IRON AND STEEL MULTIMEDIA
ENVIRONMENTAL COST-EFFECTIVENESS
MODEL FEASIBILITY STUDY
PEDCo ENVIRONMENTAL
-------
IRON AND STEEL MULTIMEDIA
ENVIRONMENTAL COST-EFFECTIVENESS
MODEL FEASIBILITY STUDY
by
PEDCo Environmental, Inc.
Cincinnati, Ohio 45246
Contract No. 68-02-3173
Task No. 15
Project Officer
R. C. McCrillis
Metallurgical Processes Branch, MD-62
INDUSTRIAL ENVIRONMENTAL RESEARCH LABORATORY
U.S. ENVIRONMENTAL PROTECTION AGENCY .
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27711
July 1980
-------
DISCLAIMER
This report has been reviewed by the Industrial Environmental
Research Laboratory, U.S. Environmental Protection Agency, and
approved for publication. Approval does not signify that the
contents necessarily reflect the views and policies of the U.S.
Environmental Protection Agency, nor does mention of trade names
or commercial products constitute endorsement or recommendation
for use.
11
-------
CONTENTS
Page
Figures iv
Tables v
1. Management Summary 1
Overview of the needs analysis 1
Overview of current systems 2
Feasible alternatives 3
Summary of the analysis 4
2. Needs and Evaluation Criteria 6
3. Alternative Systems 10
Manual 10
Semiautomatic 10
Fully automatic 11
State-of-the-art automatic 11
4. Evaluation of Alternatives 13
5. Recommendations 25
Next steps 25
References 33
Appendix A-l
111
-------
FIGURES
Number Page
2-1 Expanded Optimization Model 7
4-1 Development Man-Hours by Alternative 16
5-1 Overview of System 27
5-2 Project Schedule 30
A-l Linear Cost Function A-9
A-2 Exponential Cost Function A-9
A-3 Single-Point Efficiency Cost Function A-9
A-4 Discrete Efficiency Values of a Cost Function A-9
A-5 Nonlinear Cost Function Within a Small Range of
Efficiency A-9
IV
-------
TABLES
Number Page
4-1 Advantages and Disadvantages of Each Alternative 14
4-2 Development Man-Hours by Development Stage and
Alternative 15
4-3 Development Cost by Development Stage and Alter-
natives 15
4-4 Annual Program Maintenance Cost by Year of Oper-
ation and Alternative 18
4-5 Annual Data Maintenance Cost by Alternative 18
4-6 Operating Hours Per Analysis by Problem Size and
Alternative 21
4-7 Operating Cost Per Analysis by Problem Size and
Alternative 22
4-8 Total Annual Costs by Number of Analyses and
Alternative 22
4-9 Cost Per Analysis in the First Three Years of
Operation 24
5-1 User Needs Analysis Ranking Matrix 26
v
-------
SECTION 1
MANAGEMENT SUMMARY
OVERVIEW OF THE NEEDS ANALYSIS
The steel industry is a major source of air and water pollu-
tion and also generates large quantities of solid waste. The
Environmental Protection Agency (EPA) has devoted considerable
resources to the study of this industry over the past 10 years.
These studies have been directed to all aspects of the problem
including basic research, applied research, standards setting,
economic impact analysis, and demonstration plants.
In general, the pollution problem in a given media has been
considered on an independent basis, and even a single media has
been basically considered on a single process basis. Recently,
however, the interrelationships between media and between proc-
esses have become of more interest. This approach is reflected
in EPA policies such as the bubble concept and the unified permit
procedure. A vast amount of data have been accumulated that
allow advanced analysis techniques to be applied to assist the
agency in policy analysis, research prioritization, and data
organization.
The Industrial Environmental Research Laboratory (IERL),
with support from the Office of Air Quality Planning and Standards
-------
(OAQPS), Effluent Guidelines Division (EGD), Office of Planning
and Evaluation (OPE), and others, has authorized this feasibility
study to determine if an advanced computerized model is the most
cost-effective means to continue strategy analysis on a multimedia
basis.
In September 1978 in response to a request from OAQPS, IERL
began development of a pollution control cost-effectiveness model
for coking facilities. The model was limited in scope to air
pollutants from coke plant operations. Subsequent development
and use of the model revealed a widespread interest and need for
this type of comprehensive analytical approach in evaluating
tradeoffs. The proposed multimedia cost-effectiveness model will
address all pollutants from all processes in the entire iron and
steel industry. The magnitude of such an analysis suggests the
use of computer techniques.
OVERVIEW OF CURRENT MODELS
\
No existing model addresses the problem in its entirety.
The library of programs, however, for the EPA UNIVAC computer at
the National Computation Center (NCC) in Research Triangle Park,
North Carolina, has a variety of mathematical optimization
(linear programming) packages that can be used for this project.
It is proposed that the new model incorporate the mixed integer
linear programming package for the UNIVAC machine. Although this
package provides the core of the analysis, a supporting structure
for data handling and input/output is also necessary in order to
broaden the usefulness of the model.
-------
The initial cost-effectiveness model for coking facilities
provides a foundation for the analysis and establishes useful
concepts, but the scope of the proposed model far exceeds the
capability for expansion of the original model in its present
form. The scope of the proposed model will include, for example,
25 air pollutants (compared with 4 in the original model), 40
water pollutants (none in the original model), and 20 solid waste
parameters (none in the original model). Additionally, at least
80 sources will be considered compared with 16 in the original
model.
