Prediction Of Urban Intersection Turning-Level Traffic Congestion Based On Deep Learning

With the rapid economic growth,urbanization speeding up,urban motor vehicle ownership is rising year by year,whose growth rate reached five times that road construction.This phenomenon leads to traffic congestion gradually aggravate,causing a huge economic loss and environmental problems,making traffic congestion has become one of the international major social problems.The management of traffic congestion and the avoidance of traffic congestion are dependent on the acquisition and prediction of traffic condition information.With the development of the Internet and Artificial Intelligence,massive trajectory data and deep learning technology can be used to obtain a wide range of urban traffic condition information,and adaptive learning of high-dimensional complex rules can be used to obtain more accurate prediction ability.However,the existing researches have some limitations.From the perspective of method,the accuracy of traffic status information acquisition is limited by the trajectory data sampling frequency and the prediction model is difficult to effectively model the traffic spatial pattern.From the perspective of application,most of them analyze and predict the value of traffic flow or traffic condition grade,but they lack the detailed study on urban congestion in a fine scale,and it is difficult to reflect the impact of congestion intensity and delay.Therefore,with the trajectory data and deep learning technology,this paper focuses on the urban intersection areas,and studies the prediction method of turninglevel congestion information at intersections,so as to realize the short-term prediction of traffic speed and queue length at each steering direction,and proposes improvement and optimization methods according to the shortcomings of existing studies.The main contents and research work of this paper can be summarized as follows:(1)Introduce the background and significance of traffic congestion prediction in detail,and summarize related researches;The concepts and techniques of traffic flow,short-term traffic prediction,spatio-temporal big data and deep learning are introduced in detail.(2)In order to break through the limitation of low-frequency trajectory data and effectively improve the accuracy of traffic condition information,a queue-starting point estimation method is proposed to obtain queue start point accurately,and a wavelet analysis method is introduced to filter out the noise caused by data sparsity and instability.(3)In order to improve the accuracy and reliability of prediction by modeling the temporal and spatial rules of congestion at the traffic network,a turning-based graph construction method is proposed to construct the topological structure between the intersections.A deep neural network model is designed to capture the spatio-temporal pattern of intersection turning by means of GCN,multi-scale historical data and LSTM.(4)The method was implemented with the Spark engine and Keras deep learning framework,and Wuhan taxi trajectory data were used for experiment.The experimental results were presented and analyzed.The experimental results show that by improving the extraction method of traffic condition information,the quality of traffic condiction information can be improved effectively.The accuracy of the prediction method proposed in this paper is better than that of the benchmark method,which indicates that this method is effective and advanced.