Research On Link Average Speed Prediction Considering Spatio-temporal Correlation

The rapid economic development has induced more travel demands,and the problem of traffic congestion in cities has become increasingly serious.Intelligent transportation system(ITS)is considered to be one of the effective methods to alleviate congestion and improve efficiency.Traffic forecasting can provide a data basis for the intelligent traffic system to achieve the functions of traffic information issuing and guidance,help travelers plan their routes in advance,and relieve congestion to a certain extent.Taxi GPS data provides information on the dynamic changes of urban traffic such as the location,speed,and driving direction of mass passengers.It provides powerful data support for mining the spatio-temporal correlation between links in the road network and improving the accuracy of traffic parameters and state prediction.Existing research mainly focuses on short-term traffic flow prediction and congestion state prediction,and speed is the core indicator reflecting the road state,and speed prediction is also an important content of traffic parameter prediction.Based on the taxi GPS data,the paper predicts the average driving speed of urban links.First,the GPS data processing flow using the relational database MySQL is proposed,including time coordinate conversion,road division method,map matching,average travel speed estimation and numerical interpolation.In map matching,based on the advantages of existing algorithms and the characteristics of the data set in this paper,this paper makes full use of information such as road network topology,geometric factors and vehicle trajectories to match GPS trajectory points and correct offset points.Then,based on feature selection and deep learning knowledge,a feature selection method based on adaptive genetic algorithm and a Gated recurrent units neural network were introduced to establish a link travel speed prediction model considering multi-link spatio-temporal correlation.Finally,data experiments were carried out on the link travel speed of urban roads in Jinan.After regular analysis of the speed data set,the link travel speed was predicted,and the predictions of various algorithms were compared from the perspective of feature selection,spatial correlation and prediction model.The research in this paper can enrich the field of traffic parameter prediction in terms of spatial correlation and prediction model.The prediction results of the model in this paper are good,which helps to improve the scientific level of traffic management.