具备预判能力和向最小距离学习的粒子群算法

Particle Swarm Optimization Algorithm with Predictive Ability and Learning from Minimum Distance

针对粒子群算法求解问题时易早熟收敛,精度不高等问题,提出了具备预判能力和向最小距离学习的粒子群算法(MDPSO)。该算法中粒子的社会学习部分是从各个粒子的个体极值中提取有效信息进行学习,首先粒子对所有粒子的个体极值(包括自己)的信息进行分析,确保下一次寻优过程中向正确的方向飞行,即预判能力,防止了粒子向错误方向飞行而浪费太多时间;其次,粒子从预判的方向上选取一个最小距离来指导粒子社会部分的学习,使粒子较快的收敛到下一代较好食物的位置。最后,结合两策略的特点,可以有效的防止算法早熟收敛并提高其精度。MDPSO算法在CEC2017年版基准测试函数上的实验结果显示出该算法相比于其他的PSO算法的优势更为显著。

In order to solve the problem of particle swarm optimization,it is easy to converge prematurely and the accuracy is not high.A particle swarm optimization algorithm(MDPSO)with predictive ability and learning from the minimum distance is proposed.The social learning part of the particle in the algorithm is to extract effective information from the individual extremum of each particle for learning.First,the particle analyzes the individual extremum(including itself)information of all particles to ensure that the next op⁃timization process is correct.Fly in the direction of the particle,the predictive ability,prevents the particle from flying in the wrong di⁃rection and wasting too much time;secondly,the particle selects a minimum distance from the predictive direction to guide the learning of the social part of the particle,so that the particle quickly converges to Good food location for the next generation.Finally,combining the characteristics of the two strategies can effectively prevent the algorithm from premature convergence and improve its accuracy.The experimental results of the MDPSO algorithm on the CEC2017 benchmark test function show that the algorithm has more significant ad⁃vantages compared to other PSO algorithms.