Abstract Traffic safety state clustering has always been the focus of traffic safety research and the foundation of real‐time crash potential prediction. How to mine effective latent crash risk information and improve clustering effect are the goals and difficulties of traffic safety state clustering task. The conventional methods adopt independent feature extraction and clustering processing, which leads to mismatch problems and decrease clustering effect. To deal with the problems, a novel traffic safety state deep clustering network (TSDCN) is proposed. TSDCN integrates the feature extraction and clustering into an end‐to‐end deep hybrid network. A custom autoencoder is constructed to extract expressive risk feature and iteratively optimize clustering effects and feature extraction using a deep clustering layer. The three‐stage multitask strategy is designed to joint‐adjust shared network parameters and ensure convergence at different stages. The comparative experiments show the TSDCN achieves more outstanding cluster performance than those existing models. Moreover, the traffic safety state cluster results are statistically analysed and the crash risk level is quantified for each safety state. The risk‐quantized results are consistent with the real road crash situation and this confirms the safety state clustering effectiveness of TSDCN.
TSDCN: Traffic safety state deep clustering network for real‐time traffic crash‐prediction
2021
Aufsatz (Zeitschrift)
Elektronische Ressource
Unbekannt
Metadata by DOAJ is licensed under CC BY-SA 1.0
TSDCN: Traffic safety state deep clustering network for real‐time traffic crash‐prediction
Wiley | 2021
|Real-Time Traffic Network State Prediction for Proactive Traffic Management
Transportation Research Record | 2019
|Real-Time Crash Prediction Model for Application to Crash Prevention in Freeway Traffic
Transportation Research Record | 2003
|Real-Time Crash Prediction Model for Application to Crash Prevention in Freeway Traffic
Online Contents | 2003
|