Optimization of multipurpose reservoir operation using evolutionary algorithms / Mohammed Heydari

Mohammed , Heydari (2017) Optimization of multipurpose reservoir operation using evolutionary algorithms / Mohammed Heydari. PhD thesis, University of Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (1647Kb)
Preview	PDF (Thesis PhD) Download (5Mb) | Preview

Abstract

Today, the water resources are among the great human treasures. Optimal reservoir operation, due to the numerous needs, shortcomings and restrictions on the use of these resources is necessary. The main purpose of this study was presenting a model for an optimal operation of multi-purpose dams of water resources systems. In this study, a hybrid evolutionary algorithm model (HPSOGA) and linear programming (LP) has been developed for optimizing the operation of reservoirs with the objectives of maximizing hydroelectric power generation, meeting the water demand for agricultural purposes and predicting the cost and estimating amount of agriculture products. An improved particle swarm algorithm (HPSOGA) is used to solve complex problems of water resources optimization. One of the main problems of this method is premature convergence and to improve this problem, the compound of the particle swarm algorithm and genetic algorithm were evaluated. The basis of this compound is in such a way that the advantages of the Particle Swarm Optimization (PSO) algorithm and Genetic Algorithm (GA) have been applied simultaneously. Two efficient operators of Genetic Algorithm, that is, mutation and crossover are used in the obtained algorithm, the mutation causes an increase in the diversity of the population and the intersection of information between the particles of the population. To evaluate the hybrid algorithm, optimization of hydro-power energy of Karun dams were considered. Cases studied in this research were reservoirs of Karun I, Karun III and Karun IV. The three dams are located in a consecutive series of Karun River in Iran. In order to optimize, 41 years of the common statistical period were used. Then, the optimal output of the problem in the form of curves that represent the desired amount of discharge from the reservoir at a specified time interval were prepared and compared with the Lingo model. The regression analysis and artificial neural networks (ANN) were used to check the quality of the results. By using the Weibull distribution, the base year which is consistent with the percent probability of agricultural needs was determined for downstream of the Karun III dam. To achieve the best cultivation pattern, initially the arable land was categorized into 6 classes and only 2100 hectares of agricultural irrigable land that had the best agricultural conditions were studied. The amount of water allocated to the mentioned land was about 6.240 MCM. Seventeen important agricultural products of the region were used for the modelling. The optimization problem was modelled with the aim of maximizing the ultimate value of agriculture in terms of the number of acres of each crop. The described model was resolved by linear programming and evolutionary algorithms in Microsoft Excel (Solver). The results showed full compliance of these two methods. To estimate and predict the cost of the different stages of farming, and the cost of fertilizers needed for agricultural products, the obtained results of cultivation pattern per acre multiplied to cost breakdown values in tables taken from the ministry of agriculture. Comparing the results of the combination of the PSO and GA algorithms makes clear that the obtained algorithm increased flexibility and improving the ability of the PSO algorithm to create the population with high-speed convergence and it is very applicable to solve the problems of operation optimization of water resources. To compare the accuracy of the results, three criteria were used for RMSE, NRMSD and CV. In all the obtained results, i.e. optimum release, optimum storage and the produced energy, for all dams, the accuracy of HPSOGA was better than GA and GA accuracy was remarkably better than PSO. However, exceptionally, the accuracy of the GA algorithm was approximately 34% better than the HPSOGA algorithm for only the optimal storage capacity at Karun IV Dam. The overall results show that the optimal values have higher importance in the preparation of the rule curve, especially in periods of drought.

Actions (For repository staff only : Login required)