径向基函数神经网络指导的粒子群优化算法求解多峰优化问题  

Radical Basis Function Neural Network Guided Particle Swarm Optimization Algorithm for Solving Multimodal Optimization Problems

作　　者：张潇 宋威 ZHANG Xiao;SONG Wei(School of Artificial Intelligence and Computer Science,Jiangnan University,Wuxi 214122,China;Jiangsu Provincial Engineering Laboratory of Patern Recognition and Computational Intlligence,jiangnan University,Wuxi 214122,China)

机构地区：[1]江南大学人工智能与计算机学院,江苏无锡214122 [2]江南大学江苏省模式识别与计算智能工程实验室,江苏无锡214122

出　　处：《小型微型计算机系统》2023年第11期2529-2537,共9页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(62076110,61673193)资助;江苏省自然科学基金项目(BK20181341)资助。

摘　　要：面对多峰优化问题粒子群优化算法因多样性不足和搜索动作选取不合理,难以找到问题的全局最优解.为此本文提出一种径向基函数神经网络指导的粒子群优化算法.首先设计子群划分方法,将种群划分成多个子群,子群中心作为子群粒子的学习目标,指导其搜索.该方法充分考虑种群多样性,选择能代表子群搜索特性的粒子作为子群中心,并使之远离存在的中心,通过选择合适的子群中心,实现子群划分.不同子群粒子在各自子群中心指导下搜索,呈现多样的搜索特性.其次,利用子群中心设置隐藏层节点,并在输出层输出粒子加速系数的调整动作.最后引入强化学习来训练网络.在CEC2013的15个多峰函数上开展实验,结果表明本文方法明显提高了多峰优化问题的求解精度.When facing multimodal optimization problems,particle swarm optimization algorithm is difficult to find the global optima.Because it lacks sufficient swarm diversity and cannot select appropriate search actions.Therefore,this paper proposes a radical basis function neural network guided particle swarm optimization algorithm.Firstly,a subswarm division method is designed to divide the whole swarm into multiple subswarms,and each subswarm center is selected as the learning target of particles in the subswarmto guide their search.The designed method fully considers the swarm diversity,which selects the particle reflecting the search characteristic of the subswarm as its center,and then makes the center distant from the existing centers.By selecting the appropriate subswarm centers,the whole swarm is divided into multiple subswarms.Particles of different subswarms search under the guidance of their respective subswarm centers,showing the diverse search characteristics.Secondly,the hidden layer node is set according to the subswarm center,and the output layer outputs the action for each particle which guides the adjustment ofthe acceleration coefficient.Finally,reinforcement learning is introduced to train the network.The experimental results on 15 multimodal functions of CEC2013 test suite show that the proposed algorithm can significantly improve the solving accuracy of multimodal problems.

关 键 词：径向基函数神经网络 粒子群优化算法 学习目标 加速系数 多峰优化 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]