Short Term Traffic Flow Prediction Of Road Network Based On Deep Learning

In order to solve the increasingly serious traffic congestion problems, Intelligent Transportation Systems has been widely used to alleviate the traffic jams and improve the road traffic efficiency in dynamic traffic management. Timely acquiring the real-time traffic data in the road network can be realized with the development of traffic data collection technology. A large amount of traffic information provides the data for analysis and prediction of the traffic state in the road network. Traffic flow forecasting plays an important role in intelligent traffic management and dynamic control, and it is the key to traffic guidance. Real-time and accurate short-term traffic flow prediction of road network contributes to analyze traffic condition, and plays an essentially important role in road network transportations planning and designing efficient control strategies.For large traffic network, traditional single section traffic flow prediction can not fully indicate the real-time traffic condition. Aiming at the shortcomings of current short-term traffic flow prediction methods, this paper proposes a forecasting method of short-term traffic flow prediction of road network based on deep learning. Firstly, analyzing the temporal and spatial characteristics of traffic flow data, according to the time characteristics of traffic flow, the data can be divided into two categories: weekend and non weekend, based on the correlation of traffic flow in space, sections are grouped by setting different correlation coefficient threshold; Secondly, we decompose traffic flow into a trend and a random fluctuation by using the spectral decomposition method; Once more, we decompose the traffic flow of network by CX and apply the correlative road sections to construct the compression matrix of road network; Then, we combine the deep belief network model in deep learning with the support vector regression to build DBN-SVR model of short-term traffic flow prediction; Finally, the feasibility of the method is verified by regional road network traffic data of actual freeway from Transportation Research Data Lab in US.Through the simulation analysis, it is concluded that:(1) trend signal can make an affect on prediction accuracy, and optimal spectrum threshold can make the prediction error reduce 5%;(2) the network running time of compressed prediction is reduced and saved 90%;(3) proposed prediction model in this paper outperforms other forecasting models in prediction accuracy. Compared with SVR model, the prediction error decreases 8%, and the average prediction accuracy of each section in road network can reach 92%.