Improvement Research Of Differential Evolution Algorithm

Differential Evolution(DE) algorithm, proposed by Storn and Price, is a new intelligent optimization algorithm, which is based on real number encoding and has a good capability of global search. This algorithm has the advantages of simple operating principle, few control parameters, strong stability, high robustness and easy implementation, etc. As a result, the algorithm has gotten more and more recognition in the field of evolutionary computation and its applicability is wider and wider. This thesis firstly expounds differential evolution algorithm in detail and analyzes its advantages and disadvantages. Then some relevant improvements have been proposed aiming at the existing problem of DE and some adaptive adjustments of control parameters have been made. Finally the relevant numerical test proves the feasibility and effectiveness of the improved algorithm.The main research results of this thesis are as follows:(1)The research significance, basic principle, operation process, working schematic diagram and pseudo code of DE is elaborated. In addition, several common extended forms of DE are introduced. Through analyzing domestic and foreign research progress of DE, relevant improvements of DE are classified and summarized. Finally compare DE with other algorithms and analyze the advantages and disadvantages of it.(2)Pattern search algorithm, a kind of algorithm with strong local search ability, was integrated into the searching process of DE, which can make up the defects of weak local searching ability of DE. So a new algorithm whose performance goes beyond its maternal algorithm was born. At the same time, the new algorithm introduces a mechanism which can judge premature convergence. Firstly, the new algorithm performs basic operations of DE algorithm. If the algorithm traps into local convergence in the later phase, make a pattern search to these individuals and update the best individual and the optimal solution after pattern search, which can improve the optimization performance of the algorithm further.(3)An improved parameter self-adaptive differential evolution algorithm is introduced in the basis of DE algorithm. The scale factor is adjusted according to the functional fitness of individuals and changes as the position of the two individuals which generated difference vector in the scope of feasible solution. The crossed factor is adjusted adaptively according to variation individual vector and their fitness. Meanwhile this thesis will choose the new generation after mutation, crossover and selection to compete with new random group, which ensure that individuals can approach to the optimal value stably and quickly. Finally, use the improved algorithm and DE algorithm to do a performance test respectively and the test results prove the feasibility and effectiveness of the new algorithm.