FEASIBLE ALTERNATIVES
The four alternatives examined in this study range from a
manual system to a state-of-the-art system. Note that "system"
will from hereon refer to the complete scope of data storage,
retrieval analysis and reporting. The study shows that greater
use of computerization decreases the long-term cost of the system.
The manual system involves a high cost because of the vast amount
of data and the almost incalculable mathematical operations that
must be performed on the data. None of the alternative systems
requires additional hardware; in fact, the proposed alternative
takes advantage of existing software. The four systems investi-
gated are:
(A) Completely manual
(B) Computerized data storage--manual analysis
(C) Computerized data storage—trial and error
analysis
(D) Computerized data storage—optimization analysis
-------
Each of these systems is self explanatory except for the
distinction between trial and error analysis and optimization
analysis. The original coke model essentially used a trial and
error analysis. In this approach, all possible scenarios are
first computed and the one best meeting the input criterion is
selected. This approach is feasible for small problems, but
computing time and cost rapidly increase as the size of the
problem approaches that contemplated. The optimization analysis,
however, uses well proven mathematical algorithms to rapidly
focus on the solutions that are optimum according to the input
criterion.
SUMMARY OF THE ANALYSIS
The analysis concludes that a fully computerized system
(Alternative D) results in the lowest cost. The fully comput-
erized system is the only alternative capable of considering all
the sources and control alternatives. In the first year of
operation, the estimated annual cost of Alternative D is $163,000,
compared with $183,000 for Alternative C. After 3 years of
operation, the total cumulative annual cost of Alternative D
(assuming 20 analyses/year) is $191,000, compared with $293,000
for Alternative C. Analysis of Alternatives A and B shows that
they are infeasible.
Alternative D is recommended as the most feasible system for
a multimedia cost-effectiveness model for the iron and steel
industry. This system will fully meet EPA's need to coordinate
the data from all interested divisions by using the latest
-------
techniques of analysis to examine policy alternatives, control
scenarios, and prioritize data development needs.
This system can be used by anyone having minimal knowledge
of iron and steel processes and their controls, and access to a
terminal which can be used on the NCC computer. This system does
not require any changes in the present operation of the computer
center itself, or any additional staffing.
-------
SECTION 2
NEEDS AND EVALUATION CRITERIA
The cost-effectiveness model for pollution control at coking
facilities (as presented in Report No. EPA-600/2-79-185) deter-
mine the cost-effectiveness of air pollution control at coking
facilities by identifying an optimum configuration of control
devices. The model considered each pollutant separately; it was
confined to one medium and to coking facilities only.
The proposed model will enlarge the earlier one by con-
sidering multimedia emission data (air, water, and solid wastes).
It will also consider facility data for iron and steel production
in addition to coke production. Figure 2-1 shows a configuration
of the scope of the model. The model will enable the user to
determine by defining the optimum configuration of control
devices in order to reduce pollutants in all media from all
emission sources in the iron and steel industry at a minimum
cost. The following are the objectives of the cost-effectiveness
model:
To make cost comparisons between potential regulation con-
trol strategies applied to iron and steel facilities
To determine the appropriate combination of control to
achieve a specified emission reduction at the minimum cost
To determine the appropriate combination of controls to
achieve the lowest possible emissions at a specified cost
To calculate cross-media impacts and energy impacts
6
-------
CONTROL
STRATEGIES
MULTIMEDIA
EMISSION
DATA
FACILITIES
DATA
COST
DATA
PROCESS\ SOLID
FACTORS] WASTE
ENERGY
CONSUMPTION
TOTAL
COST
SOLID
WASTE
QUANTITIES
WASTEWATER
EFFLUENTS
AIR
EMISSIONS
Figure 2-1. Expanded optimization model.
-------
Alternatively, the model should help determine the optimum con-
figuration of control devices to achieve the maximum possible
reductions in total pollutants for a given capital or annual
tot 1 cost. The model should be capable of considering individ-
ual plants or regions or the industry as a whole. The model
should be modularized so that each component of the industry can
be considered separately or integrated with the rest. The
following are the advantage of a computerized model approach:
Linkage between media impacts and energy impact
Ability to quickly evaluate alternative strategies
Flexibility for growth and updating
Analysis of optimum solutions
Ability to perform sensitivity analysis
Ability to identify R&D needs and priorities
Several alternative methods are available for solving the
optimization problem. All the criteria must be fulfilled in
order to have a useful system. The criteria listed below were
used to evaluate alternative methods. The assignment of weight-
ing factors to each individual criterion is not practical for
this system.
Technical expertise required by the developers and the
users. The amount of expertise needed by each depends upon
the sophistication level of the system. A more sophisti-
cated system requires more expertise in the development
phase, but less expertise of the user. Because the system
is intended for use by a broad cross-section of program
offices, it must be designed so that potential users can
readily familiarize themselves with its use and application.
System use, interpretation of results, and management of
problems. Less sophisticated systems will give clear
results, but they may not be easier to use and are likely to
be limited in scope. Program offices should be able to
eventually use the system directly from local terminals.
-------
Length of time and cost required to produce an economical
solution to the problem. This is perhaps the most important
criteria in selecting any system. Even though a system may
cost very little to develop, it is not useful if it cannot
produce solutions within a given amount of time. Conversely,
a system that costs slightly more to develop may be able to
produce one or several solutions in a very short time.
Additionally the system should allow for unanticipated
problems. Finally, the system should be easy to update with
new or modified data.
