Short-term Traffic Flow Prediction Based On Spatio-tempral And Semantic Correlation

With the increasingly serious urban traffic problems,intelligent transportation system becomes more and more important.Intelligent transportation system can effectively alleviate the problem of traffic congestion,improve the efficiency of road network,and also provide information services for people’s transportation.Short-term traffic flow prediction is one of the key problems in intelligent transportation system.It is the basis for analyzing the traffic conditions(speed,flow and density)of urban road network,mining traffic patterns and predicting the future traffic conditions of road network.However,due to the complex spatial,temporal and semantic correlation of traffic flow,short-term traffic flow prediction is a challenging task.In order to solve the above challenges,this paper proposes a new short-term traffic flow prediction scheme based on graph convolutional network,cyclic neural network extension,and deep learning techniques such as attention mechanism.The main research work of this paper is as follows:1.The original traffic flow data often have data missing or abnormal problems,and the pretreatment of original data directly affects the accuracy of traffic flow prediction.This paper studies a traffic flow data repair method based on two-way RNN.Taking into account the timeseries characteristics of traffic flow data and the influence of traffic data before and after missing data,the paper introduces a two-way structure recurrent neural network to repair the missing data of traffic flow,so as to realize the time series interpolation of abnormal traffic data,And compared with the traditional data repair method based on statistical correlation,it shows that this method can repair missing data more effectively.2.Considering the complex spatial and temporal dependence of short-term traffic flow prediction,In this paper,a Dense Net based Graph Convolution Long and Short-term Memory Network(DGLN)for traffic flow prediction is studied.The undirected graph method is used to represent the urban traffic network,and a short-term module composed of multiple space-time blocks is constructed.The spatial correlation and temporal sequence of traffic data were dynamically obtained by using the graph convolution network based on attention mechanism and the short and long time memory network,and the features were transferred between the spatiotemporal blocks in the way of dense connection.A cycle module based on long-term and short-term memory network and periodic attention mechanism is constructed to mine the periodic characteristics of traffic data.The fusion method of near-term module and periodic module is studied,and the traffic flow prediction method based on graph convolution long and short time memory network based on dense connection is proposed to better reflect the time correlation and space dependence of traffic flow.3.In addition to spatiotemporal correlation,there is semantic correlation between roads in traffic flow data.For example,roads around business districts often have similar traffic patterns.Based on the DGLN model,this paper studies a traffic flow prediction method based on Semantic Based Graph Convolution Long and Short-Term Memory Network(SGLN).Considering the historical relevance and functional similarity,road connectivity graph is used to represent the spatial and semantic relevance of traffic flow,so as to comprehensively consider the connectivity between roads that are not directly connected;The crawler technology is used to collect the traffic information from the website of Guangzhou Transportation Bureau,and experiments are carried out on the collected data set.The results show that the traffic flow prediction method considering semantic association in this chapter has better prediction effect.