Urban Road Traffic Speed Prediction Based on Fusion of Multiple Feature Neural Networks
In response to the issue of incomplete analysis of urban road traffic speed characteris-tics,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 me-teorological 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 percent-age error are 1.38,5.32％and 1.74 respectively,which are superior to other models.
intelligent transportationdeep learningtime series predictionspeed prediction
NETL
NSTL
万方数据
