Академический Документы
Профессиональный Документы
Культура Документы
Based Approach
A K Gupta, Non-member
Dr R K Shrivastava, Fellow
This paper develops and demonstrates the solution for integrated water treatment plant design using genetic algorithm (GA). The GA approach
is found to be useful optimization tool, capable of providing optimal design parameter estimates while incorporating design and effluent quality
constraints. The GA approach has been presented with a case study and the solution obtained using GA is compared with the solution obtained
by dynamic programming (DP) . The results obtained confirm the capability of proposed method for optimal design of water treatment plant.
The CPU time required is almost one-tenth and storage space required is very-very less eliminating the problem of ‘curse of dimensionality’ posed
by DP programming. Additionally the information provided by GA can be utilized to test the model sensitivity.
GS : velocity gradient in flocculation unit, 1/s Recently the application of genetic algorithm is increasing in the
area of environmental engineering and water resources engineering.
TS : detention time in flocculation unit, s A genetic algorithm is a stochastic search procedure that uses proba-
bilistic rather than deterministic search rules. The objective func-
SOR : surface over flow rate in sedimentation tank, m3/m2/h tion magnitude, instead of derivative information is used directly
in search, therefore allowing GA to be applied to non-convex, highly
FR : filtration rate in filter unit, m3/m2/h
non-linear and complex problems6. Accordingly, this paper, pre-
H : head loss in filter unit, m sents a genetic algorithm based solution to optimal water treat-
ment design problem with design and effluent quality constraints.
Frun : time of filter run, h
In this paper, GA has been applied to a water treatment plant
INTRODUCTION design problem and the results obtained are compared with the
results obtained by dynamic programming. These results show ca-
Water treatment plant (WTP) is a part of urban water supply pability of GA to provide good solution to WTP design problem
system accounting 15% to 20% of the total expenditure incurred along with a solution to problem of ‘curse of dimensionality’.
on urban water supply. The principal unit processes involved in
conventional water treatment system include coagulation-floccu- OPTIMIZATION MODEL FORMULATION
lation, sedimentation, filtration and disinfection. The traditional
practice in water treatment plant design is to design individual treat- The primary objective of the water treatment plant design is to
ment unit in the treatment train based on standards. As the perfor- determine design and operating parameters of the process such
mance of each treatment unit affects the efficiency of subsequent that the total cost is minimal and effluent quality meets the set
treatment units (eg , increased detention time in flocculation might standards. In this study an optimization model has been devel-
result in larger floc formation, better removal in sedimentation, oped, which combines the GA with simulation model for integral
and larger filter runs), ideally therefore decision should be made water treatment analysis developed by Mhaisalkar,etal 5. The
with regard to interaction between various unit operations1. The details of the simulation model are illustrated in Mhaisalkar, et al 5
funds needed for construction, maintenance and operation of the and CPHEEO Manual7. The model considers monodispersed and
system imply that a good compromise between technical and eco- homogeneous suspension and its mass concentration as sole
nomic aspects must be achieved. Lawler, et al 2 first presented inte- parameter to describe the quality of suspension and measure of
grated simulation model of WTP later modified by Ramaley, et al 1. performance of treatment unit. For a conventional water treat-
Few attempts have been made for optimal design of integrated ment system that include coagulation-flocculation, sedimentation
water treatment plant based on non-linear programming3,4 and and filtration, the optimization problem is expressed as
Explicit bound constraints are used to set the limit on the explicit
decision variables of the design problem. The explicit decision mechanics of evolution in nature and in essence, consist of a popu-
constraints for various treatment units are presented in Table 1. lation of bit strings transformed by three genetic operators, such
as, selection, crossover and mutation. Each string (chromosome)
Implicit Constraints represents a possible solution to the problem being optimized and
each bit (or group of bits) represent a value of variable of the
Implicit constraint includes design constraints and effluent quality
problem (gene). A population of chromosomes represents a set of
constraints as shown in Table 2.
possible solution. These solutions are classified by an evaluation
Cost Function function, giving better values, or fitness, to better solutions.
