Citation

Xiong X, Liu Q, Shen Y, Cai Y, Chen L. China Saf. Sci. J. 2022; 32(5): 170-176.

Copyright

(Copyright © 2022, China Occupational Safety and Health Association, Publisher Gai Xue bao)

DOI

10.16265/j.cnki.issn1003-3033.2022.05.1602

PMID

unavailable

Abstract

In order to reduce highway traffic accidents, a traffic risk state prediction model was studied by using LSTM neural network and BF. Firstly, time dependency existing in historical traffic flow risk data was studied through LSTM module. Then, real-time risk prediction performance was improved by BF module integrating LSTM prediction results. Finally, with accident data and traffic flow data of Ningbo roundabout in 2020 as an example, multi-step feature variables were constructed in the form of migration time window based on spatio-temporal data between upper and lower port stations within 20 min prior to accidents, and 5-fold cross-validation was carried out. The results show that the precision and recall rate of LSTM model are higher than that of random forest (RF) algorithm, and F1 score of the final prediction result is close to 0. 80 by adding BF module. © PHYSOR 2022 China Safety Science Journal. All rights reserved.

Language: zh