Urban Short-term Traffic Flow Prediction Based On Deep Learning And Quening Network

The intersection is the bottleneck of the urban traffic network.Improving the operational efficiency of the intersection(especially under traffic congestion conditions)is a realistic demand for traffic management and a research hotspot for intelligent transportation systems.Mastering the intersections' traffic demand of future through short-term traffic flow prediction is the basis for effective management and control of urban intersections.The traditional short-term traffic flow prediction model has less prediction accuracy due to insufficient data volume and less consideration of spatial correlation to traffic flow.The development of deep learning provides a new solution to the short-term demand prediction problem.The long short-term memory(LSTM)model has strong learning ability for long-term sequence data.And the short-term traffic flow prediction result based on LSTM has high precision.It has a good application prospect.However,this model is highly dependent on historical data and does not have consideration in the spatial correlation of short-term traffic flow,so it has some limitations.The queuing network model can simulate the process of vehicles passing through the intersection,which revealing the law of operation when vehicles passing through intersections.It can effectively describe the spatial correlation of short-term traffic flow.Therefore,in order to consider the temporal and spatial correlation of the short-term traffic flow prediction model.Based on the existing research results,this paper introduces LSTM in deep learning to consider the temporal correlation of traffic flow.Considering the spatial correlation of short-term traffic flow,the queuing network model is Introduced.By combining the long short-term memory network model and the queuing network model,the accuracy of the prediction results can be effectively improved.This paper firstly obtains the traffic flow data of some intersection entrance lanes by the intelligent transportation system of Kunshan City,Jiangsu Province,and preprocesses the data.Secondly,based on the preprocessed traffic data,the LSTMbased urban intersection entrance lane is constructed.The short-term traffic flow prediction model is compared with other short-term traffic flow prediction models to verify the efficiency of the model.After that,a comprehensive offline prediction model based on LSTM model and queuing network model is built.Finally,a case study was carried out with short-term traffic flow on the west entrance of Heilongjiang Road,Qianjin East Road,Kunshan City,Jiangsu Province.The dynamic weights were determined by MAPE to obtain the final prediction results.The example analysis shows that the prediction accuracy of the combined prediction model is 94.65%,which is higher than LSTM prediction model(prediction accuracy is 90.66%)and queuing network prediction model(prediction accuracy is 85.55%).The MAPE between the predicted value and the actual value is 5.35%.The result shows that the prediction accuracy of the combined prediction model is effectively improved.