0.5 A/div),
(b) Machine current i« (10 A/div)
127
-------
Fig. 92 First step of speed optimization by FLC-1 for wind velocity varying from 6 m/sec
to 9 m/sec, speed reference is 700 RPM, generated power is 370 W, (a) Line-side voltage
va (120 V/div) and line-side current i. (2 A/div), (b) Machine current ia (20 A/div)
128
-------
Fig. 93 Second step of speed optimization by FLC-1 for wind velocity varying from 6
m/sec to 9 m/sec, speed reference is 875 RPM, generated power is 575 W, (a) Line-side
voltage v, (120 V/div) and line-side current i» (2 A/div), (b) Machine current i, (20 A/div)
129
-------
Fig. 94 Third step of speed optimization by FLC-1 for wind velocity varying from 6 ni/sec
to 9 m/sec, speed reference is 1050 RPM, generated power is 840 W, (a) Line-side
voltage Va (120 V/div) and line-side current i. (2 A/div), (b) Machine current ia(20 A/div)
130
-------
When the power optimization with FLC-1 is finished the flux optimization by FLC-
2 takes over Fig. 95 shows the steady-stale chart for each search step, when the wind
velocity is constant at 6 m/sec and the generator speed is at the optimum point of 623
RPM The excitation current is decreased from 9.0 A (rated) to 5.0 A, and the generated
power boosts from 190 23 W to 210.29 W This is an enhancement of 10.54 % in the
power generation, which for a large wind system installation is very significant. Of course,
the enhancement is better for light load, e g. for lower wind velocities. As the excitation
current (i
-------
/
Wind Velocity (6 m/sec)
Generator Speed (623 RPM)
-A —* A *
a)
b)
210.29 W
Generated
Power
201.16 W
195.5 W
207.01 W
190.23 W
c)
9.5 A
8.5 A
7.5 A
6.5 A
5.5 A
Excitation
Current (i*) e)
5.0 A
4.5 A
-5.02 A
£ - -5.54 A
-6.15 A
-6.73 A
Torque
Current (i^)
-7j8 A , -7.63 A -7.S5 A
f)
Fig. 95 Fuzzy logic flux enhancement optimization of induction generator,
(a) Wind velocity (m/sec), (b) Generator Speed (RPM), (c) Generated Power (W),
(d) Excitation Current i* (A), (e) Torque current (A)
132
-------
Fig. 96 First step of flux optimization by FLC-2 for wind velocity at 6 m/sec, generator
speed = 623 RPM, (a) Machine voltage v4 (150 V/div) and current i. (10 A/div),
(b) Phasor diagram for such operating conditions
133
-------
* Fig. 97 Second step of flux optimization by FLC-2 for wind velocity at 6 m/sec, generator
speed = 623 RPM, (a) Machine voltage v, (150 V/div) and current i, (10 A/div), (b)
Phasor diagram for such operating conditions
134
-------
(b)
Fig. 98 Third step of flux optimization by FLC-2 for wind velocity at 6 m/sec, generator
speed tik = 623 RJPM, (a) Machine voltage v, (150 V/div) and current ia (10 A/div),
(b) Phasor diagram for such operating conditions
135
-------
Fig. 99 Wind transient step-up from 4.75 ra/sec to 7.25 m/sec, (a) Wind velocity (m/sec),
(b) Generator speed (RPM), (c) Line-side generated power (W),
(d) Dc-link voltage (V)
136
-------
A system's step down response on the wind velocity is plotted in Fig. 100. The
wind velocity jumps from 9 m/sec to 7 m/sec as Fig. 100 (a) shows. The generator speed
goes down as shown in Fig. 100 (b) and the power stabilizes at the optimum point for the
new wind velocity as shown in Fig. 100 (c). The dc-link voltage response for such
transient is recorded in Fig. 100 (d) Despite the fact that step responses are usually used
for testing loop responses, a wind velocity will never vary so abruptly. Therefore, a ramp
wind velocity profile as given in Fig. 101 is closer to the real wind variation. Fig. 101 (a)
shows the wind going up from 6.125 m/sec to 9.125 m/sec in a ramp profile. The speed
goes up as Fig. 101 (b) shows, while the wind velocity increases, showing that the fuzzy
controller FLC-1 is capable of tracking a slowly random variation of wind, as usually
happens in the nature. As the wind velocity reaches the maximum, the fuzzy controller
FLC-1 automatically decreases the step size and the wind system settles down to the new
operating condition with the generated power as shown in Fig. 101 (c) The dc-link
voltage variation during such transient is shown in Fig 101 (d)
The flux optimization by fuzzy controller takes over the system after the power
optimization settles down as already discussed. Fig 102 shows the optimization transient
for a wind velocity constant at 5 0 m/sec, the generator speed was frozen at 520 RPM.
