Multi-step traffic flow prediction method based on the Conv1D + LSTM

Abstract

In view of the present traffic flow prediction model only takes into consideration the time series of traffic flow, and make one-step prediction rather than multi-step prediction, while ignoring the influence of external factors (actual factors) on traffic flow, this paper designed a multi-step traffic flow prediction model, which combines one-dimensional convolution (Conv1D) and Long Short-Term Memory Network (LSTM). Considering with external factors such as weather, time information and holidays, this model uses Conv1D to model the time feature, period feature and local correlation feature of traffic flow, and makes multi-step prediction through LSTM. The experimental results show that the prediction accuracy of Conv1D+LSTM model is obviously higher than baseline methods, which verifies the validity of multi-step prediction of traffic flow considering external factors.