基于独立成分分析的多目标分布估计算法

针对复杂的强耦合、非线性连续多目标优化问题，提出了一种基于独立成分分析方法（IC A ）的多目标分布估计算法。假设其概率图模型为非高斯的，采用ICA进行分离产生独立的各分量，接着采用基于拥挤距离排序和NSGA-Ⅱ的非支配排序，选择出优秀个体作为新的种群。与多目标ICA-UMDA 和vbICA-MM 的比较实验表明，该算法在测试函数ZDT2-2、ZDT4-2、ZDT6-2和F5上获得的Pareto解集具有较好的收敛性与多样性，且数据没有服从高斯分布的限制。

Multiobjective Distribution Estimation Algorithm Based on Independent Component Analysis

Multiobjective estimation of distribution algorithm based on independent component analysis（ICA-MOEDA） is proposed for solving multiobjective optimization problems .The non-Gaussian probabilistic graphical model is introduced in the nonlinear variable linkage continuous optimization problems .Then the ICA model is performed on the parent population to get the new independent population which is clustered .The selection procedure is based on the non-dominated sorting of NSGA-Ⅱ and crowding distance which choose the best offspring enter next generation .Compared with other two evolution-ary multiobjective algorithms ,simulation results show that the improvement algorithm has good convergence and diversity performance on ZDT2-2 ,ZDT4-2 ,ZDT6-2 and F5 benchmark instances and the variables are not required to subject to Gaussian distribution .