径向基函数神经网络的一种两级学习方法

建立RBF(radial basis function)神经网络模型关键在于确定网络隐中心向量、基宽度参数和隐节点数.为设计结构简单,且具有良好泛化性能径向基网络结构,本文提出了一种RBF网络的两级学习新设计方法.该方法在下级由正则化正交最小二乘法与D-最优试验设计结合算法自动构建结构节俭的RBF网络模型;在上级通过粒子群优化算法优选结合算法中影响网络泛化性能的3个学习参数,即基宽度参数、正则化系数和D-最优代价系数的最佳参数组合.仿真实例表明了该方法的有效性.

A two-level learning hierarchy for the radial basis function networks

The key to construct a radial basis function(RBF)network is to select reasonable hidden center vectors,RBF width and hidden node number.In order to design a RBF network with parsimonious structure and good generalization,a new two-level learning hierarchy for designing RBF networks is proposed.At the lower level in this method,a parsimonious RBF model is constructed by an integrated algorithm(ROLS D-opt)which combines regularized orthogonal least squares (ROLS)with D-optimality experimental design(D-opt).At the upper level,particle swarm optimization(PSO)is used to search the optimal combination of three important learning parameters,i.e.,the RBF width,the regularized parameter and D-optimality weight parameter,which influence the network's generalization ability.Simulation results show the effectiveness of the proposed method.