The excitation current decreased from the rated value as shown in Fig. 102 (c) and the
generated power boosted as shown in Fig. 102 (d). Another flux optimization for wind
velocity at 6,5 m/sec is indicated in Fig. 103 (a) with the generator speed at 675 RPM as
shown in Fig. 103 (b). The excitation current and the power boost are indicated in Fig.
103 (c) and Fig. 103 (d).
A time domain sequencing of the fuzzy controllers FLC-1 and FLC-2 is shown in
Fig. 104. The wind profile in Fig. 104 (a) goes down sinusoidally from 9.35 m/sec to 6.35
m/sec with a superimposed wind vortex. Such wind signal was made possible by a look-up
table in the host communication interface. The system performed very well, as indicated
by the generator speed in Fig. 104 (b) and the power in Fig. 104 (d). After the settling
down of the power optimization search the flux excitation current decreased as shown in
Fig. 104 (c), with a corresponding small power boost indicated in Fig. 104 (d).
137
-------
i % 8.0
L300
ll
g & 950
V
«
1 1
3.
if
Time (sec)
12
16
b)
20
Pig. 100 Wind transient step-down from 9 m/sec to 7 m/sec, (a) Wind velocity (m/sec),
(b) Generator speed (RPM), (c) Line-side generated power (W),
(d) Dc-link voltage (V)
138
-------
9.0
*o w'
| M 7.5
6.0
1350
900
BIT
c E.
Vi
450
700
100
303
297
0
4
12
Time (sec)
Fig. 101 Wind transient ramp-up profile from 6.125 m/sec to 9.125 m/sec, (a) Wind
velocity (m/sec), (b) Generator speed (RPM), (c) Line-side generated power (W),
(d) Dc-link voltage (V)
139
-------
5.3
0 4 8 12 16
Time (sec)
Fig. 102 Flux excitation current programming by FLC-2, (a) Wind velocity at 5 m/sec, (b)
Generator speed at 520 RPM, (c) Excitation current u, (A),
(d) Line-side generated power (W)
140
-------
6.8
£
-a -
,i >¦
? 'i
u
>
w a.
S *
u
E "3
a
o H.
t/5
§ <
cj c
U
^ t
k b
r i
u !>
"O >
V ii
.s *
6.5
6.2
850
425
0
10.0
5.0
0.0
315
270
225
a)
Time (sec)
12
16
Fig. 103 Flux excitation current programming by FLC-2, (a) Wind velocity at 6.5 m/sec,
(b) Generator speed at 675 RPM, (c) Excitation current i^ (A),
(d) Line-side generated power (W)
141
-------
10.0
17 25
Tune (sec)
34
42
Fig. 104 Time domain sequencing of FLC-1 and FLC-2, (a) Wind velocity sinusoidally
going down from 9.35 m/sec to 6,35 m/sec with wind vortex, (b) Generator (RPM), (c)
Excitation current u, (A), (d) Line-side generated power (W)
142
-------
The system was run for several operating points with the optimization of both
fozzy controllers FLC-1 and FLC-2, and the performance was compared with a fixed
generator speed reference Such comparison was done because in a regular wind
generation system the speed reference for the generator is kept constant, and the user only
casually changes the set-point Table 10 shows the values for power at fixed speed and the
corresponding variation of power for FLC-1 operation and FLC-2 operation. The table
has the values in per-unit because it is better to compare and visualize the effective
efficiency in normalized numbers. The table summarizes the efficiency improvement for
each operating point, and the average efficiency enhancement due to the utilization of
flizzv control (instead of using a fixed generator speed) is computed. The efficiency
improvements due to FLC-1 and FLC-2 are plotted in Fig. 105, showing that the
operation of FLC-1 gives higher power, except when the fixed generator speed is
optimum for that wind velocity, and the operation of FLC-2 always give some
improvement, but with decreasing efficiency as the wind velocity increases because it is
more effective at light load.
