FKCN优化的RBF神经网络

FKCN Optimization of Radial Basis Function Neural Network

ＦＫＣＮ（Ｆｕｚｚｙ Ｋｏｈｏｎｅｎ ｃｌｕｓｔｅｒ ｎｅｔｗ ｏｒｋ）将模糊隶属度的概念用于Ｋｏｈｏｎｅｎ 神经网络的学习和更新策略中，改善了Ｋｏｈｏｎｅｎ 网络的性能，是一种更为快速有效的聚类网络。作者将ＦＫＣＮ用于优化ＲＢＦ（Ｒａｄｉａｌｂａｓｉｃ ｆｕｎｃｔｉｏｎ）神经网络基函数的中心，并将优化后的ＲＢＦ网络用于曲线拟合和非线性时间序列预测，同时与基于Ｃ－ＭＥＡＮＳ的ＲＢＦ网络进行比较。实验结果表明：采用ＦＫＣＮ优化的ＲＢＦ网络具有更好的拟合和预测能力，尤其在曲线拟合实验中，ＦＫＣＮ优化的ＲＢＦ网络可以达到最小学习误差，比Ｃ－ＭＥＡＮＳ的网络小一个数量级，可见用ＦＫＣＮ优化ＲＢＦ神经网络可以较好地提高ＲＢＦ神经网络的性能。

Fuzzy Kohonen cluster network(FKCN) is a fast and effective clustering neural network. It improves Kohonen neural network by using fuzzy membership as its learning rate. This paper provides FKCN to optimize the centers of the radial basis function(RBF) neural network, then applies the network to function approximation and nonlinear time series prediction. Compared with the RBF networks obtained by C MEANS clustering, the optimized RBF network has better approximation and prediction feasibility. Especially in function approximation, the least learning error of the optimized RBF network is about 1/10 of that of the RBF networks obtained by C MEANS clustering.