Research On Hybrid Evolutionary Algorithm Based On Differential Evolution And Estimation Of Distribution

Global optimization problems are used almost in each subject, engineering field and business. For example, an engineer needs to design cars with best performance, so he must optimize configuration parameter of the cars in order to realize this target. The search of best configuration parameter belongs to global optimization category, and a lot of jobs always devote to solve relative global optimization problems, but the main challenge of continuous global optimization problem is that there are always many local optimal solutions in this problem. Differential evolution algorithm has pretty well performance in solving this continuous global optimization problem. It leads further search through using obtained distance and direction information by people nowadays. The probability model of differential evolution algorithm is extracted from a promising solution set and generated new sample according to it. Three hybrid algorithm models based on DE and EDA integrate the advantages of these two algorithms in this paper, to solve continuous global optimization problem. Three algorithms are tested through classic benchmark functions, and the best DE algorithm is compared with EDA algorithm. The simulations prove that the three models are better than DE and EDA algorithm, and they are analyzed from each other through investigating the performance parameter results.Three hybrid algorithm models are proposed based on DE and EDA algorithm in this paper as following:1) gene hybrid model:The genes in genetic chain partially come from DE algorithm and partially come from EDA algorithm if each individual is compared to a genetic chain, this model makes individuals have more diversity; 2) individual hybrid model:The individuals in society partially come from DE algorithm population and partially come from EDA algorithm population if the population is compared to a society, this model makes population has more plurality; 3) EDA guides the mutation direction of DE:The temporary individual generated according to statistical probability by EDA guides individuals to mutate toward excellent direction as a leader if the population is compared to a team, this model makes local and global research have better performance.Three algorithm models are tested through classic benchmark functions finally in this paper, and the performance parameter results are investigated through comparing DE algorithm with EDA algorithm. The 5480 groups of test results prove that the three models have better performance than DE and EDA algorithm in multimodal function and cheat function. The performance parameters of each model are analyzed through comparing the characteristics of three models.