The picture shown in Fig. 106 shows the experimental set-up in the laboratory
with an overall view of the double PWM converter, computer and equipment used during
the experiment.
143
-------
Tnble 10 Power enhancement due tb fuzzy logic control
Wind
Velocity
(pu)
l 1
Power for
fixed to,*
(0.575 pu)
APo due
FLC-1
(pu)
APo due
FLC-2
(Pu)
Efficiency
Improvement
due FLC-1
Efficiency
improvement
due FLC-2
Total
efficiency
improvement
due ftjzzy
Average
efficiency
improvement
due FLC-1
Average
efficiency
imprdvement
due FLC-1
Total averag
efficiency
improYemer.
due fuzzy
(%)
(%)
control
operation (%)
operation (%)
operation (%)
control
operatior.(%)
0.5300
.0.0581.
0.0626
0.0343
107.67
28.45
136.12
0.5691
0.1250
0.0990
0.0466
79.25
20.81
100.06
0.6083
0.1740
0.0893
0.0390
51.36
14.82
66.18
0.6475
0.2383
0.0643
0.0321
27.01
10.63
37.64
0.6866
0.2941
0.0269
0.0250
9.17
7.78
16.95
0.7258
0.3302
0.0024
0.0193
0.73
5.81
6.54
35.525
8.616
44.141
0.7650
0.3604
0.0064
0.0183
1.77
5.00
6.77
0.fe04J
0.4330
0.0399
0.0151
9.21
3.20
12.41
0.W
0.4500
0.0890
0.0145
19.78
2.69
22.47
0 8825
0.5000.
0.1522
0.0132
30.44
2,02
32.46
0.9216
0.5300
0.2135
0.0124
40.28
1.67
41.95
0.9608
0.6250
0.3099
0.0041
49.58"
0.44
50.02
-------
6 80
Efficiency improvement
due to FLC-1
Efficiency improvement
due to FLC-2
WIND VELOCITY (pu)
Fig 105 Efficiency enhancement of wind energy system due to fuzzy logic control
145
-------
Fig, 106 Experimental set-up picture
146
-------
8. Conclusion and Implementation Aspects for a ZOO kW System
A complete fuzzy logic control based wind generation system was analyzed,
designed, and a proof of concept experimental setup was built. The performances were
studied extensively by simulation and experimental validation. There were three fuzzy
logic controllers in the system. Controller FLC-1 searches on-line the optimum generator
speed so that aerodynamic efficiency of the wind turbine is optimum. A second fuzzy
controller FLC-2 programs the machine flux by on-line search so as to optimize the
machine-converter system efficiency. A third fuzzy controller FLC-3 performs robust
speed control against turbine oscillatory torque and wind vortex, Advantages of fuzzy
control are that it is parameter insensitive, provides fast convergence, and accepts noisy
and inaccurate signals. The fuzzy algorithms are universal and can be applied retroactively
in any system. System performance, both in steady state and dynamic conditions, was
found to be excellent. The system is ready to be reproduced in a field installation,
contributing to the performance and efficiency improvement of a wind energy generation
system.
The system was designed to be reproduced and applied into a 200 kW system.
Concerning the hardware supply, the DSP's controllers are Texas Instruments
TMS320C30 based, which are already an industrial choice. The DSP and 10 boards are
packaged by Spectrum, well known as a DSP third party company. The hardware
designed for interfacing with Manning chip, speed sensor, currents and voltages sensors,
are standard logic based However, the availability of the Harming chip must be checked
with the European manufacturer. A printed circuit board was,designed for interfacing with
the Powerex modules and it can be used for a higher power inverter. The Powerex
Intellimod modules are very reliable and easily purchased in the U.S. market.
The double-PWM converter must be assembled carefully. The IGBT devices must
be rated for 1200 V in order to have a 500 V dc-link loop. The leakage inductance in the
dc-link bus should be minimal, this might require a flat bus-bar for connecting the
modules. If the leakage inductance is very small in the final construction, it is possible to
147
-------
discard the utilization of snubbers across the IGBT devices, but high frequency capacitors
in parallel with the dc-link bus (as close to the modules as possible) are still required.