-------
SECTION 3
ALTERNATIVE SYSTEMS
Several alternative systems are available for producing the
multimedia cost-effectiveness model for the iron and steel in-
dustry. These alternatives include manual, semiautomatic, fully
automatic, and state-of-the-art automatic systems. All of these
systems are required to analyze the same site-specific parameters
for iron and steel plants to determine the cost-effectiveness of
air pollution controls by identifying the optimum configuration
of control devices. The general concepts and features of each
system are described below.
MANUAL
This system consists of all necessary plant specific infor-
mation organized in a very stringent and controlled manual
library. Data retrieval, compilation, and problem analysis must
all be performed manually in this system. The major feature of
this alternative is that only limited amounts of information can
be retrieved and processed.
SEMIAUTOMATIC
In this system the detailed site-specific data will be
organized in a computerized library to allow automatic data
10
-------
retrieval and compilation for a manual analysis task. The major
features of this alternative are a sophisticated computerized
library and automatic compilation of the retrieved data. The
problem analysis must be performed as it was in the manual
system.
FULLY AUTOMATIC
In a fully automatic system the detailed data will be
organized in a computerized library that allows automatic re-
trieval, compilation, and analysis of selected data. The data
storage features of this alternative are the same as those of a
semiautomatic system, but also include the capability of auto-
matic analysis through the use of the same iterative approach
used by the cost-effectiveness model for pollution control at
coke facilities (as presented in Report No. EPA-600/2-79-185).
The iterative approach is to examine every possible solution
through completion and select the one which best meets the input
criteria.
STATE-OF-THE-ART AUTOMATIC
The description and features of this alternative are iden-
tical to those of the fully automatic system with one major
difference: a mixed integer programming (MIP) package (see
Appendix A) will replace the iterative technique. Unlike the
iterative approach, the MIP will not look at every possible solu-
tion, but, based on input criteria, will quickly find its way to
the optimum solution using mathematical methods.
11
-------
This system will also incorporate sophisticated data manage-
ment techniques, e.g., the System 2000 for the Univac 1100. Be-
cause the iron and steel industry is dynamic; the data must be
continually updated. Data management is therefore an important
element of the system that enables rapid and economical updating
and data retrieval. The essential features of data management
will be the use of key fields, hierarchical relationships, and
data security. Data handling is described schematically in
Section 5.
12
-------
SECTION 4
EVALUATION OF ALTERNATIVES
This section evaluates each alternative described in Section
3 as measured against the criteria set forth in Section 2. Table
4-1 summarizes the advantages and disadvantages of each alterna-
tive.
Development costs for all data processing projects must in-
clude manpower and computer usage during development. For
clarity, the development stage of a project can be broken down
into five major categories:
(1) System design
(2) Coding and compilation
(3) Testing and debugging
(4) Documentation
(5) System testing, installation, and training
Table 4-2 presents a man-hour breakdown for development of
the four alternatives. The manual and semiautomatic methods
require long training sessions to educate the user about mathe-
matical analysis procedures needed to derive a solution. Con-
versely, the fully automatic methods require the majority of time
during the initial design phase to insure system completeness,
i.e., clear results, quick response, and uncomplicated proce-
dures. Figure 4-1 is a graphical representation of the total
development time in man-hours. Table 4-3 translates these
13
-------
TABLE 4-1. ADVANTAGES AND DISADVANTAGES OF EACH ALTERNATIVE
Alternative
Advantages
Disadvantages
Manual
Semiautomatic
Fully automatic
State-of-the-art-
automatic
Quick retrieval of small
data sets
Low development cost
Quick retrieval of all
data
Fewer personnel required
for maintenance
Data can be accessed from
various locations via
communications network
Data can be compiled in
preparation for analysis
Same advantages as semi-
automatic system
Plus
Problems can be solved faster
than by manual analysis
More reliable than manual
analysis
Less technical expertise
required for analysis
Larger range of problems
can be solved
Analyses can be performed
of unanticipated problems
Same advantages as fully
automatic system
Plus
Total range of problems
can be solved by using
a fast and efficient
"screening" type of
analysis
Usable by a broad variety
of personnel from vari-
our program offices
Faster than all other
alternatives
More constraints can be
applied to problems
Low operating costs
No additional staffing
required
Easy to update and expand
Large amount of manpower re-
quired for maintenance and
operation
Retrieval of large data sets
is time consuming
Data can be accessed in only
one location
High technical expertise re-
quired for problem analysis
Analysis time Increases expo-
nentially with size of
problem
Limited range of problems
can be solved
Difficult to update or expand
High operating cost
High technical expertise
required for analysis
Analysis time may be
extensive
Higher development cost
than manual system
Limited range of problems
can be solved
Higher development cost than
manual or semiautomatic
Range of problems still
limited even though larger
than that for manual
analysis
Analysis time longer than
that for state-of-the-art
automatic due to iterative
"plodding" type of analysis
High development cost
More technical expertise re-
quired in developmental
stage
14
-------
TABLE 4-2. DEVELOPMENT MAN-HOURS BY DEVELOPMENT
STAGE AND ALTERNATIVE
Development stage
System design
Coding and compilation
Testing and debugging
Documentation
Installation and
training
Total
Manual
250
100
1000
1350
Semi-
automatic
500
300
250
200
1100
2350
Fully
automatic
1500
600
800
400
200
2.500
State-of-the-art
automatic
1500
1000
800
400
300
4000
TABLE 4-3. DEVELOPMENT COST BY DEVELOPMENT STAGE AND ALTERNATIVE'
(dollars)
Development stage
System design
Coding and compilation
Testing and debugging
Documentation
Installation and
training
Total
Manual
8,250
3,300
33,000
44,550
Semi-
automatic
16,500
11,175
9,313
6,600
36,725
80,313
Fully
automatic
49,500
22,350
28,100
13,200
7,450
120,600
State-of-the-art
automatic
49,500
37,250
29,800
13,200
11,175
140,925
Adapted from Guidelines for Preparing and Reviewing Feasibility Studies.
