Prediction of the vehicle lane-changing distance in an urban inter-tunnel weaving section based on wavelet transform and dual-channel neural network

Abstract

Vehicle lane-changing behaviour is often regarded as transient traffic behaviour while ignoring behavioural characteristics of the lane-changing process. A combined prediction model based on wavelet transform (WT) and dual-channel neural network (DCNN) is proposed to explore the selection behaviour of lane-changing distance by taking lane-changing behaviour in an urban inter-tunnel weaving section. Firstly, the extracted lane-changing data are analysed for correlation and noise reduction, and the main factors affecting lane-changing distance are taken as input variables of the model. The trajectory data of the inter-tunnel weaving section of the “Jiuhuashan-Xi'anmen” tunnel in Nanjing, China, are used to improve the prediction of vehicle lane-changing distance by training the model. The results show that the proposed WT-DCNN model has high prediction performance when compared with existing artificial neural network (ANN), DCNN and wavelet neural network (WNN) models. The characterization and study of the typical lane-changing behaviour in the weaving section can lay the theoretical foundation for the development of an urban inter-tunnel weaving section management scheme.