A Novel Deep Learning Model Based on the Improved Loss Function for Traffic Flow Forecasting

Traffic flow prediction is of great significance to solve the problem of traffic congestion. Traffic sequence has time correlation and space correlation, and the correlation is constantly changing with time. How to better capture spatio-temporal dynamics is the key to traffic flow prediction. In this paper, spatial attention mechanism is designed to capture spatial correlation, and sequence to sequence (seq2seq), composed of Gated Recurrent Unit (GRU) units, is designed to capture temporal correlation. Moreover, the information fusion layer is added to capture the external influence, and the improved loss function is used for model training. The experimental results show that the designed network has achieved good results.