Because there are power circuits connected to logic circuits, the overall lay-out
must avoid ground loops. Clean pulses for driving the IGBTs are vital for the system.
Therefore, the system must be immune to EMI interference, and all cables should be
shielded.
Some extra hardware for protection and fault interface need to be designed for the
200 kW system Such hardware can be either interfaced with the DSPLINK bus on each
DSP board or can be plugged in the ISA PC bus. Either way is feasible and easy to be
controlled by the resident C programs If the protection interface is built in the DSPLINK,
C-pointers can address the hardware ports. If the protection interface is built in the ISA
PC bus, the DSP dual memory can address the interface. It is strongly recommended to
use the ISA PC bus, because the host program can easily supervise the fault indication in
the background while communicating with and controlling both boards for shut-down, in
case of fault. Some output ports must also interface with the system in order to
electronically disable the inverters firing pulses (during start-up), command the circuit
breakers, turbine yaw angle, and the relay for by-passing the series dc-link resistance (for
charging the capacitor). The following protection/fault indications should be implemented
in the final system:
INPUT PORTS
OUTPUT PORTS
Low wind speed (below threshold)
High wind speed (above safe value)
Over-temperature
Inverter tripping indication
Disable inverter firing pulses
Command circuit breaker
By-pass relay for series dc-link resistance
Turbine yaw angle control
With such hardware for protection and fault indication, the complete control
coordination given in section 3.2 can be fully implemented. The software is modularized
and all the interface with the protection and fault indication is easily implemented. The
required vector control parameters can be changed by C preprocessor directives indicated
by #define. However, the PI gains for the control loops must be fine tuned as usually. The
scaling for the fuzzy controllers is explained in appendix B.
148
-------
REFEREN CES
149
-------
REFERENCES
[1] "Time for action; Wind energy in Europe," European Wind Energy Association,
Rome, Italy, Oct. 1991.
[2] H. Le-Huy, P. Viarouge and J. Dickinson, "Application of power electronics in
windmill generation systems," ENERGEX'82 International Energy Conference, Regina,
Canada, May 1982, pp. 1080-1088.
[3] M. E, Ralph, "Control of the variable speed generator on the Sandia 34-m vertical axis
wind turbine," Windpower' 89, San Francisco, CA, Sept 1989.
[4] T A Lipo, "Variable speed generator technology options for wind turbine generators,"
NASA Workshop, Cleveland, OH, May 1984.
[5] C. V. Nayar and J. H, Bundell, "Output power controller for a wind-driven induction
generator," IEEE Trans, on AES, vol. 23, May 1987, pp. 388-401.
[6] P. G. Casielles, J. G. Aleixandre, J. Sanz and J. Pascual, 'Design, installation and
performance analysis of a control system for a wind turbine driven self-excited induction
generator," ICEM'90, Cambridge, MA, Aug. 1990, pp. 988-993.
[7] D A Torrey, ^Variable-reluctance generators in wind-energy systems," IEEE Power
Electr Specialists Conf, Seattle, WA, June 1993, pp. 561-567.
[8] C. S. Brune, R. Spee and A K. Wallace, "Experimental evaluation of a variable-speed,
doubly-fed wind-power generation system," IEEE Trans. Ind Appl, May/June 1994, vol.
30, pp. 648-655.
[9] "IEEE recommended practices and requirements for harmonic control in electric
power systems," Oct. 1991, Project IEEE-519
[10] Gary L. Johnson, Wind Energy Systems, Prentice-Hall, Englewood Cliffs,NJ, 1985.
[11] G.C.D. Sousa, B K Bose and J.G. Cleland, "Fuzzy logic based on-line efficiency
optimization control of an indirect vector controlled induction motor drive,"
IEEE/IECON'93 Conf. Rec , pp. 1168-1174, 1993.
150
-------
[12] N R. Zargari and G. Joos, "Performance investigation of a current-controlled
voltage-regulated PWM rectifier in rotating and stationary frames," IEEE Trans, on Ind.
Electr., vol. 42. no, 4, pp. 396-401, August 1995:
[13] B. K. Bose, Power Electronics and AC Drives, Prentice-Hall, Englewood Cliffs, NJ,
1986.
