BP神经网络的改进算法及其应用

BP（Back-propagation neural network）神经网络是目前应用最为广泛和成功的多层前馈神经网络之一，分析了BP算法的基本原理，指出了BP算法具有收敛速度慢、易陷入局部极小点等缺陷以及这些缺陷产生的根源，并针对这些缺陷，通过在标准BP算法中引入变步长法、加动量项法等几种方法来优化BP算法。仿真实验结果表明，这些方法有效地提高了BP算法的收敛速度，避免陷入局部最小点。同时．将改进得BP神经网络算法应用于脱机手写体汉字识别系统的实现。使系统较好地回避了汉字结构复杂、变形难以预测等问题，提高了识剐率。

The Improves on the Standard BP Algorithm and their Use

Back propagation neural network is one of the extensively apphed muhilayer feedforward neural network in artificial neural network. Basic principle of BP algorithm is analyzed firstly, then some defects such as slow convergence rate and getting into local nfinimum in BP Algorithm are pointed out, and the root of the defects is presented. Finally, in view of these limitations, several methods such as genetic algorithm and simulated annealing algorithm etc. are led to optimize BP algorithm.Simulation Experiment results show that these methods increase efficiently the convergence performance of BP algorithm and avoid local minimum. The Chinese character recognition system is realized as a result of the use of improved BP networks.Experimental results show that the system based on the improved BP algorithms can successfully overcome obstacles out of the complexity of the Chinese characters＇ structures,and of difficulties to foresee written forms＇ changes.