摘要
针对径向基函数神经网络参数难以设置以及因此而导致的网络隐层结构不明朗的问题,提出了一种应用控制种群多样性的微粒群(ARPSO)优化径向基函数神经网络(RBF)的方法。通过引入"吸引"和"扩散"因子对基本微粒群算法进行改进,并将改进的微粒群算法用于RBF聚类半径的优化,进而能够合理地确定RBF的隐层结构。将用ARPSO优化的RBF神经网络应用于非线性函数逼近,经实验仿真验证,与基本微粒群(PSO)算法、收缩因子微粒群(CFA PSO)算法优化的RBF神经网络相比较,在收敛速度和识别精度上有了显著的提高。
Aiming at the problems that parameters of radial basis function neural network are difficult to be set up and thus lead to network hidden layer structural uncertain,a novel radial basis function neural network method based on a diversity-guided particle swarm is pro-posed. By introducing the ＂attract＂ and ＂proliferation＂ factor,the basic particle swarm algorithm is improved. The RBF hidden layer structure can be reasonably determined by using the improved particle swarm optimization for clustering radius. The new training algo-rithm is used to approximate polynominal function,compared with PSO,and CFA PSO,the algorithm improves the velocity of conver-gence and recognition accuracy.
出处
《计算机技术与发展》
2014年第11期43-46,共4页
Computer Technology and Development
基金
山西省研究生教改课题资助项目(201002034)
关键词
微粒群算法
吸引
扩散
RBF神经网络
最近邻聚类方法
Particle Swarm Optimization （PSO）
attractive
repulsive
radial basis function neural network
nearest neighbor cluster algo-rithm
