改进的分布估计算法求解多目标优化问题

Improved Estimation of Distribution Algorithm for Multi-objective Optimization Problems

为了进一步提高多目标分布估计算法的收敛性和多样性,提出了一种改进的多目标分布估计算法,其基本思想是:通过正交设计来产生初始化种群,使初始种群均匀地分布在可行解域;引入精英策略,防止最优解的丢失,同时利用小生境技术来维护精英种群,避免早熟现象;加入遗传算法来进化种群,在算法初期使用分布估计算法进行快速的全局搜索,在算法后期则主要利用遗传算法的交叉变异进行局部寻优,增强算法的局部搜索能力。在数值仿真实验中选取4个测试函数进行实验,并同其他算法进行了多方面的比较,结果表明所提算法具有良好的收敛性和多样性。

In order to improve the convergence and accuracy performance of multi-objective estimation of distribution algorithm,and enhance the local search capability,an improved multi-objective distribution optimization algorithm has been proposed.The basic idea of new method is using orthogonal design to initialize the population,which makes the algorithm can search in the whole feasible space,introducing the improved elitist strategy to avoid the loss of the optimal solution,while using the niche technology to maintain elite populations and prevent premature,importing genetic algorithm to evolve populations,the estimation of distribution algorithm makes use of in the early stage of the algorithm to search the global space quickly and the genetic algorithm is mainly used to local optimization in the later stage. Four test functions are used in numerical experiment. The numerical results show that the proposed algorithm has a better convergence and diversity performance by compared with two other algorithms.