BP神经网络算法的改进及收敛性分析

研究BP神经网络的数学理论,详细分析几种流行的BP神经网络学习算法的优缺点.针对一般BP算法收敛速度慢,易陷入局部极小值的缺陷,受Fletcher-Reeves线性搜索方法的指引,提出基于改进共轭梯度法的BP算法.从理论方面对算法进行深入的分析,介绍算法的详细思路和具体过程.并将算法训练后的BP神经网络运用到函数逼近中去.仿真结果表明,这种改进方案确实能够改善算法在训练过程中的收敛特性,而且提高收敛速度,取得令人满意的逼近效果.

BP Neural Network Algorithm Improvements and Convergence Analysis

This paper studies the theory of BP neural network, analyzes the advantages and disadvantages of several popular training algorithms. To deal with the defects of the steepest descent in slowly converging and easily immerging in partial minimum frequently, after analyzing the linear hunting method developed by Fletcher and Reeves, the improved conjugate gradient algorithm is brought forward to solve the problem. This paper analyzes the algorithm deeply in theory, introduces the idea and process. Then the BP neural network trained by this algorithm is applied into function approximation. The results show that this algorithm improves the convergence of training process and achieves excellent identification effect.