Citation

Wang K, Ma C, Huang X. Int. J. Mod. Phys. C 2023; 34(10): e2350127.

Copyright

(Copyright © 2023, World Scientific)

DOI

10.1142/S0129183123501279

PMID

unavailable

Abstract

Traffic speed is an essential indicator for measuring traffic conditions. Real-time and accurate traffic speed prediction is an essential part of building intelligent transportation systems (ITS). Currently, speed prediction methods are characterized by insufficient short-term prediction accuracy and stability, nonlinear, nonstationary, strong fluctuation and relatively small sample size. To better explore the traffic characteristics of the road networks, a hybrid prediction model based on wavelet transform (WT) of the autoregressive moving average model (ARIMA) and gate recurrent unit (GRU) was constructed. First, this model decomposes the original traffic speed data into low-frequency data, and high-frequency data by WT. Second, the ARIMA and GRU models are used to model data predictions in two frequency bands, respectively. Finally, the prediction result of the predicted value is fused. In addition, in this paper, traffic speed data of four sections in Guangzhou from 1 August to 31 September 2016 are taken as examples to test the validity, applicability, and practicability of the model. The results show that compared with ARIMA, LSTM, GRU, RNN, and other single models and hybrid models, the prediction method proposed in this paper has higher prediction accuracy and can provide a more scientific decision-making basis for urban traffic management.

Language: en

Keywords

ARIMA; deep learning; GRU; Traffic speed prediction; wavelet transform