一种用于RBF神经网络的支持向量机与BP的混合学习算法

A Hybrid Learning Algorithm for RBF Neural Networks Based on Support Vector Machines and BP Algorithms

基于支持向量机与径向基(RBF)神经网络在结构上的相似性,提出了一种用于RBF网络的支持向量机与BP的混合学习算法.算法分为2步:首先采用序贯最小优化算法学习训练支持向量机,得到RBF网络较优的初始结构和参数;随后由BP算法调整优化RBF网络参数.混合学习算法结合了支持向量机小样本学习、学习训练快捷以及BP算法在线修改网络参数的特点.仿真研究表明,混合学习算法学习效率高,网络性能优良,应用于函数逼近时效果优良.

<Abstrcat>Support vector machine (SVM) resembles RBF neural networks (RBFNN) in structure. Considering their resemblance, a new hybrid learning algorithm for RBFNN was proposed. The proposed learning algorithm is based on SVM and BP algorithms and includes two steps: the first step is SVM learning using sequential minimum optimization, and this will obtain a good initial structure and parameters of RBFNN; in the second step, BP algorithms is applied to optimize RBFNN parameters. This hybrid learning algorithm has a number of advantages: its training process is fast and efficient, and it can optimize parameters online. Examples are simulated to demonstrate the superiority and performance of the proposed hybrid learning algorithm.