Traffic Flow Prediction Method Based on Fast Statistics of Traffic Flow and Graph Convolutional Network

Traffic flow prediction is an important issue in the intelligent transportation system, and traffic flow statistics is the data source of traffic flow prediction. The commonly used target detection and tracking traffic flow statistics methods are vulnerable to the impact of light, background, etc. in the actual scene, resulting in the problem of target loss and inaccurate tracking results. This article proposes a fast statistical method for vehicle flow, which uses trigger sensor devices in the road network to obtain vehicle images. In order to solve the problem of sensor counting errors caused by fast vehicle speeds and reduce the amount of data processed subsequently, the obtained vehicle images are first classified into target categories, with a target number greater than one and a target number equal to one. Then, target detection is performed on images with a target number greater than one, Finally, use statistical formulas for traffic flow statistics. Experiments have shown that using this method can reduce detection data by 30.8%, greatly improving detection efficiency. On this basis, this paper further uses the graph convolutional network(GCN) combined with the moving horizon to predict the traffic flow, analyzes the relationship between the topological structure of the traffic network and the time series data, effectively captures the time-space characteristics and evolution mode of the traffic flow, and establishes the latest characteristic mode of the traffic flow. This article uses a total of 16992 data points from the public datasets PeMS04 from January to February 2018 for training and testing. The experimental results showed that the root mean square error(RMSE) and mean absolute error(MAE) of the graph convolutional network traffic flow prediction model combined with the moving horizon decreased by 2.1 % on average and 2.3% on average.