Abstract
Accurate traffic flow forecasting is an essential component of the Intelligent Transportation System (ITS). However, existing traffic forecasting methods using deep learning pay little attention to the pandemic's repercussions. This paper proposes a multiscaled deep learning framework called VMD-LSTM-ARIMA, which couples the variational mode decomposition (VMD) algorithm, long short-term memory (LSTM) neural network, and autoregressive integrated moving average (ARIMA) to accurately predict traffic flow time series. Just like any hybrid model, the proposal takes advantages of each one of these approaches, which enhances the performance of the overall forecasting model. Experiments were conducted on a US public traffic datasets, and the results showed that VMD-LSTM-ARIMA effectively increased the prediction accuracy.