基于多尺度分析与神经网络的交通流预测

针对实际交通系统时变复杂和变化的不确定性所带来的交通流量随机因素影响大、非线性强、规律性不明显的特征;采用小波多尺度分解的方法,将含有综合信息的时间序列分解为多个分量特征不同的时间序列,然后采用神经网络对各个分量分别进行预测,最后用实测数据进行了验证分析。结果表明,基于多尺度分析与神经网络预测模型比单神经网络预测模型预测精度高,可用于交通流的实时动态预测。

Traffic Flow Prediction Based on Multi-scale Analysis and Neural Network

Aiming at characteristics of nonlinearity,strong interference and no obvious regularity of traffic flow caused by the complexity and uncertainty of time variance in current traffic system,a new approach is proposed for traffic flow prediction.Firstly,the traffic flow sequence made of different frequencies is decomposed into low and high frequencies in the multi-resolution analysis and the trend components are restored according to the reconstructing principle of wavelet coefficients.Secondly,the artificial neural network is used to forecast these coefficients respectively.Finally,the real detecting traffic data are used to testify the precision of the model.The results show that the forecast model based on multi-scale analysis and neural network has higher accuracy than the traditional artificial neural network model does,and this new model can be used in the dynamic forecast of traffic flow in real time.