[14]Y.F. Li and C.C, Lau, "Development of fuzzy algorithms for servo systems," IEEE
Control Systems Magazine, April 1989.
[151GCD. Sousa and B. K. Bose, "A fuzzy set theory based control of a phase
controlled converter dc machine drive," EEEE-IAS Annu. Meeting Conf. Rec., Oct. 1991,
pp. 854-861.
[16] "Sirnnon User's Guide", SSPA Systems, Goteborg, Sweden, 1992,
[17] G C D Sousa, B.K.Bose, J. Cleland, R J.Spiegel and P J Chappell, "Loss modeling of
converter induction machine system", IEEE-DECON Conf, Rec., pp, 114-120, 1992.
151
-------
APPENDICES
¦s
152
-------
Appendix A - Inverter interface board layout
rW
JJii
i;
It PH3 1 Bffl iffl
mBkmmkmmmm1 | I
£
f H Tf^fTf
nor
A
INTERFACE WITH
POWEKEX MODULE
MGS 07/25/1995
153
-------
Appendix B - Fumy scaling for p.u. (per-unit) operation
Fuzzy algorithms are universal because the fuzzy inference engine is performed
with fuzzy variables in a normalized universe of discourse (from -1 to +1) However, there
are fuzzy scaling gains for input fuzzification and output defuzzification that need to be
supplied, and such gains will vary depending on the application
Fortunately, such gains usually have a physical interpretation, and at least an initial
choice can be supplied by the user and,with a few iterations and observation of the
system's response, they can be fine tuned.
One good example is the choice of the fuzzy normalization gains for the fuzzy PI
control (FLC-3). It requires KE and KCE for scaling the error and change-in-error and
one output gain (KTE) for change-in-torque. The gain KE is the maximum allowed speed
step in the system and the gain KCE is the maximum allowed acceleration step in the
system (which are both known by an experienced user) Such natural requirements define
a maximum variation of torque in the system, which defines the output gain KTE Since
the fuzzy controller takes error and change-in-error signals and integrates to build the
torque signal, KE and KCE can also be associated to respective stable integral and
proportional gains of a regular PI controller. With the definition of an initial set of KE,
KTE and KCE the user must perform some step responses to fine tune the gains.
The fuzzy algorithm that searches the optimum speed (FLC-1) has a fuzzy
computation of the gain of power (KPO) and gain of speed (KWR) The fuzzy gain
computation takes a weighted average of six gains, because there are six fuzzy sets for the
speed input The user must provide six such gains, which can be done by knowing how
the turbine power is related to the generator speed for several wind velocities. Such
information can be provided by the turbine manufacturer or by numerical evaluation of
performance curves. Fig. 107 shows how the six fuzzy gains for power and speed can be
chosen with the turbine output power curves in terms of the generator speed and wind
velocity.
The fuzzy controller FLC-2 (flux optimization) uses linear equations for
calculation of the scaling gains The input and output gains are defined by (b 1) and (b 2)
154
-------
u
*
o
a.
SMAtUSM
SPeed
¦* KWRSMAL^Bir,
'•* ~ KWRmeo,
uwsmall
<—~ kwiu,
UM/B1G
« ~ KWR,IGSMAa
« "KWR,,^
Fig. 107 Choice of fuzzy scaling gains for FLC-1
¦ Pb = a*e* + b (b.l)
^ = €1*0,-02*1^ + 03 (b.2)
where the torque estimate Tia, is given by (b.3)
Tic* = Kt'id.%, (b.3)
In accordance to the final report of the previous project*, the coefficients are defined on the
basis of the machine no-load current (Ini), rated torque (T^), base speed (cy, rated machine
efficiency t]im, and rated power P0 as (b.4) - (b.8).
"Turner, M.W., V. E. McCormick, and J, G. Cleland. Efficiency Optimization Control of AC
j
Induction Motors: Initial Laboratory Results. EPA-600/R-96-008 (NTIS PB96-153424),
February 1996.
155
-------
cl = 0.156
'inl^
c2 = 0.44
v®b/
f , \
nl
lrtd y
c3 = 0.221m
0.09
PpHim
100
^im
100
1
©b
/ p ^olim ^
0.035
100
'Him
100 )
(b.4)
(b.5)
(b.6)
(b.7)
(b 8)
156
------- |