EPA 68-01-3836. April 1977. Man-hour costs have been inflated from
$25/h to $33/h.
15
-------
4000
3000
2000
1000
MANUAL SEMI- FULLY- STATE-OF-
AUTOMATIC AUTOMATIC THE-ART
AUTOMATIC
Figure 4-1. Development man-hours by alternative.
16
-------
hours and associated computer usage into dollars by development
stage for each of the four alternative systems. Labor costs were
based on an estimated 1980 - 1981 rate of $33/h.
Annual costs are another major aspect of an evaluation of a
data processing system. Annual costs may be broken down into
three categories:
(1) Program maintenance
(2) Data maintenance
(3) Operating cost per analysis
Program maintenance cost in the first year is defined to be
2
10 to 12 percent of the development cost. program maintenance
cost in the second year is 6 to 10 percent of development cost,
and in each successive year, it is 5 percent. Table 4-4 presents
annual program maintenance. (Note that a manual system will have
no program maintenance costs.)
Data maintenance is a very significant part of the cost-
effectiveness model. Plants may eventually change equipment
configurations, increase throughput, or add controls. In addi-
tion, new emission and pollutant data are continually being
generated, new control systems are continually under development,
and cost data are changing because of new technologies and
inflation. This information should be entered into the data
library of all of the alternatives to maintain data integrity.
Table 4-5 summarizes the cost of data maintenance and data stor-
age. Data storage cost is based on an estimated total of 2,800,000
characters, 7000 characters per track, and a rate of $0.025 per
track per week. (Note that no computer costs are involved in the
manual system.)
17
-------
TABLE 4-4. ANNUAL PROGRAM MAINTENANCE COST BY YEAR OF
OPERATION AND ALTERNATIVE
Year of
operation
1
2
3
4
Semi-
automatic
8,800
6,400
4,000
4,000
Fully
automatic
13,300
9,600
6,000
6,000
State-of-the-art
automatic
15,500
11,300
7,000
7,000
TABLE 4-5. ANNUAL COST OF DATA MAINTENANCE AND DATA STORAGE BY ALTERNATIVE
Cost
category
Manpower
Computer
Storage3
Total
Manual
6,600
b
6,600
Semi
automatic
2,400
2,000
520
4,920
Fully
automatic
2,400
2,000
520
4,920
State-of-the-art
automatic
2,400
2,000
520
4,920
Includes program storage.
There is a cost for manual storage, including such items as file cabinets
and floor space, but no attempt has been made to estimate these costs.
18
-------
Up to this point all alternatives have been treated as
equally capable of performing the same task, and it would appear
that the manual system has a definite advantage as far as devel-
opment and maintenance costs are concerned. Any decision to
select a manual alternative based on information presented thus
far could prove to be extremely costly in the long run. Signifi-
cant differences appear when operating costs are examined. To
understand these differences, a review of the complexity of the
problem is in order. For a problem involving a total of 85
pollutants, 70 emission sources, and an average of 4 control
alternatives per source, there are approximately 320 decision
variables that have up to 350 constraints (see Appendix). The
objective is to find the optimum configuration of control devices
that will keep multimedia pollutant emissions below specified
levels at the lowest cost. The manual and semi-automatic alter-
natives are very limited in their capabilities and could not
solve this problem in a reasonable amount of time because both
require manual analyses. Furthermore, few qualified personnel
would be able to perform the sophisticated mathematical proce-
dures required. Therefore, when directly comparing manual and
automatic methods all succeeding tables and figures should be
used cautiously for all but a very limited range of problems.
The EPA's UNIVAC 1100 machine, which is located at the
technical center in Research Triangle Park, North Carolina, is
capable of solving this problem. The EPA has available the
UNIVAC FMPS (Functional Mathematical Programming System) package,
which can readily solve problems of the size described earlier.
19
-------
Preliminary estimates have been made of run time for a plant
model (350 constraints and 320 variables with 160 integer vari-
ables) . The run time will be about 2 to 3 minutes for problems
of normal complexity. The run time may exceed 10 minutes for
complicated problems. The FMPS has provisions to terminate
problems if no solutions are found within a reasonable time.
Table 4-6 illustrates that although the manual and semiauto-
matic alternatives provide solutions to very small problems, they
become unmanageable as the size of the problem increases. Prob-
lems involving only 4 pollutants, 10 sources, and 3 control
alternatives would require a long time for manual analysis. It
is estimated that any manual analysis effort would increase
exponentially as the size of the problem increases. Since the
exact number of analyses that may be performed during a given
time period is not known, Table 4-7 lists the cost per analysis.
Computer costs were estimated using a rate of $215 per computer
hour. This is considered to be a long-term representative value
for the NCC Univac Computer for a complex system. If n analyses
were to be performed during a given year, the direct annual
operating cost would be equal to:
program maintenance cost from Table 4-4
+ data maintenance cost from Table 4-5
+ (n times operating cost per analysis from Table 4-7).
