Short-time traffic flow prediction based on a hybrid model of Bi-LSTM-CNN

Intelligent traffic control and induction is now an important means to achieve convenient and effective travel, among which, real-time and accurate short-time traffic flow prediction is the key factor. Accurate prediction of short-term traffic flow can help people choose paths, reduce travel time and traffic congestion and so on. In terms of prediction,  Long Short-Term Memory (LSTM) model is more and more favored by scholars at home and abroad. However, at present, most scholars compare the difference between the prediction performance of LSTM and other models, and more research focuses on building other models with better spatial performance. Few scholars study the improvement of traffic flow prediction performance of LSTM models and their variants combined with deep learning models. In view of this point, this paper proposes a hybrid model of double-layer Bi-LSTM and Convolutional Neural Network (CNN) for short-term traffic flow prediction. The experimental results show that the prediction errors of the hybrid model are 9.8607, 0.0041 and 0.0397, respectively under the Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) evaluation indexes, which are 6.1532, 0.007 and 0.0214 higher than that of the Gated Recurrent Unit (GRU) model with relatively poor performance in the baseline model. The findings of this paper can reflect the excellent performance of the hybrid model for short-term traffic flow prediction and lay a foundation for the further development of the deep learning hybrid model for predicting traffic flow.