First Vehicle Arrival Time Prediction at Signalized Intersection Based on Wavelet-Elman Neural Network

• Published in: NeuroQuantology Vol. 16; no. 6
• Main Authors: Ao, Guchang, Zhang, Huiling, Yang, Linyu, Jiang, Wuting
• Format: Journal Article
• Language: English
• Published: Bornova Izmir NeuroQuantology 2018
• Subjects: Neural networks; Optimization; Time series; Traffic intersections; Wavelet transforms
• ISSN: 1303-5150; 1303-5150
• DOI: 10.14704/nq.2018.16.6.1657

There is a close relationship between the vehicle arrival time at signalized intersection and the vehicle delay at intersection entrance. When the red light of the signalized intersection is on, the first vehicle stopping at the stop line of the entrance is taken as the study object, and the time difference in the red light’s turning on and the first vehicle’s arriving at the stop line is defined as the first vehicle arrival time at signalized intersection. There is great randomness in the first vehicle arrival time series at signalized intersection. Firstly, the wavelet transform (WT) method is adopted to decompose the non-stationary original arrival time series into low-frequency signal and high-frequency signal. Then the dynamics and fast feedback of Elman neural network are used to predict different signals respectively. Finally, a final predicted result of the first vehicle arrival time is obtained when they are subject to linear superposition. The result shows that the error of the first vehicle arrival time prediction at signalized intersection based on wavelet-Elman neural network is small, and the predicted value is highly consistent with the actual value, which can provide reliable data source for delay parameter extraction and signal timing optimization at signalized intersection.