一种前向神经网络快速学习算法及其在系统辨识中的应用

A FAST LEARNING ALGORITHM OF FEEDFORWARD NEURAL NETWORKS AND ITS APPLICATION TO SYSTEM INDENTIFICATION

提出一种基于最小二乘的前向神经网络快速学习算法．与现有同类算法相比，该算法无需任何矩阵求逆，计算量小，较适于需快速学习的系统辨识和其他应用．文中推导了算法，并给出一种更为简便的局部化算法．

In this paper,we propose a fast learning algorithm of feedforward networks based on the least squares.Compared with existing similar algorithms,the present algorithm does not require any matrix inversion,therefore,it has a less computational cost and can be better suited for system indentification and other areas where fast learning is required.We derive the algorithm and also give an even simpler and more convenient localized algorithm.Simulation results for system identification show the effectiveness of the algorithm.