A Hybrid Deep Learning Framework for Long-Term Traffic Flow Prediction
Yiqun Li, Songjian Chai, Zhengwei Ma, Guibin Wang
Abstract
An accurate and reliable traffic flow prediction is of great significance, especially the long-term traffic flow prediction e.g., 24 hours, which can help the traffic decision-makers formulate the future traffic management strategy. However, the long-term traffic flow prediction imposes great challenges for decision-makers due to the nonlinear and chaotic feature of traffic flow. Therefore, in this paper, we proposed a hybrid deep learning model based on wavelet decomposition, convolutional neural network-long and short-term memory neural network (CNN-LSTM), called W-CNN-LSTM, to prediction next-day traffic flow. The wavelet decomposition technology is used to decompose the original traffic flow data into high-frequency data and low-frequency data for the improvement of predictive accuracy. The decomposed sequences are fed into a CNN-LSTM deep learning model, where the long-term temporal features of traffic flow can be well captured and learned. The numerical experiment is carried out against five benchmarks based on England traffic flow dataset; the results show that the proposed hybrid approach can achieve superior forecasting skill over the benchmarks.