一种带有加速策略和变异策略的粒子群算法

A Kind of Particle Swarm Optimization Algorithm with Acceleration Strategy and Mutation Strategy

不同于对粒子群控制参数的改进,在标准粒子群的基础上提出了双种群粒子群算法,粒子种群每次进化时都按适应值排序并分组,最好的一组粒子采用适应值最接近的两点连线上的更好点替代当前粒子位置的加速策略,以此加大算法的局部搜索能力,为了平衡算法的全局和局部搜索能力,另一组最差的粒子采用随机背驰当前全局最优粒子的速度方向进行变异策略更新,以此保证种群的全局搜索能力,该算法采用一般非线性惯性权重和固定学习因子.新算法在23个三类经典测试函数的实验中都找到了最优值,与其它算法比较,结果表明该新算法在三类问题上都有更好的性能,特别是在多模函数的优化中更为显著.

The double population particle swarm optimization is proposed based on the standard PSO,which is different from the improvement of control parameters of PSO.The particle population is sorted and grouped according to adaptive value at every time evolution.In order to increase the ability of local search algorithm,new particles displace the current position of original particles,and the new particles adopt a better point which is in a line of connecting the nearest two points.In order to balance the algorithm of the global search ability with the local search ability,the worst group of particle is update,and the worst particles use random velocity direction which contrary to the global optimal particle.This algorithm uses general nonlinear inertia weight and constant learning factor.In the experiments of 3kinds of problems which contain 23 kinds of classical test function,the new algorithm can find the optimal value of all functions.Compared with other algorithms,the result shows that the new one has better function,especially in the optimization of multimodal function.