A Combined Adaptive Differential Evolution Algorithm with Ensemble of Mutation Strategies for Reducing Irrigation Deficiencies

The optimal utilization of dams water resources in order to meet the water needs of different departments is an important problem in the management and engineering of water resources. Evolutionary algorithms have provided great success to create balancing in supply of water resources shortages with the goal of controlling the water resources storage and dams release rates. Differential evolution (DE) is an efficient search technique and method to solve numerical optimization problems and real-world applications. DE suffers from numerous weaknesses that usually due to the use of a mutation strategy and consideration of constant settings for parameters. In this paper, an adaptive DE algorithm titled Combined Improved Multi-Population Ensemble DE (CIMPEDE) is suggested. In CIMPEDE, the entire population is divided into four sub-populations; three mutation strategies are considered to achieve a better, and two adaptive and self-adaptive schemas are proposed for setting parameters. CIMPEDE has been able to achieve the optimal solution with the highest quality by choosing the best mutation strategy in each repetition, dynamic adjustment of parameters and exchange of information between sub-populations. For prove the performance of CIMPEDE, extensive experiments have been performed on mathematical benchmark functions (entitled CEC 2013) and a real-world problem for modeling of the single-reservoir system of Golestan dam in Iran with the aim of decreasing irrigation deficiencies. CIMPEDE has also made comprehensive comparisons with DE-based applied approaches. In the case of Golestan dam problem, mean absolute error between real demand and released water by CIMPEDE was zero MCM. While, the comparable algorithm in second rank could not get a value better than 1.81E-220 MCM. Also, in test of reliability and vulnerability indices and test of mean violations intensity between demand and released water in an annual average, the best results were recorded among the comparable evolutionary algorithms for CIMPEDE.