摘要
结合改进的免疫算法和最小二乘法,提出了一种设计径向基函数(RBF)网络的两级学习方法.该方法利用免疫算法确定RBF网络隐层的非线性参数,能够有效克服进化算法的未成熟收敛现象.改进的免疫算法针对RBF网络的特点,采用基于矢量距离的亲和度计算方法,克服了原有基于信息熵计算方法存在的计算复杂、参数难于确定的缺陷.将这种方法设计的RBF网络用于Mackey-Glass混沌序列预测的仿真实验证明了该方法的有效性.
A two-level learning method combining improved immune algorithm and least square method was proposed to design a radial basis function (RBF) network. In this method, the nonlinear parameters of RBF hidden layer are determined by an immune algorithm, which can effectively overcome the immature problem in the evolutionary algorithm. According to the characteristic of RBF network, an affinity computation based on vector distance is used in this improved immune algorithm, which overcomes the flaw of the original entropy-based computation method, such as the problems in computation complexity and parameter determination. The application of the RBF network in Mackey-Glass time series prediction problem demonstrates the effectiveness of the proposed training algorithm.
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
2004年第5期768-770,774,共4页
Journal of Shanghai Jiaotong University
关键词
径向基函数网络
免疫算法
最小二乘法
radial basis function (RBF) network
immune algorithm
least square method
