基于改进 BP 神经网络的铁路货运量预测

在分析铁路货运量预测方法的基础上，针对标准BP神经网络的不足，提出改进的BP神经网络预测模型。首先，利用动态陡度因子来改变激励函数的陡峭程度，以此来得到更好的激励函数响应特征以及更好的非线性表达能力；其次，利用附加动量因子，通过对以前经验的积累，既降低了神经网络对误差曲面的局部细节敏感特性，又较好的遏制了神经网络易于限于局部最小的缺陷；最后，采取改变学习率的方法，给定一个较大的学习率初始值，在学习的过程中学习率不断减小，网络最终趋于稳定。改进BP算法既可以得到更优的解，还能够缩短训练时间。利用全国铁路货运量的相关数据对改进BP神经网络进行了验证。验证的结果表明，改进的BP神经网络预测模型在相对误差和迭代次数上有较大改善，对铁路的货运量预测很有效。

Forecasting Railway Freight Volume Based on Improved BP Neural Network Model

An improved BP neural network is proposed for the shortage of standard BP neural network based on the analysis of railway freight volume forecasting methods .Firstly, this model introduces dynamic steepness factor to change the steepness of the activation function and obtain better response characteristics of activation function and better ability of nonlinear expression .Secondly , it uses additional momentum factor to accumulate previous experience ,reduce network on the error surface detail sensitivity characteristics , and effectively trapped in local minimum;Thirdly,it uses learn algorithm of changing the learning rate , given a large initial learning rate value , learning rate decreases in the learning process , and the network tends to be stable network .The improved BP algorithm can get better solution ,and can also shorten the training time .The improved BP neural network is verified by using the relevant data of railway freight volume .The results show that the improved BP neural net-work prediction model has greatly improved the relative error and the number of iterations , and it is very effective for the forecasting of railway freight volume .