基于双变异策略的自适应骨架差分进化算法

Self-adaptive bare-bones differential evolution based on bi-mutation strategy

骨架差分进化算法能够较好规避差分进化算法控制参数和变异策略选择问题。针对基于双变异策略的经典骨架差分算法(MGBDE)没有根据个体进化差异选择适合的变异策略和考虑早熟收敛的问题,提出一种改进算法。该算法引入变异策略选择因子,并借鉴自适应差分进化算法的设计思想,将选择因子随个体共同参与进化,使个体执行当前最为适合的变异策略,克服原始算法进化过程的盲目性,同时选择因子的动态自适应特性保持了骨架算法近似无参数的优点;该算法加入停滞扰动策略,降低陷入局部最优的风险。采用18个标准测试函数进行实验,结果表明,新算法在收敛精度、收敛速度和顽健性上整体优于多种同类骨架算法以及知名的差分进化算法。

Bare-bones differential evolution （BBDE） can elegantly solve the selection problem of control parameters and mutation strategy in differential evolution （DE）. MGBDE is a classical BBDE based on bi-mutation strategy. However, it randomly assigns a mutation strategy to each individual, not considering their differences during evolution process, meanwhile it may suffer from premature convergence. To overcome these drawbacks, a modified algorithm based on MGBDE was proposed. A mutation strategy choice factor that guided the individual to choose a preferable mutation strategy at each mutation operation was introduced, avoiding the evolution blindness brought by the random selection of mutation strategy. To retain the almost parameter-free characteristic of bare-bones algorithm, the tuning of choice factor to be adapted was involved in the individual evolution, inspired by the concept of self-adaptive DE. The algorithm also included a well-designed stagnation perturbation mechanism to reduce the risk of trapping into the local optimal. Expe-rimental results on 18 benchmark functions show that the proposed algorithm generally achieves better performance than state-of-the-art BBDE variants and several wel l-known DE algorithms in terms of convergence and robustness.