Traffic Flow Pattern Mining And Its Application Based On Spatiotemporal Features

According to the data of China’s National Bureau of statistics,the ownership of civil cars and private cars in China continues to grow rapidly every year.The speed of highway construction is far from keeping up with the growth rate of car ownership.More and more provinces or cities begin to explore the temporal and spatial correlation and distribution characteristics of traffic flow in the road network with the help of intelligent transportation system,so as to promote the construction and improvement of Expressway infrastructure.However,under the joint influence of space-time conditions such as settlement distribution,holidays and meteorological factors,the traffic flow in the expressway network presents a variety of flow change patterns.Studying these change patterns is of great significance for infrastructure optimization and business decision-making.For example,these change models can be used to predict the hot toll stations with high traffic flow in the road network in the future,and give priority to the construction and improvement of these stations to improve the vehicle passing efficiency in peak periods.Traditional work emphasizes the similarity of adjacent space,while the traffic flow of Expressway stations has weak correlation with adjacent space,and stations with large geographical differences may have similar patterns.This brings difficulties in traffic flow modeling and prediction.At the same time,the traditional traffic flow prediction methods do not consider the similarity of time series,which limits the prediction accuracy of station traffic flow and hotspot discovery.To solve the above problems,the following research work is carried out in this thesis:(1)Highway traffic flow pattern mining method based on temporal and spatial characteristics.In view of the weak correlation between the nearest neighbors and the traffic flow of Expressway stations,the stations with large geographical differences may have similar patterns,and the description of various spatio-temporal change patterns of station traffic flow is insufficient.In this thesis,the traffic flow time series of toll stations in the whole network are divided into multiple clusters through cluster analysis,so that the traffic flow time series belonging to the same cluster are as similar as possible,and the traffic flow differences between different clusters are as large as possible.At this time,the traffic flow time series clustered into the same cluster can be regarded as belonging to the same traffic flow pattern.(2)Expressway hot spot toll station discovery based on traffic flow model.The traditional methods do not consider the similarity of time series,which limits the prediction accuracy of station traffic flow and hot spot discovery.Based on the traffic flow pattern mining,this thesis takes the pattern of traffic flow as the feature,and introduces the influencing factors such as holidays,weekends and extreme weather,designs and implements an integrated learning model to predict the traffic flow of the whole website of Expressway in the next 30 days.Divide the top 10% of the predicted traffic into hot charging stations.Experiments show that compared with KNN,SVR and other common baseline models,this method has a good effect on improving the prediction accuracy,and can meet the business needs of hot spot toll station discovery.(3)The prototype system is designed and implemented in the actual project scenario.The system can help relevant departments analyze and evaluate the spatial-temporal correlation and distribution characteristics of the existing expressway traffic flow,mark the stations under each mode on the map,facilitate the discovery of the spatial distribution of stations in the same mode,and more intuitively display the prediction results of hot charging stations in the map.