在模式识别领域中,如何实现更高精度的分类一直是个核心问题。本文提出了将自适应RBF神经网络与小生境遗传算法相结合的方法,其中自适应RBF神经网络通过对样本判断,自动实现对RBF网络添加新的隐层节点或者将样本归于已存在的隐层节点所属的类;小生境遗传算法用于寻找最优的网络宽度值。两者相结合最后确定一个隐层节点数与类别数相同的俭省的网络。用歼击机故障数据进行仿真,比较结果表明此方法能实现更高精度的故障认定。
How to obtain a more accurate class separability is a key question in the field of classification application. An adaptive radial basis function(ARBFN) neural network is combined with the niche genetic algorithm(NGA). The ARBFN is used to add new hidden layer neurons or to determine the certain class and the input vector belongs to the class. The niche GA is used to search for the best value for the parameter of RBFN by estimating the input vectors. The method can select a parsimonious network architecture. Compared with other methods, the result shows that the method can achieve fault diagnosis with more high accuracy.