融合多特征神经网络的城市道路交通速度预测

Urban Road Traffic Speed Prediction Based on Fusion of Multiple FeatureNeural Networks

针对城市道路交通速度特性分析不全面的问题,本研究考虑到气象因素和空气污染因素对交通速度的影响提出了一种多源特征融合结合Informer模型的道路交通速度预测模型(MFInformer,Multi Feature Informer)。采用成都市浮动车数据及气象数据、空气污染数据,通过提取数据集的交通流特征,天气特征以及空气污染特征对道路交通速度进行预测。研究表明:在以10分钟为间隔的预测中,基于浮动车数据结合外源因素的多特征数据的Informer模型相较于仅使用浮动车数据的Informer模型以及常用的时间序列预测模型循环神经网络和长短期记忆网络来说预测的准确度有着显著的改善,结合多源特征的Informer模型预测交通速度平均绝对误差,均方误差和平均绝对百分比误差分别为1.38、5.32%、1.74,均优于其它模型。

In response to the issue of incomplete analysis of urban road traffic speed characteristics,this study proposes a multi-source feature fusion combined with Informer model for road traffic speed prediction(MF Informer,Multi Feature Informer)considering the impact of meteorological and air pollution factors on traffic speed.Using floating vehicle data,meteorological data,and air pollution data from Chengdu,road traffic speed is predicted by extracting traffic flow characteristics,weather characteristics,and air pollution characteristics from the dataset.The research shows that in the 10 minute interval prediction,the Informer model based on the multi feature data of floating car data combined with external factors has significantly improved the accuracy of prediction compared with the Informer model using only floating car data and the commonly used time series prediction models Recurrent neural network and Long short-term memory network.The Informer model combined with multi-source characteristics predicts the Mean absolute error of traffic speed,The Mean squared error and the average absolute percentage error are 1.38,5.32%and 1.74 respectively,which are superior to other models.