一种基于邻域改进的分解多目标进化算法

Decomposition Multi-objective Evolutionary Algorithm Based on Neighborhood Improvement Strategy

基于分解的多目标进化算法(MOEA/D)是将多目标优化问题分解为若干个简单子问题进行并行求解的方法.然而M OEA/D对不同子问题均采用固定邻域求解,这不利于算法在邻域范围内选择到合适的解替换更新.针对此问题,本文提出一种新的调整邻域大小分配的分解多目标进化算法,以平衡算法的收敛性和多样性.该算法根据子问题距离中心区域的偏离程度,动态调整选择邻域和替换邻域大小.在算法性能对比实验中,将本文提出的算法与MOEA/D、MOEA/D-GR、MOEA/DDRA及M OEA/D-DU在二维ZDT测试函数和三到五维DTLZ测试函数进行性能测试.实验结果表明,本文所提算法与其他几种经典算法相比,在测试函数上解集的整体质量显著提高.

The decomposition-based multi-objective evolutionary algorithm(MOEA/D)decomposes the multi-objective optimization problem into a set of simple optimization sub-problems.However,the sub-problems in M OEA/D use fixed neighborhood size,w hich is not conducive to selection and replacement of solutions.Aiming at this problem,this paper proposes a decomposition multi-objective evolutionary algorithm based on neighborhood improvement strategy.According to the degree of deviation from sub-problems to the central region,it dynamically adjusts the neighborhood size of the selection and replacement neighborhood.In the performance comparison experiment,the algorithm proposed in this paper is tested with M OEA/D,M OEA/D-GR,M OEA/D-DRA and M OEA/D-DU in the 2-objective ZDT test suite and the 3,4,and 5-objective DTLZ test suite.The experimental results show that the algorithm proposed in this paper is better distributed and closer to the Pareto front.The IGD and HV of the algorithm are better than other algorithms.