Short-term traffic flow prediction of parallel roads based on transfer learning

Abstract

Short-term traffic flow prediction is of great significance for road construction and operation. Aiming at the possible problems of missing data or insufficient sample data, the short-term traffic flow prediction of parallel road sections is studied using the traffic flow data collected in Shandong Province and the transfer learning idea. Firstly, the spatiotemporal correlation of the traffic flow of the parallel road section is analyzed, and the correlation coefficient is calculated; Secondly, the long-term and short-term memory network model is used to predict the section flow of expressway; Finally, the model based transfer learning is used to predict the cross section flow of parallel roads. The prediction results show that the average absolute error, the mean square error, the root mean square error and the average absolute percentage error of the transfer learning based prediction algorithm decrease by 2.56%, 3.59%, 1.81% and 4.41% respectively compared with the direct prediction of the short-term memory network model. In addition, the prediction model based on transfer learning can significantly improve the prediction efficiency, and can provide feasible methods for traffic flow prediction in areas lacking equipment collection or data.