A hybrid particle swarm optimization algorithm using adaptive learning strategy
A hybrid particle swarm optimization algorithm using adaptive learning strategy
复制标题
DOI：
10.1016/j.ins.2018.01.027
复制
发表时间：
2018-04
期刊：
影响因子：
--
通讯作者：
Feng Wang;Heng Zhang;Kangshun Li;Zhiyi Lin;Jun Yang;Xiao-Liang Shen
中科院分区：
文献类型：
--
作者：
Feng Wang;Heng Zhang;Kangshun Li;Zhiyi Lin;Jun Yang;Xiao-Liang Shen
现实中的许多优化问题变得越来越复杂，这推动了对不同优化算法的改进研究。粒子群优化算法(PSO)已被证明是解决各种优化问题的有效工具。然而，对于基本的粒子群算法，其更新策略主要是为了学习全局最优，在许多复杂的优化问题上，特别是对于多峰问题，往往会出现早熟收敛和性能较差的问题。提出了一种采用自适应学习策略(ALPSO)的混合PSO算法。在ALPSO算法中，我们采用了基于自学习的候选生成策略来保证算法的探索能力，采用基于竞争学习的预测策略来保证算法的可拓展性。为了更好地平衡勘探能力和开发能力，设计了一种基于容差的搜索方向调整机制。在40个基准测试函数上的实验结果表明，与五种典型的PSO算法相比，ALPSO在更多的情况下无论在收敛精度还是收敛速度上都明显优于其他算法。
Many optimization problems in reality have become more and more complex, which promote the research on the improvement of different optimization algorithms. The particle swarm optimization (PSO) algorithm has been proved to be an effective tool to solve various kinds of optimization problems. However, for the basic PSO, the updating strategy is mainly aims to learn the global best, and it often suffers premature convergence as well as performs poorly on many complex optimization problems, especially for multimodal problems. A hybrid PSO algorithm which employs an adaptive learning strategy (ALPSO) is developed in this paper. In ALPSO, we employ a self-learning based candidate generation strategy to ensure the exploration ability, and a competitive learning based prediction strategy to guarantee exploitation of the algorithm. To balance the exploration ability and the exploitation ability well, we design a tolerance based search direction adjustment mechanism. The experimental results on 40 benchmark test functions demonstrate that, compared with five representative PSO algorithms, ALPSO performs much better than the others in more cases, on both convergence accuracy and convergence speed.