熵值理论在多目标演化中的应用研究

Application and research of entropy theory in multi-objective evolution

为了克服传统多目标演化算法在进化后期遗传操作可能失效使算法性能降低以及基于概率建模的多目标算法在演化初期由于种群分布尚未呈现一定规律,采样产生的新个体的搜索方向同目标方向存在差异,提出一种基于熵值的多目标演化算法(entropy based multi-objective evolutionary algorithm,EB-MOEA)。算法利用种群进化过程中,个体分布存在从无序到有序的现象,设计了一种基于熵值理论的种群分布计算方法,并将其作为种群从无序到有序过渡的判定准则,指导遗传操作和概率建模操作切换的时机。新算法采用ZDT、DTLZ系列测试集进行实验,通过与NSGA-Ⅱ以及RM-MEDA算法的实验对比,证明了新判断准则的有效性,EB-MOEA具有更好的寻优性能。

In the later stage of multi：objective evolutionary algorithm, the traditional genetic operation may be invalid, thus the performance of the algorithm will be reduced, while at the early stage of probabilistic modeling, the search direction of the sam- piing new individuals may be differ from the exact direction for the lack of distribution rule of the population. The paper proposed an entropy-based multi-objective evolutionary algorithm（EB-MOEA） ,it used the phenomenon that the population went from disorder to order in the process of evolution. It designed a distribution calculation method and used as the criteria which could guide： the switching time of genetic operation and probabilistic modeling operation. The new algorithm adopted ZDT, DTLZ test suits to conduct the comparison experiment with NSGA- II and RM-MEDA,results show that the effectiveness of the new criterion and the proposed algorithm has better optimization performance.