Abstract: To solve the shortcomings of low accuracy and low speed in convergence of MOEA/D’s Differential Evolution（DE）, a novel DE model based on Controlling Dominance Area of Solutions（CDAS） and dynamic scaling factor is proposed. It can guide explicit and implicit search. And an improved MOEA/D-iDE algorithm based on a new differential evolution model is realized. The new DE model classifies the individuals of the neighborhood by non-dominated sorting of CDAS and produces a vector difference of evolutionary direction to achieve explicit convergence speed guide in the different evolutionary period. Analysing decision space by PCA, the scaling factor is adjusted dynamically to achieve the implicit convergence accuracy guide. The MOEA/D-iDE algorithm is compared with 6 optimization algorithms in IGD+ and ER indicators by testing ZDT, DTLZ and WFG benchmarks. The experimental results show that MOEA/D-iDE has a better performance in speed and accuracy of convergence than other algorithms, and validate the effectiveness of the new DE model.

Key words: differential evolution, controlling dominance area of solutions, principal component analysis, Multi-Objective Evolutionary Algorithm based on Decomposition（MOEA/D）

摘要： 针对MOEA/D算法中差分进化操作收敛精度不高且速度较慢的不足，提出了一种综合基于可控支配域的向量差生成策略和基于主成分的动态缩放因子的新型差分进化模型，均衡显性与隐性搜索引导；并实现了一种基于新型差分进化模型的MOEA/D改进算法（MOEA/D-iDE）。新型差分进化是借助基于可控支配域的非支配排序对邻域进行分层，根据分层信息生成与不同进化阶段相匹配的向量差，实现对种群收敛速度的显性引导；同时对决策空间进行主成分分析，动态调整差分进化缩放因子，实现对种群收敛精度的隐性引导。实验选取ZDT、DTLZ和WFG等为测试问题，以IGD+，ER作为评价指标，将MOEA/D-iDE算法与6个同类算法进行对比实验，结果表明新算法在保证多样性的同时具有更好的收敛速度与精度，从而验证了新型差分进化模型的有效性。

关键词: 差分进化, 可控支配域, 主成分分析, 基于分解的多目标进化算法