The growth of cities, associated with the lack of investment in basic infrastructure, has rendered water supply systems complex and difficult to operate. The efficient operation of such systems is a fundamental tool for extending the system’s service life as much as possible, thus ensuring a reliable service to the consumers while keeping electrical energy and maintenance costs at acceptable levels. Efficient operation requires knowledge of the system, supported by tools such as models for hydraulic simulation, optimization, and definition of rules, provides the operator with proper conditions for the rational operation of the system’s units. This paper aims to develop a computational model for the optimal operational control of macro water distribution systems using the EPANET2 hydraulic simulator, SPEA (Strength Pareto Evolutionary Algorithm) multiobjective genetic algorithms, and data mining to extract operational rules for the system. The studies were conducted on the macro system of the city of Goiânia, Brazil, chosen due to its complex characteristics, showing that solutions for its satisfactory operation can be quickly produced as a substitute to the personal judgment of the operator. Introduction and Rationale The concept of systems operation, understood by laypersons as a mere sequence of equipment commands whose objective is to meet the demand (Zahed Filho, 1990), is actually far more complex, involving aspects of planning, control and supervision, and infrastructural consumer support and services, considered simultaneously and interdependently. The operation plan requires that at least four basic conditions be met: a) a clear definition of the objectives to be achieved; b) the availability of mathematical analysis models; c) equipment to process these models; and d) knowledge of the system (Luvizotto Jr, 1995). Seeking greater reliability in the establishment of the system operating rules, new hydraulic techniques associated with optimization algorithms have been developed. Righetto (2002) emphasizes that the interface between models for 1 Designer Manager, SANEAGO S.A., Goiânia, Brazil and Honorary University Fellow, Department of Engineering, School of Engineering and Computer Science, University of Exeter, North Park Road, Exeter, EX4 4QF, UK; phone 44 1392 263646; tinil@sc.usp.br and I.B.Carrijo@ex.ac.uk 2 Associate Professor – Sao Carlos School of Engineering, University of Sao Paulo, Av. Trabalhador Sao Carlense, 400, Sao Carlos, Sao Paulo, Brazil; phone 55 16 2739545; fernanda@sc.usp.br 3 Professor, Department of Engineering, School of Engineering and Computer Science, University of Exeter, North Park Road, Exeter, EX4 4QF, UK; phone 44 1392 263633; G.A.Walters@ex.ac.uk 4 Professor, Department of Engineering, School of Engineering and Computer Science, University of Exeter, North Park Road, Exeter, EX4 4QF, UK; phone 44 1392 263637; D.Savic@ex.ac.uk hydraulic simulation, optimization and definition of operating rules must be built carefully to make the model transparent, facilitate the introduction of restrictive inequalities and obtain objective function values in the successive steps required by an optimizer. The purpose of this work is to present a methodology to achieve the optimal operation of water distribution systems, essentially macro systems (skeleton), concerning the costs of the operation and the hydraulic benefits. It represents an attempt to provide appropriate operation rules in order to minimize costs and maximize hydraulic benefits. Based on the knowledge of the system, provided by technical and commercial georeferenced records, the purpose is to optimize its operation through multiobjective genetic algorithms (MOGAs). This is supported by a realistic hydraulic simulation model of the system behavior, and the production of operational rules through the data mining process. Methodology The optimization model implemented here takes into account two objectives: the minimization of the operational costs and the maximization of the hydraulic benefits. Where the hydraulic benefits are considered as the index of demand met, adequate levels of water in the tanks, and minimum and maximum pressures at the demand points for a 24-hour period of analysis. Hydraulic Simulation of the System.The hydraulic simulation evaluates the system’s response to operational decisions in terms of the state variables, i.e., pressure, flow rate and tank level. It is therefore an essential tool for the computational routine, which evaluates the established objectives. EPANET2, via Toolkit Library, Rossman (2001), is used for this purpose. Operational Optimization using Multiobjective Genetic Algorithms (MOGAs). According to Deb(2001) and Deb et al.