基于随机邻域策略和广义反向学习的自适应差分进化算法

全局探索和局部开发能力之间的平衡以及对控制参数的整定是影响差分进化(differential evolution, DE)算法性能的主要因素。针对这两个问题, 提出一种基于随机邻域策略和广义反向学习的自适应DE算法。首先, 在每一代进化过程中, 算法从当前种群为每一个体随机选择相应的邻域, 其中最优个体作为基向量执行变异操作, 邻域中个体数量随进化动态更新。其次, 采用基于历史存档的自适应参数整定方法, 进化进程中根据“精英”信息动态更新算法各参数。最后, 在初始化和每一代进化结束阶段, 执行基于广义反向学习策略的种群初始化和种群“代跳”操作。通过基于27个标准测试函数的3组仿真实验, 验证了所提算法具有寻优精度高、收敛速度快、鲁棒性强的优点。

Self-adaptive differential evolution algorithm with random neighborhood-based strategy and generalized opposition-based learning

The balance between global exploration and local development and the tuning of control parameters can be two main factors that extremely influence the performance of differential evolution (DE) algorithm. To solve these two problems, a self-adaptive DE algorithm with random neighborhood-based strategy and generalized opposition-based learning is proposed. Firstly, at each generation, the neighbors of the individuals from current population are selected at random, in which the finest one is selected as the base vector to execute the mutation operation, and the number of each individual in the neighborhood is dynamically updated with evolution process. In addition, a history-driven parameter self-adaptation method is implemented to adaptively update parameters during the evolution process of DE with the elite information. Finally, at the phase of initialization and the end of each generation, the generalized opposition-based learning strategy is applied to execute the initialization and generation jumping of population. Through three groups of simulation experiments based on 27 benchmark functions, the proposed algorithm is proved to have high searching accuracy, fast convergence speed and strong robustness.