基于自适应学习速率的改进BP神经网络

摘要

As a nonlinear systems of identification tool in pattern recognition,system-control and other areas,BP neural network has got widely used,but there is still some flaws in the BP,that it can fell into the local minimum during the training process.This article introduces the adjustment of neurons conversion functions,and the learning rate self-adjustment to overcome network training is so slow and not easily convergence to a global optimization.According to the simulation experiment,the improved BP network method is proved.%BP神经网络自被提出以来,便作为非线性系统的辨识工具在模式识别、系统控制等多领域得到广泛应用。但BP算法仍存在缺陷,在学习时容易陷入局部极小。本文采用调整神经元转换函数的方法,并采用学习速率自适应调整来克服网络训练速度慢、不易收敛到全局最优等缺点。通过仿真实验表明,改进后的方法可行。