基于BP网络的一种改进算法及仿真

An improve gradient algorithm of BP neural network

针对BP网络存在收敛速度慢及目标函数容易陷入局部极小值的缺点,本文研究分析了BP神经网络模型的缺陷并提出了优化策略。最后提出了一种提高BP网络学习速度的新方法,即自适应调整学习率和下降梯度。仿真结果表明,该方法大大地提高了收敛速度,而且算法简单、易行。

We analyze the advantages and disadvantages of several popular training algorithms of BP Neural Network.To deal with the defects of slow convergence rate and easily immerging in local minimum,we propose the improved gradient algorithm.The experiment shows that the algorithm can improve the conver-gence rate and is easy to use.