摘要
OL S训练方法应用在径向基 (RBF )神经网络里时 ,存在当训练数据量很大时速度很慢的问题 ,并且 OL S方法不能自动确定基函数的平滑参数。本文针对此问题提出了一种基于快速模糊 C-均值算法 (A FCM)与 OL S算法相结合的 AF OL S训练算法 ,该算法使用 AF CM方法对数据进行聚类 ,并获取基函数的平滑参数 ,然后使用 OL S方法从聚类结果中选取网络中心。利用实测的 4类飞机目标数据对其进行性能检验 ,试验结果验证了该训练算法可提高网络的训练速度 ,缩小网络规模 ,提高网络的分类能力。
In this paper, a novel training algorithm called AFOLS for RBFNN is presented based on accelerated fuzzy Cmeans and orthogonal least squares(OLS). Investigations reveal that this algorithm possess desired property. It quickens the learning process and reduces the size of network. The spreads of centers are autodefined according to the training data. Experiments for recognition of four kinds of aircraft targets are carried out. Results show that the AFOLS algorithm achieves faster speed and higher resolvingpower than the OLS algorithm does.
出处
《现代电子技术》
2005年第3期18-20,共3页
Modern Electronics Technique
关键词
正交最小二乘算法
快速模糊C-均值算法
径向基神经网络
AFOLS算法
orthogonal least squares(OLS)
accelerated fuzzy Cmeans(AFCM)
radial basis function neural network(RBFNN)
AFOLS algorithm
