A new traffic flow prediction based on multi-scale convolutional neural network combined with LSTM neural network

Abstract

In this study, we proposed a novel traffic flow prediction model aimed at providing more accurate and effective traffic information. The proposed model utilized a multi-scale convolutional neural network in conjunction with a long short-term memory (LSTM) neural network to efficiently extract spatial and temporal features from the data. The network model was trained using the L1 loss function to optimize its performance.To evaluate the performance of the proposed model, we conducted experiments on two public datasets, PeMSD4 and PeMSD8. The results demonstrated the model's strong competitiveness in the field of traffic flow prediction. Specifically, on the PeMSD4 dataset, our model achieved a mean absolute error (MAE) of 20.94, a mean absolute percentage error (MAPE) of 13.32%, and a root mean square error (RMSE) of 30.96. These findings suggested that the proposed model hold significant promise in the realm of traffic flow forecasting.