一种基于多子群的动态优化算法

Multi-swarm based optimization algorithm in dynamic environments

针对动态粒子群优化算法的群体多样性问题,提出一种新的度量方法。为了提高群体多样性,在每次迭代前,子群内部各粒子以一定的概率飞离局部最优粒子,以保持子群内部粒子多样性。在此基础上,提出一种动态粒子群优化算法,即在每次迭代前,要淘汰超规模子群中的低适应值粒子,进一步增强整个群体的多样性水平,提高算法的鲁棒性。用标准测试函数MPB测试该算法跟踪动态全局最优值的能力,实验结果表明:该算法能有效跟踪5维以上的动态全局最优值,子群内部多样性水平提高60%以上。

A diversity measure was proposed for multi-swarm particle swarm optimization algorithms in dynamic environments. In order to improve the diversity of the swarm, each particle flew randomly away from its best neighbor particle before updating its velocity and location, and the redundant particles in each sub-swarm were replaced with random particles in search space. Subsequently, the change was detected by re-evaluating the objective function at the memorized best location of each particle. The results show that the diversification method can improve the diversity of sub-swarms with 60% more than that of a representative algorithm, and the proposed algorithm can efficiently track changing global optimum in high dynamic environments.