To incorporate multiple design criteria in the optimization problem, While many coding schemes of gene (decision variable) is possible,
four design criteria to be evaluated from the cost function are Michaelewicz10 indicates that for real valued numerical optimization
(1) Annualized capital cost; (2) Operation and maintenance cost; problems, floating point representation outperforms binary repre-
(3) Energy requirement cost; and (4) Land requirement cost. The sentations because they are more consistent, more precise and lead
relationship for above criteria were taken from Dharmappa, et al 4, 8 to faster execution. They are convenient in keeping one-gene-one-
and Clark9. variable correspondence, in addition to the possibility of finding
best point regardless of initial population. Also for most applica-
GENETIC ALGORITHMS tions of genetic algorithms to constrained optimization problems,
the real coding technique is used to represent the solution to a
Genetic algorithms are search algorithms based on mechanics of problem.
natural selection and natural genetics. Philosophically GAs are based
on Darwin's theory of ‘survival of the fittest’. Genetic algorithm A simple GA starts with initial population of randomly generated
has the following advantage over traditional methods. chromosomes and improve upon an initial population of strings
representing a set of possible solutions by repeated application of
(i) GAs search from a population of points, not a single point. genetic operators to search efficient solution of the problem at
Hence GAs are said to be global optimization techniques. hand. A simple GA capable of producing good practical results is
(ii) GAs use only the value of objective function. The derivatives more commonly composed of three genetic operators, such as,
are not used in the search process. selection, crossover, and mutation.
(iii) GAs use probabilistic transition rules, not deterministic rules. The application of GAs involves the appropriate choice of
the alphabet for representation of the decision variables of the
Genetic algorithms are, by far most popular form of evolutionary problem as well as an efficient selection, crossover and mutation
algorithms. They derive their behaviour from a metaphor of the scheme. In spite successful application of GAs to several kinds of
problems they do not guarantee the identification of global opti-
Table 1 Explicit constraint on decision variables
mum. The solution found by GAs must thus be considered to be
Unit Process Design Variable Notation Unit Range ‘near optimal’11.
-1
Rapid-mix unit Velocity gradient GR s 700 ≤ GR ≤ 1000
Real Coded GA
Detention time TR s 20 ≤ TR ≤ 40
Slow-mix unit Velocity gradient GS s-1 5 ≤ GS ≤ 100 As the independent variables are continuous floating numbers, it
Detention time TS s 600 ≤ TS ≤ 3600
has been found suitable to encode independent variables as a string
of floating point numbers Michaelewicz10. In floating point imple-
Sedimentation unit Surface overflow rate SOR m/h 1 ≤ SOR ≤ 1.6
mentation each chromosome was coded as a vector of floating
Filtration unit Filtration rate FR m/h 4.8 ≤ FR ≤ 7.2
point numbers, as the same length as the solution vector.
4 IE(I) Journal−EN
The precision of such representation depends on underlying ma- mutation probability (pm ) to prevent premature convergence to the
chine, but is much better than binary representation. Of course, local optima. When used sparingly with the reproduction and cross-
introducing more bits can always extend the precision of binary over operators, mutation serves to safeguard against a premature
representation but this considerably slows down the algorithm. loss of important genetic material at a particular position. For a
Parents are initialized by selecting random number in the range given parent X, if the element Xk of it is selected for mutation, the
(0,1) for each element of chromosome. Each chromosome Xi = 1, resulting offspring is X’ = [Xl… X’k… Xn], where, X’k is a random
2,… ng in the population is converted in the form appropriate for number in the range (0,1).
evaluation (actual problem variables, Xa ) and is then assigned a
fitness value, according to the objective function. Moreover, the In this study probability of mutation (pm) is kept dynamic, ie, the pm
lower and upper bound inequality condition has been embedded is varied from generation to generation in the range of 0.5-0.01.
in the coding it self by the elements of chromosomes in the range The large value of pm transform the GA into a pure random search
(0,1) using following encoding scheme. initially while the small value of pm at later stage avoids conver-
gence to local optimum and also avoids disruption of near optimal
Xa = X l + ( X u − X l )× X (3) solution12. The probability of mutation (pm) at any generation ‘t ’
can be expressed as
where Xl and Xu are lower and upper bound respectively on
problem variables and Xa is the actual solution vector and X is the pm = 0.5 − (0.5 × t / t max ) (4)
normalized solution vector. The GA operators (selection, cross-
over and mutation) are applied for several generations until one of where tmax= maximum number of generations.