Table 4-8 presents the direct annual operating costs by
number of analyses and alternatives for a large problem during
the first year of operation. Any extra expense incurred in the
development and maintenance of the state-of-the-art system versus
20
-------
TABLE 4-6. OPERATING HOURS PER ANALYSIS BY PROBLEM SIZE AND ALTERNATIVE
Operation task
Small problem
Retrieval
Compilation
Analysis
Total
Medium problem
Retrieval
Compilation
Analysis
Total
Large problem
Retrieval
Compilation
Analysis
Total
Manual
10
10
500
520
15
30a
2000a
2045
20
50
8000a
8070
Semi-
automatic
0.01
0.01
500.00
500.02
0.01
0.01
2000a
2000.
0.01
0.01
8000a
8000.
Fully
automatic
0.01
0.01
2.50
2.52
0.01
0.01
5.00
5.02
0.01
0.01
10.00
10.02
State-of-the-art
automatic
0.01
0.01
0.05
0.07
0.01
0.01
0.01
0.12
0.01
0.01
0.15
0.17
Analysis time is very difficult to estimate for problems of this size. The
numbers shown are based on engineering judgment.
21
-------
TABLE 4-7. OPERATING COST PER ANALYSIS BY
PROBLEM SIZE AND ALTERNATIVE
Problem
size
Small
Medium
Large
Manual
17,200
67,500a
226,300a
Semi -
automatic
16,500
66,000a
264,000a
Fully
automatic
550
1,100
2,200
State-of-the-art
automatic
40
65
90
Operating costs are tentative for problems of this size.
TABLE 4-8. DIRECT ANNUAL OPERATING COSTS BY NUMBER .OF
ANALYSES AND ALTERNATIVE3
Number of
analyses/year
10
20
50
100
Manual
2,669,600
5,332,600
13,306,200
26,636,600
Semi-
automatic
2,653,700
5,293,700
13,213,700
26,413,700
Fully
automatic
40,200
62,200
128,200
238,200
State-of-the-art
automatic
21,300
22,200
24,900
29,400
These costs are for large problems only in the first year of operation.
22
-------
a manual analysis system will be recovered in the first year
after, at most, 6 small size analyses. In all successive years
of operation a manual system will always be more expensive.
The manual and semiautomatic systems are not feasible alter-
natives because of the limited range of solvable problems, the
time restrictions of obtaining a solution to these problems, and
the extremely high operating costs. The fully automatic systems
remain for consideration.
As an example, a very moderate usage of the remaining
alternatives was considered over a period of years, e.g., 20
analyses per year. Large problems were assumed in order to
obtain the most benefit from each alternative in terms of saved
man-hours. Table 4-9 shows the total operating costs for the
first 3 years of operation for each of the alternatives. These
estimates include development costs as well as operating and
maintenance costs. Note that in each successive year the state-
of-the-art alternative becomes increasingly less expensive. The
differences would be even more pronounced as the number of
analyses increases. Therefore, the state-of-the-art-automatic
system is the alternative that should be used when implementing
the multimedia cost-effectiveness model for the iron and steel
industry.
23
-------
TABLE 4-9. COST PER ANALYSIS IN THE FIRST THREE YEARS OF OPERATION
to
*>.
Year 1
Development
Program maintenance
Data maintenance
Analysis of 20 large problems
Total costs for first year
Cost per analysis8
Year 2
Program maintenance
Data maintenance
Analysis of 20 large problems
Total costs for second year
Total costs for first and second
years
Cost per analysis9
Year 3
Program maintenance
Data maintenance
Analysis of 20 large problems
Total costs for third year
Total costs for first, second, and
years
Cost per analysis3
Manual
44,550
0
6,600
5.326.000
5,377.150
268.857
0
6.600
5.326.000
5.332.600
10,709.750
267,744
0
6.600
5.326.000
5.332.600
16,042.350
267,372
Semi-
automatic
80,313
8,800
4,920
5.280.000
5.374,033
268,702
6,400
4,920
5,280,000
5,291,320
10.665,353
266,634
4,000
4,920
5,280.000
5,288,920
15,954,273
265,904
Fully
automatic
120,600
13,300
4,920
44,000
182,820
9,141
9,600
4,920
44 ,000
56,720
239,540
5,988
6,000
4,920
44,000
53,120
292,660
4,877
State-of-the-art
automatic
140,925
15,500
4.920
1.800
163.155
8,157
11,300
4,920
1,800
16,220
179,365
4,484
7,000
4,920
1,800
11,920
191.285
3,188
aThese figures apply only for a usage rate of 20 analyses per year.
-------
SECTION 5
RECOMMENDATIONS
Table 5-1 summarizes the ranking of each system based on the
discussion in Section 4 and relative to user needs criteria.
Although the ratings are subjective, they indicate the relative
strengths and weakenesses of each alternative. Based on these
ratings it clearly appears that the state-of-the-art automatic
alternative be used to implement the proposed multimedia cost-
effectiveness model for the iron and steel industry. This alter-
native was selected over the others for several reasons. The
high development cost of this alternative is quickly offset by
providing solutions to the largest problems, in the shortest
amount of time, for the lowest total cost. In fact, as a system
of this type is used more, the unit cost per use becomes lower
because initial development costs are spread over a larger base.
