基于改进BP网络模型的公路流量预测

针对公路客货运量预测的问题，对现有的常用预测方法进行研究，提出改进BP神经网络预测模型。该模型首先采用动态陡度因子改变激励函数的陡峭程度，改善激励函数的响应特征，得到更好的非线性表达能力；其次利用附加动量因子，通过将以前的经验进行积累，降低了神经网络对误差曲面的局部细节敏感性，较好地遏制网络陷于局部最小；再次采取变学习率学习算法，先给一个较大初值，随着学习过程的进行，学习率不断减小，网络趋于稳定。改进BP算法既可以找到更优解，又可以缩短训练时间。结合某地区的公路运量相关数据，对改进BP神经网络预测模型进行了验证。实验结果表明，该模型的相对误差和迭代次数都取得了较大的改善，对公路客货运量预测很有效。

Forecasting Highway Flow Based on Improved BP Neural Network Model

Aiming at the forecast problem of the highway flow, basbxi on the research of common forecasting method, an improving BP neural network forecasting model was put forward. Firstly,this model introduces steepness factor to dynamically change the steepness of the activation function and improve the response characteristics of the activation function to get better ability to express non-linear;Sec- ondly, it uses the method of momentum item addition to accumulate experience of previous , reduce the sensitivity of local details of the network for error surfaces and effectively trapped in local minimum; Thirdly, it adopts the learning algorithm of variable learning rate, a larger initial value was given at beginning, with the learning process progresses, the learning rate decreasing, the network is stabilized. The improved BP algorithm can find better solutions,but also can shorten the training time. With some traffic-related data,the improved BP neural network prediction model is validated. Experimental results show that the relative error and the number of iterations of the model have made great improvements. It is very effective for the forecast of highway flow.