一种改进的神经元网络及学习算法

A MODIFIED NEURAL NETWORK AND LEARNING ALGORITHM

本文提出一种改进的神经网络结构，它由线性网络和多层前向网络两部分组成。线性网络部分的参数采用递推最小二乘法辨识，多层前向网络的权值和阈值采用ＢＰ算法学习。由于线性网络的引入及递推最小二乘法的使用，大大提高了网络的学习速度。此外，该网络结构也为基于神经网络模型控制器的设计的简化提供了条件。

A modified neural network structure which is composed of a linear network and a multilayered feedforward neural network(MFNN) is presented.The parameters of linear network are identified by recursive least square and weights and thresholds of MFNN are learned by BP algorithm.The modified neural network′s learning speed is much increased because of the linear network and the recursive least square.In addition it is convenient to use the modified neural network for simplifying controller′s design.