A Traffic Flow Prediction Method Based on Road Crossing Vector Coding and a Bidirectional Recursive Neural Network

• Published in: Electronics (Basel) 2019-09, Vol.8 (9), p.1006
• Main Authors: Zhao, Shuanfeng, Zhao, Qingqing, Bai, Yunrui, Li, Shijun
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Deep learning; Feature extraction; Generalized linear models; Model testing; Neural networks; Roads; Time series; Traffic congestion; Traffic control; Traffic flow; Traffic information; Traffic intersections; Traffic models; Traffic volume
• ISSN: 2079-9292; 2079-9292
• DOI: 10.3390/electronics8091006

Aiming at the problems that current predicting models are incapable of extracting the inner rule of the traffic flow sequence in traffic big data, and unable to make full use of the spatio-temporal relationship of the traffic flow to improve the accuracy of prediction, a Bi-directional Regression Neural Network (BRNN) is proposed in this paper, which can fully apply the context information of road intersections both in the past and the future to predict the traffic volume, and further to make up the deficiency that the current models can only predict the next-moment output according to the time series information in the previous moment. Meanwhile, a vectorized code to screen out the intersections related to the predicting point in the road network and to train and predict through inputting the track data of the selected intersections into BRNN, is designed. In addition, the model is testified through the true traffic data in partial area of Shen Zhen. The results indicate that, compared with current traffic predicting models, the model in this paper is capable of providing the necessary evidence for traffic guidance and control due to its excellent performance in extracting the spatio-temporal feature of the traffic flow series, which can enhance the accuracy by 16.298% on average.