Research On Short-term Traffic Flow Forecast Based On Spatio-temporal Correlation

The rapid development of the city has driven the rapid increase in the number of cars,which has caused the problem of urban traffic congestion to become more and more prominent.In order to alleviate traffic congestion and improve road utilization efficiency,scholars in the field of transportation have proposed Intelligent Transportation Systems(ITS).The two major functions of the intelligent transportation system are traffic control and traffic guidance.Traffic flow prediction can provide an important basis for traffic guidance and is the core content of the intelligent transportation system.This paper analyzes the spatiotemporal characteristics of short-term traffic flow,and uses single-observation point multi-lane correlation and multi-observation point spatio-temporal correlation to predict short-term traffic flow.The work content and innovation results of this article are as follows:Firstly,Aiming at the relevance of multi-lane traffic flow,a short-term traffic flow prediction method based on multi-lane weighted fusion is proposed.This method analyzes the correlation between the aggregated traffic flow of a single inspection point on the California highway and the traffic flow of each lane,and found that the traffic flow of each lane has a greater correlation with the aggregated traffic flow.Then construct a two-way long and short-term memory model(Bi LSTM),select each lane traffic flow and aggregated traffic flow as training data,respectively input into the Bi LSTM model to predict each lane traffic flow and aggregated traffic flow,and input for predicting each lane traffic flow and aggregated traffic flow Ridge regression method performs weighted fusion as the final predicted traffic flow.The experimental results show that the prediction accuracy of aggregate traffic flow considering the correlation of multi-lane traffic flow is better than the traditional aggregate traffic flow prediction accuracy,and has certain application value.Secondly,Based on the dynamic spatio-temporal correlation between the traffic flow of each section of the traffic network,a graph convolution gated recurrent network prediction method based on the attention mechanism is proposed.This method aims at the shortcomings of the graph convolutional gated recurrent network model that cannot pay attention to the spatiotemporal features most relevant to the predicted target.The attention residual fusion method is introduced when the spatiotemporal feature is output,thereby enhancing thelearning performance of the graph convolutional gated recurrent network model.The graph convolutional gated recurrent network model is used to learn the temporal and spatial dependence of the traffic flow of each section of the traffic network,and the attention mechanism is integrated to calculate the influence weight of each unit before the model output.The temporal and spatial characteristics of each unit are multiplied by the influence weight to obtain attention The spatio-temporal characteristics of the nature,and finally use the spatio-temporal characteristics of each unit and the spatio-temporal characteristics of each unit with attention properties to carry out residual fusion to obtain the final prediction result.Experiments show that this method has higher traffic flow prediction accuracy than the traditional model before the improvement.It can not only capture the temporal and spatial dependence of traffic flow in the traffic network,but also pay attention to the temporal and spatial features that have the greatest impact on the predicted target when each unit is output.