the individuals of population converges to an optimal value or the
required number of generations is reached. GA FORMULATION
GR 700.0 700.0 700.0 700.0 700.0 700.0 SOR 1.5976(1.6) 1.5958(1.6) 1.5874(1.6) 1.5988(1.6)
TR 20.0 20.0 20.0 20.0 20.0 20.0 FR 7.188(7.2) 7.1808(7.2) 7.1856(7.2) 7.1856(7.2)
GS 23.525 22.575 20.39 21.435 21.815 22.0 Cost 544443.0 714000.0 631998.0 631837.0
TS 850.20 891.60 988.80 938.40 934.80 910.0 (539049.0) (707740.0) (630894.0) (630894.0)
SOR 1.5918 1.5946 1.5970 1.5988 1.5952 1.600 * Values in parenthesis corresponds to solution found by DP based procedure
6 IE(I) Journal−EN
the size of floc produced in the flocculation, which affects their 4. H B Dharmappa, S Vigneswaran, J Vernik and O Fujiwara. ‘Water Treatment
removal in sedimentation tank and consequently influent to filter unit. System Design for Turbidity Removal : Optimization.’Journal of Environmental
Engineering Division, ASCE, vol 120, no 4, 1994, p 901.
CONCLUSION 5. V A Mhaisalkar, J K Bassin, R Paramasivam and P Khaana. ‘Dynamic Programming
Optimization of Water Treatment Plant Design.’ Journal of Environmental Engineering
The genetic algorithm has been potentially investigated for its Division, ASCE, vol 119, no 6, 1993, p 1159.
application to integrated water treatment plant design. Experiment
6. D E Goldberg. ‘Genetic Algorithm in Search, Optimization and Machine
with a case study was performed to demonstrate the GAs ability to Learning.’ Addison-Wisely, Reading, Mass, 1989.
obtain optimal design parameter with multiple constraints. The
penalty function method is employed to handle the multiple con- 7. Manual on Water Supply and Treatment. CPHEEO, Ministry of Urban Development,
Government of India, New Delhi, 1991, p 495.
straints. Following the case study experiment, this study demon-
strates the capability of GA to accurately find the optimal solution 8. H B Dharmappa, J Vernik, O Fujiwara and S Vigneswaran. ‘Optimization of
Granular Bed Filtration Treating Poly-dispersed Suspension.’ Water Research, vol 26,
of WTP design problem. The method requires less computational
no 10, 1992, p 1307.
time and storage space as against DP. Furthermore, the GA popu-
lation can be used to learn more about the model sensitivity. Thus, 9. R M Clark. ‘Cost Estimating for Conventional Water Treatment.’ Journal of
Environmental Engineering Division, ASCE, vol 108, no 5, 1982, p 819.
GA has been demonstrated to be quite promising as a useful opti-
mization tool in environmental modelling. 10. Z Michaelwicz. ‘Genetic Algorithm + Data Structure = Evolutionary
Programmes.’ Springer, 1994.
REFERENCES
11. L F R Reis, R M Porto and F H Chaudhry. ‘Optimal Location of Control
1. B L Ramaley, D F Lawler, W C Wright and C R O’Melia. ‘Integral Analysis of Valves in Pipe Network by Genetic Algorithm.’ Journal of Water Resources Planning
Water Plant Performance.’ Journal of Environmental Engineering Division, ASCE, and Management, ASCE, vol 123, no 6, 1997, p 317.
vol 107, no EE3, 1981, p 547.
12. M Srinivas and L M Patnaik. ‘Adaptive Probabilities of Crossover and
2. D F Lawler, C R O’Melia and J E Tobiason. ‘Integral Water Treatment Plant Mutation in Genetic Algorithms.’ IEEE Transactions on System, Man and Cybernetics,
Design from Particle Size to Plant Performance.’ Advances in Chemistry, no 189, SMC-24 (4), 1994, p 16.
American Chemical Society, Washington D C, 1980.
13. S Baskar, P Subbaraj and M V C Rao. ‘Hybrid Genetic Algorithm Solution to
3. M R Wisner, C R O’Melia and J L Cohen. ‘Optimal Water Treatment Plant Design.’ Economic Dispatch Problem with Multiple Fuel Option.’ Journal of The Institution
Journal of Environmental Engineering Division, ASCE, vol 113, no 3, 1987, p 567. of Engineers (India), vol 82, 2001, p 177.