Multi-objective Differential Evolutionary Algorithm Based Subregion

Evolution algorithm is a type of random optimistic algorithms, studing the natural selection between biological and natural evolution. Since it is suitable for solving nonlinear problems have highly complex, evolution algorithm has been a very wide range of applications. In the real world, optimization problems are usually multi-at tribute, and therefore can be attributed to multi-objective optimization problem(MOP). Since these objectives are contradictory and conflicting, there is no solution to minimize all objectives simultaneously. Therefore, we need to employ a fitness assignment method to choose better individuals. So far, the best known of fitness value strategy is based on fast sort method and the scalar strategy. In addition to find the optimal solutions, the distribution of the solutions is most important work, among clustering density method, niche method and weighting method are the most commonly used. DE is a popular algorithm with some features that easy to understand and implement, simplicity, litter parameters. In dealing with nonlinear problems, single-objective optimization and MOP present a good robustness. Over the years, DE have successfully used to solve many practical problems, but theoretical research is still necessary to further study.A novel multiobjective DE algorithm using the subregion and external set strategy MOEA/S-DE) is proposed in this paper, in which the objective space is divided into some sub-regions and then independently optimize each subregion. An external set is introduced for each subregion to save some individuals ever found in this subregion. An alternative of mutation operators based the idea of direct simplex method of athematical programming are proposed:local and global mutation operator. The local mutation perator is applied to improve the local search performance of the algorithm and the global mutation operator to explore a wider area. Additionally, a reusing strategy of difference vector also is proposed. It re-uses the difference vector of the better individuals according to a given probability. Compared with traditional DE, the crossover operator also is improved. In order to demonstrate the performance of the proposed algorithm, it is compared with the MOEA/D-DE and the hybrid-NSGA-II-DE. The result indicates that the proposed algorithm is efficient.