筛选和记忆相结合的粒子群算法*

针对粒子群算法优化高维复杂问题出现局部最优的缺陷,提出初始粒子筛选和最差粒子记忆相结合的粒子群算法。利用熵度量粒子分量分布的均匀性,只有各分量满足均匀性要求时,该粒子才被筛选为初始粒子,以控制粒子在解空间的分布。在速度更新过程中引入最差粒子,避免粒子重复搜索曾经找到的最差位置,以提高算法的搜索效率。根据粒子寻优的成功率动态调整权重,以有效平衡深度和广度搜索能力。用本文算法优化6个经典测试函数,与3种改进的PSO算法相比,本文提出的算法不仅可以平衡局部和全局的搜索能力,还可以提高算法的搜索效率和精度。

Swarm optimization algorithm with particle selection and memory

PSO is prone to premature convergence and is difficult to balance capability of searching globally and locally. If the particles were well distributed in solution space, the search time can be reduced remarkably. Entropy is used to measure the diversity of particles in order to control the distribution. The particles which meet the need of the diverstiy are received as initial particles. To avoid searching these poor location repeatedly and increase the search efficiency of algorithm, the worst particle was introduced during velocity updating . In order to balance the global and local search ability, inertia weight was dynamicly adjusted according to success of particles evolution . The lower the success radio is , the lower the inertia weight is. To verify the validity of this algorithm, six classical test functions are optimized by the improved algorithm proposed in this paper. The result shows that the proposed algorithm can not only balance the global and local search ability,but also improve the search efficiency and accuracy of the algorithm.