Costs for the first year, including development, operation, and
maintenance, are approximately $163,000. The operation and
maintenance costs for the second year are approximately $16,000.
In all succeeding years this cost will be about $12,000.
NEXT STEPS
Figure 5-1 presents an overview of the system. The figure
indicates that the system is comprised of two elements: data and
25
-------
TABLE 5-1. USER NEEDS ANALYSIS RANKING MATRIX'
Need
Short retrieval time of large data sets
Data accessibility
Short analysis time required
Large range of problem solving
Usable by all program offices
Low technical expertise required
for development
Low technical expertise required for
analysis
Small manpower requirements for main-
tenance and operation
Ease of updating
Ease of expansion
Low development cost
Low operating cost
Rating for alternative
A
1
2
1
2
3
8
1
2
3
1
8
1
B
1
2
1
5
5
7
1
3
4
4
4
1
C
10
9
6
7
8
3
7
10
9
6
2
5
D
10
9
10
10
9
1
8
10
9
9
1
9
aRanking values based on information listed in Table 4-1 and using engineering
judgment.
1 = Low rating or less desirable relative to need.
10 = High rating or more desirable relative to need.
26
-------
U.
o
13
O
O.
EPA
INDUSTRY
VENDORS
RESEARCH
OAQPS
EGD
OPE
ECONOMIC
CONDITIONS
NEW CONTROL SYSTEMS
NEW EQUIPMENT
COSTING METHODOLOGY
CHANGES IN EQUIPMENT SIZE
\
COSTING
PROGRAM
MODULE
CONTROL
COST
FUNCTION
^ FILE _
NEW EMISSION DATA
REVISED EMISSION FACTORS
INDUSTRY
CHANGES
NEW PLANTS
PLANT CLOSURE
NEW PROCESSES
EMISSION
DATA FILE
AIR, WATER
SOLID WASTE
PRODUCTION
FACILITIES
DATA
. FILE •
INPUT:
o
i
SE
£
o.
o
SCENARIO DEFINITION
RESTRAINTS
DATA SCOPE
MASTER PROGRAM
.CONTROL _ _
"INPUT/OUTPUT "CONTROL
FEEDBACK
OPTIMUM RESULTS
ENVIRONMENTAL
IMPACTS
ENERGY IMPACTS
COST IMPACTS
STATUS
REPORTS
DATA
RETRIEVAL
Figure 5-1.. Overview of the recommended system.
27
-------
analysis. This dichotomy of structure is equally advantageous
during system design and implementation and during actual use.
During design and implementation, it enables separate work teams
to proceed nearly simultaneously, thereby shortening overall
implementation time.
This system will be administered by IERL in the EPA Tech-
nical Center at Research Triangle Park, North Carolina. A
coordinating committee consisting of representatives from program
offices will provide guidance and input during the development of
the system to insure that it is compatible with their programs
and that the system capabilities will fulfill their needs. This
approach has proven successful in the development of the coke
oven model previously mentioned.
System outputs will consist of concise summaries of the
impacts associated with either an optimum solution or a scenario
defined by the user. Environmental impacts such as total waste-
water generated, pollutant loadings in that wastewater, total
energy required, and total tons of solid waste will be calculated
from the detailed data contained in the data base. The IERL will
screen and approve all original and updated values entered into
the data base for validity and accuracy prior to entry. Printed
output can be converted to remote terminals; however, the timing
requirements at this time are not considered critical enough to
require remote terminal operation. File maintenance data,
however, will be programmed to permit either conventional batch
maintenance or terminal maintenance. An EPA coordinator assigned
28
-------
by the coordinating committee during system development will
coordinate all file maintenance.
Original input data will be derived primarily from existing
and ongoing government studies and reports. A major ancillary
engineering effort will be required to organize and evaluate the
data. The coordinating committee will monitor this effort.
Figure 5-2 shows an estimated project schedule. Actual
model development is estimated to require 6 months. This esti-
mate assumes that three full-time program analysts with miscel-
laneous support personnel are assigned to system design, file
structure, and input/output programming. This activity also
includes interfacing requirements between the control program and
existing software, i.e., FMPS.
The first step is file design. Once data file formats and
extents are determined, separate technical teams will be able to
collect, evaluate, and organize input data for the data base.
Concurrently, another technical team will be developing
control cost algorithms for the costing program module. This
module is a key element of the system because it will enable
automated cost calculations to keep pace with changing conditions
in the industry and the economy and will also calculate costs for
new technologies as they are developed. This capability of the
system will overcome the shortcomings experienced in many past
efforts wherein the cost data were out-of-date or inapplicable
shortly after they were developed.
29
-------
Task description
Model development
System specifications
Detailed incorporation
of FMPS
File structure and maintenance
Data management
Calculations of cost and
emission coefficients
Input/output structure
System documentation
Data development
Scope definition
Emission data
Cost data
Control technique
Facilities data
Process Interrelationships
Stack data
Project reporting
Draft report
Final report
Users manauals
Orientation package
Program management
Months
1
2
3
4
5
6
7
8
9
10
11
12
13
Figure 5-2. Project schedule.
-------
To achieve meaningful output from the optimization model,
the data must be as accurate and up-to-date as possible. This
requires ease of data input and modification. The data files
consequently must be structured on a detailed level that permits
ease of access to an individual data element referenced by easily
recognized coding structures. A rigorous data coding structure
will have two major advantages: (1) new and revised data can be
easily put into the data base, and (2) data retrieval for routine
purposes aside from the optimization capabilities of the system
will be facilitated.
