一种可在线学习的变结构径向基函数网络及其在被动声纳目标识别中的应用

An Novel On Line Learning Structure Variable Radial Basis Function Nets with Application on Passive Sonar Target Classification

提出了一种新颖的隐节点可调的变结构径向基函数网络，并应用进化规划最优地确定和调节变结构径向基函数网络隐层节点的数目及其核函数的中心和宽度，从而使网络具有在线学习和记忆新的目标模式的功能．并将该网络应用于被动声纳目标的识别和在线学习，实验结果表明基于进化规划的变结构径向基函数网络不仅改善了网络的泛化能力。

In this paper,a novel structure variable radial basis function networks (SVRBF networks) is proposed,whose hidden layer nodes can be modified on line,and Evolutionary Computation (EC) is used to optimally determine and modify the total number of hidden layer nodes and their core function's center and width of the SVRBF networks.The SVRBF networks are then used for passive sonar target classification and learning on line,and the result of experiment shows that the EC based SVRBF networks have better generalization performance than k means based RBF nets,and are effective in solving the problem of forgetting the old patterns in on line learning which exists in passive sonar target recognition by using conventional neural networks.