求解函数优化的新型差异演化算法

Novel differential evolution algorithm for function optimization

针对差异演化算法存在早熟收敛和后期求解效率低的缺点,提出一种新型差异演化算法。该算法基于单种群,在演化过程中直接对当前种群进行变异、交叉和选择操作,无须差异演化算法中的中间过渡种群。此外,新型差异演化算法的变异与交叉概率是时变的,其中变异概率随着迭代次数的增加而减小;交叉概率随着迭代次数的增加而增加。对几个典型的测试函数进行仿真实验表明,该算法能够有效避免早熟收敛,改善了差异演化算法的优化性能。

This paper proposed a novel differential evolution algorithm to overcome the premature convergence and slow convergent speed during the late evolution in differential evolution algorithm. The new algorithm was based on single population without intermediate population, in which mutation operation, crossover operation and selection operation were used on the current population. In addition, the parameters of mutation and crossover in the new DE were time-varying. The probability of mutation decreased with the evolution, while the probability of crossover was increasing. Results of several typical benchmark functions show the algorithm can avoid premature convergence and improve the performance of differential evolution algorithm in optimization.