Given well-documented and comprehensive data, the system
enables the user to evaluate alternative scenarios; to determine
the economic, environmental, and energy impact of possible con-
trols; and to assure that controls are applied in the most cost-
effective manner possible.
Determining the optimum configuration of control alterna-
tives is essentially a linear programming problem with the addi-
tional complication of requiring integer values for some of the
variables and continuous values for others. For example,
whether a particular control option should be included or not can
be represented by a variable that takes on values of 0 or 1 only.
If it has a value of 0, the option is not to be included; if it
has a value of 1, the option is to be included. Some other
variables may take on continuous values. For example, if a
control alternative can be introduced at any specified level, the
value for that variable should be capable of taking on any value
31
-------
between 0 and 1. Problems requiring both integer and continuous
values for the variables are called Mixed Integer Programming
(MIP) problems. The multimedia optimization problem is essen-
tially solved using MIP.
The master program controls all system functions and pri-
marily translates user scenario definitions from English to
mathematical terms. This is an important feature that permits
all users to communicate in their normal language without having
to be familiar with computer language. The control program
essentially sets up each problem in coded form for processing by
FMPS, the optimization program. The system can also function as
a data base retrieval and processing system without optimization,
Finally, the control program translates the results back into
English and produces clear, concise, easy-to-read reports.
32
-------
REFERENCES
1. Kemner, F. Cost Effectiveness Model for Pollution Control
at Coking Facilities. EPA-600/2-79-185. August 1979.
2. Guidelines for Preparing and Reviewing Feasibility Studies
EPA 68-01-3836. April 1977.
3. Personal Communication between R.M. Livingston and G. Ball,
EPA NCC. June 5, 1980
33
-------
APPENDIX A
MIXED INTEGER PROGRAMMING (MIP) PROBLEM
In a linear programming problem values have to be found for
certain decision variables so that a given objective function of
the variables is maximized (or minimized) and the variables and
some functions of the variables are subjected to predefined
restrictions (constraints). Usually when the constraints and the
objective function are linear, the problem is solved through
linear programming. Linear programming has found wide applica-
tions in business and industry. If some of the variables are
restricted to be integers, the problem is solved by MIP. (See
Reference A-l for a comprehensive discussion.)
An advantage possessed by MIP is that interdependencies and
unusual restrictions can be easily formulated. For example, if
Alternative A cannot be used simultaneously with Alternative B,
a constraint of the form
A + B <_ 1
will indicate an interdependency provided A and B are restricted
to be integers taking on values of either 0 or 1. In a more
complicated case, if dry quenching (Q) occurs pushing control (P)
will occur also; however, this situation does not occur in re-
verse. This situation can be represented as
P - Q >_ 0
A-l
-------
Even more complex relationships can be easily modeled into MIP's
because of the integer restrictions on these variables. The
flexibility and capability make MIP a very powerful modeling tool
in many diverse situations.
Solving MIP is more difficult than solving a linear program,
mainly because it involves developing a number of linear pro-
grams, keeping them in store, and solving each until the best
solution is found. Many efficient procedures have been developed
that are commercially available. The EPA's computer system has
the FMPS package that is capable of solving such MIP's. The time
required to solve an MIP increases rapidly with the size of the
problem, especially with the number of variables required to be
integers. This aspect makes it imperative to exercise great care
in modeling, so that all essential relationships are captured
with the least number of extra constraints and integer variables.
In the initial stages the costs of control devices are
assumed to be linear (if they can be introduced for different
efficiencies). The costs of these devices can be calculated
using the parameters A and B as described in the earlier report
C = A XB
A-2
where X is the annual production quantity. It will be further
assumed that quantities of a pollutant produced by each source
can be added together to give the total emission and that one
pollutant does not affect another.
A-2
-------
To describe the mathematical formulation, the following
symbols are defined:
i - pollutant ( i = 1,2,...,I)
j - source (j = 1,2,...,J)
k - control alternative (k = 1,2,...,K)
1 - product (1 = 1,2,...,L)
Each product (in an integrated plant) will have some sources
specific to that product.
Sources i...j correspond to product 1 (e.g., coke)
j ...j correspond to product 2 (e.g., iron)
j ,...JT correspond to product L (e.g., iron)
L~l L
e.. - efficiency of control device k when used on source
1-' j for pollutant i while producing product 1
u.., - quantity of pollutant i produced by emission
1-) source j per ton of product 1 produced
X - quantity of product 1 produced per year
E. ., .. - quantity of pollutant i produced by source j with
Jijkl
control device k while producing product 1
E. ., .. = (1.0 - e. ., n ) -u. .. -X.
ijkl
c., - cost of including control option k on source j
JK
x., = 1 if control option k is included on source j
-1 = 0 otherwise
y. - fraction of control option k included on source j
3*
P. - total specified quantity of pollutant i
w. - undesirability weight to be associated with pol-
1 lutant i
3 - available budget for introducing pollution control
devices
A-3
-------
Two models will be developed: one for minimizing cost to
achieve desired levels of pollutants and the other for maximizing
reduction in pollutants for a given cost.
MODEL I
The objective in this model is to find the optimum configu-
ration of control devices that will keep multimedia pollutant
emissions below specified levels at the lowest cost.
Minimization of z = I. I. c.. -x., + Z £ c. Y..
j k Dk Dk j k Dk Dk
subject to:
-------
MODEL II
The objective of this model is to find the optimum con-
figuration of control alternatives to produce the minimum level
of pollutants for a given total cost.
