基于成绩标量函数搜索的分解多目标进化算法

An achievement scalarizing function search-based multi-objective evolutionary algorithm

基于分解的多目标优化算法在整个进化过程中由于种群规模和权向量保持不变,容易导致种群多样性下降和早熟收敛。针对这一问题,提出了一种基于成绩标量函数搜索的分解多目标进化算法。为使基于分解的多目标优化算法对决策空间均匀探索,首先通过分析当前种群的稀疏度,设计了一种自适应基于成绩标量函数的局部搜索策略,动态地增加种群规模和权向量;其次,提出了一种具有自适应缩放因子的改进混合差分算子,平衡算法的全局探索和局部开发能力。通过对10个函数进行数值实验,并与5种经典的多目标优化算法和6种基于分解的多目标优化算法相比,数值结果表明所提算法能够获得分布均匀且充分靠近Pareto最优前沿的解集。

An achievement scalarizing function search-based multi-objective evolutionary algorithm is presented to overcome the decrease of population diversity and premature convergence of multi-objective evolutionary algorithm based on decomposition caused by the fixed population size and set of weight vectors during the evolutionary process.In the proposed algorithm,the smaller initial population is first set,and a local search strategy based on achievement scalarizing function with adaptive preference is designed to enhance the search of the sparse region and dynamically increase the population size and weight vector.Then a hybrid differential evolution operator with adaptive scaling factor is proposed to balance global exploration and local exploitation.Different from the existing algorithms,variable population size and dynamically increasing weight vector can avoid the decrease of population diversity and premature convergence.The proposed algorithm is compared with 11 typical algorithms on 10 benchmark functions.Experimental results show that the proposed algorithm obtains a uniform distribution set close to the Pareto front.