The 20th and 21st joint COTA International Conference of Transportation Professionals (CICTP 2020-21)

The rapid development of artificial intelligence provides a new way for the research of transportation systems. Aiming at the problems of short-term traffic flow prediction such as lagging, insufficient time variable characteristics extraction, and low prediction accuracy, this paper uses the correlation of highway traffic flow in time as the basis to extract 4 types of variables closely related to time, and establish 6 Long-Short-Term Memory (LSTM) models respectively. The results show that a combination model that simultaneously considers multiple time variables can effectively reduce the lag in time series prediction. In addition, we establish two comparison models. The results show that the selected variables have both temporal characteristics and non-temporal characteristics. Capturing these characteristics can help improve the accuracy of the model. Finally, the Random Forest (RF) algorithm is used to rank the importance of variables, which further shows that the combined model has a certain feasibility.

CICTP