基于多目标进化算法混合框架的MOEA/D算法

针对混合多目标进化算法中如何设计全局搜索算法和局部搜索策略结合机制的难点问题以及提高多目标进化算法的求解性能,基于反馈控制思想,提出了一种系统化、模块化的全局优化与局部搜索相结合的混合MOEA/D算法,算法中设计了一种基于拥挤熵的种群多样性度量方法;提出了基于简化二次逼近的局部搜索策略,以及针对MOEA/D的种群多样性增强策略。数值实验表明所提算法具有良好性能,可以兼顾算法求解的多样性和收敛性,所提混合框架可有效提升现有多目标进化算法的求解性能。

MOEA/D Algorithm Based on the Hybrid Framework for Multi-objective Evolutionary Algorithm

Aimto the difficulties of designing the bonding mechanism of global optimization algorithm and local search strategy for hybrid multi-objective evolutionary algorithm, and of improving the performance of multi-objective evolutionary algorithms, based on the feedback control idea, a systematic and modular hybrid MOEA/D algorithm combining the global optimization and local search is proposed. In the algorithm, a diversity measure method based on crowded entropy is designed; a local search strategy based on simplified quadratic approximation and population diversity enhancement strategy for MOEA/D is proposed. The numerical experiments show that the proposed HMOEA/D can achieve a balance between diversity and convergence of algorithm. The proposed hybrid framework can effectively improve the performance of existing multi-objective evolutionary algorithms.