This formulation is more difficult because it requires a
weighting scheme for combining the different pollutant quanti-
ties. It may be necessary to define the undesirability of each
pollutant (e.g., w.) so that the total undesirability of all the
pollutants produced can be described, which could then be minimized.
Minimization of z = I, I, I. (E. ..•x.1 ) *w.
i j k i:)kl Dk 1
(total undesirability should be minimum) subject to:
(i) I I cik'X-ik + £ * cik*jik- 3
j k Dk DK j k DK DK
(total cost of control should be less than budget)
(ii) E x < 1 (j = 1,2,...J)
k DK
(there can be only one control option for each source)
(iii) Other constraints for specific interdependencies)
(iv) 0 < x < 1 (j = 1,2,...J, k = 1,2,...K)
~ DK
(integrity constraints)
Although the number of variables is the same as in Model I,
the total number of constraints will be considerably less in this
model (1-1 or 84 fewer constraints). This model, however, re-
quires the development of a weighting scheme to determine the
undesirability of pollutants.
A-5
-------
The problem of developing a weighting scheme may be avoided
by resorting to goal programming, in which desired goals are
specified for pollutant levels and priorities are assigned for
deviations from those goals. It is not yet feasible, however, to
solve goal programming by using integer variables (see Reference
A-2 for a brief discussion).
Both models described above are for one integrated plant
producing coke, iron, and steel. The problems will be smaller
for specialty plants or for miniplants.
The above models could be extended by including all plants
in the industry or region. There are two approaches for modeling
the industry or region.
First Approach
Each plant is treated independently and the model is enlarged
by increasing the number of decision variables (320 times the
number of plants) and the constraints (350 times the number of
plants). Although this approach is feasible, it would be an
extremely difficult problem to solve. For example, if 50 plants
are to be considered, the problem will have 16,000 decision
variables and 17,500 constraints. Solving this problem may
require pushing the existing computer technology to its limit.
Additionally it is doubtful that any new insights will be gained
by solving the problem in this manner for the whole industry.
Such a formulation, however, may be suitable for a region with no
more than 10 or 12 plants; in which case the problem is solvable
by current technology. This approach would minimize pollutant
levels within a specific region or industry.
A-6
-------
Second Approach
The characteristics of each plant in the industry or region
are combined to yield a hypothetical aggregate plant, whose
production will be the sum of production of all plants and whose
cost parameters are the sum of costs, etc. Although individual
plant differences are masked, this approach provides useful in-
formation regarding control options that are preferred for the
industry or the region. Additionally the problem could be solved
with available computer technology. Individual plant configura-
tions could then be derived from the regional or industry solu-
tion, or the regional or industry solution could be used as a
standard for comparison.
These two approaches or other modifications will be devel-
oped depending upon user requirements.
Cost Functions
In the models formulated above, the costs and efficiencies
of pollutant removals of different control alternatives are
assumed to be linearly related. In actual practice this is not
true in a number of situations.
Cost of a control option, first of all, depends on the size
of the plant (see Reference A-3, p. 83). Even for the same
plant, the cost of a control option may depend on the efficiency
desired. In the following paragraphs, five typical cost functions
are described.
A-7
-------
(1) Linear—the control alternative can operate at any
level of efficiency (between 0 and 100%), and the cost
of the alternative is linearly related to the effi-
ciency (i.e., c = k-efficiency). (See Figure A-l.)
This is the simple linear cost assumed in linear
programming. This cost function is suitable for the
y. variables described in the models.
D*
(2) Exponential—The control alternative can operate at all
levels as in (1) above, but the cost increases ex-
ponentially with increases in efficiency (see Figure
A-2). Such cost functions are usually handled by
breaking the function into smaller linear segments and
using the piecewise linear functions in the model.
(3) Single-point efficiency—The control option can operate
only at one efficiency; it does not have to be in-
cluded; and it costs a specific amount (see Figure
A-3). This function is typical for an integer variable
(xjk).
(4) Discrete efficiency values—The option can operate at
specific efficiency levels by addition of certain add-
on features (see Figure A-4). This function can be
treated as a number of integer value variables; each
efficiency point corresponds to one system and one
cost.
(5) Nonlinear cost functions within a small range of effi-
ciency—The option can operate within a range of
efficiencies, and the cost is nonlinearly related to
the efficiency (see Figure A-5). This function can be
handled similar to (2) above.
While developing the MIP, the actual cost functions for each
control option will be examined in detail and will be appro-
priately transformed for use in the model.
A-8
-------
CO
o
o
0 EFFICIENCY, % 10°
Figure A-l. Linear cost function.
CO
o
o
° EFFICIENCY, % 10°
Figure A-2. Exponential cost function.
co
O
o
EFFICIENCY, %
100
Figure A-4. Discrete efficiency
values of a cost function.
CO
o
EFFICIENCY, %
100
Figure A-3. Single-point efficiency
cost function.
CO
o
o
EFFICIENCY,
100
Figure A-5. Nonlinear cost function
within a small range of efficiency.
A-9
-------
REFERENCES FOR APPENDIX
A-l Wagner, H. Principles of Operations Research. 2nd ed.
A-2 Lee, S.M. Linear Optimization.
A-3 Kemner, W.F. Cost-Effectiveness Model for Pollution Control
at Coking Facilities. EPA-600/2-79-185. August 1979.
A-10
------- |