(2002), since 1993, different evolutionary algorithms have been proposed for the solution of multiobjective optimization problems. The Multi-Objective Genetic Algorithm (MOGA), the Niched-Pareto Genetic Algorithm (NPGA), and the Non-dominated Sorting Genetic Algorithm (NSGA), were the precursors of this technique, whose basic characteristics are: evaluation of the members of a population based on the Pareto dominance concept and on preservation of the diversity of solutions. Although these algorithms have proven efficient in obtaining multiple non-dominant solutions to various engineering problems, researchers have suggested the introduction of elitism to improve their convergence properties. Several algorithms stand out among the multiobjective evolutionary algorithms that consider elitism, i.e., the Strength Pareto Evolutionary Algorithm (SPEA and SPEAII), the Pareto Archived Evolution Strategy (PAES), the elitist GA of Rudolph, Pareto Envelope-based Selection Algorithm (PESA and PESAII ) and Non-dominated Sorting Genetic Algorithm (NSGAII). This work uses the elitism-based SPEA method. Uniform crossover and nonuniform mutation were adopted, following an analysis of the results of several tests using various different operators (Cheung et al., 2003). Extraction of Rules using Data Mining. There is a set of methods known as expert systems or knowledge-based systems whose classification models can be developed according to two main routes. The first obtains rules for the model through interviews based on experts and the inclusion of previous knowledge in the system. The second creates an inductive model through the generalization of a large record of collected and classified data. According to Bessler et al. (2003), the method called data mining used in this work belongs to the second aforementioned route, which creates a classification model through the discovery and analysis of patterns that can be found in the data records. To apply the algorithms, several specific characteristics of the data must be analyzed. All the information about the cases (or examples) has to be presented in the form of attributes and each case is allocated to a discrete predefined class. The main function of a data mining program is generally to construct classification models as decision trees for later application. That, however, is not the main objective in this work. The classifier called rulesets is used to extract operational rules from a set of examples (cases) supplied by the optimization model (Pareto front). The decision tree tool SEE5, which is the most recent version of the C4.5 inducer described by Quinlan (1993), is used for this purpose. Description of the Problem Focusing on the development of a flexible tool that is easily handled by water supply systems operators, clearly providing a set of operational rules according to the working conditions of each unit of the system, part of the macro piping system of Goiânia, Brazil, was considered for analysis and evaluation of the results. For a clearer picture of the proposed application a diagram is shown of the system under study (Figure 1). Definition of the Objective Functions Several studies developed in the past showed that, of all the parameters relating to the operational costs, the most relevant one is the cost of electric energy consumption at water pumping stations. Another possibility is the system’s reliability in meeting consumer needs coherently. In this case, several parameters can be listed. The reliability of water supply systems can be considered from a hydraulic or mechanical standpoint. The former involves physical parameters that vary according to the operational changes in the system, while the latter involves the possible interventions on equipment. As in Walters et al. (1999), this work evaluates two basic objectives, the economic objective and the objective of hydraulic benefits of the water distribution systems. In the case of the economic objective, the intention is to minimize the costs of electrical energy consumption at the pumping stations. The daily cost for each pump at a pumping station is given by the sum of the cost of the maximum demand factor and the measured cost of consumption.
[1]
Dragan Savic,et al.
Improved design of “Anytown” distribution network using structured messy genetic algorithms
,
1999
.
[2]
Kalyanmoy Deb,et al.
A fast and elitist multiobjective genetic algorithm: NSGA-II
,
2002,
IEEE Trans. Evol. Comput..
[3]
Rudy Gargano,et al.
RELIABILITY AS TOOL FOR HYDRAULIC NETWORK PLANNING
,
2000
.
[4]
Kalyanmoy Deb,et al.
Multi-objective optimization using evolutionary algorithms
,
2001,
Wiley-Interscience series in systems and optimization.
[5]
Kamel Zahed Filho.
Previsão de demanda de consumo em tempo real no desenvolvimento operacional de sistemas de distribuição de água.
,
1990
.
[6]
Edevar Luvizotto Junior,et al.
Controle operacional de redes de abastecimento de agua auxiliado por computador
,
1995
.
[7]
J. Ross Quinlan,et al.
C4.5: Programs for Machine Learning
,
1992
.
[8]
Dragan Savic,et al.
Water Reservoir Control with Data Mining
,
2003
.
[9]
L Reis,et al.
Multi-objective optimization to the rehabilitation of a water distribution network
,
2003
.