Abstract:

In view of the frequent rear-end accidents of trucks on highways, the risk of rear-end collision of trucks on highways was assessed, and the influence of traffic flow characteristics was analysed to reduce the occurrence of rear-end accidents. According to the German HighD open-source dataset, the TTC thresholds of different conflict risk levels were used as the classification standard, and the vehicle trajectories and traffic parameters of trucks were extracted. The risk models of rear-end collision of trucks were established based on Random Forest (RF) model, Support Vector machine (SVM) model, and Artificial Neural Network (ANN) model respectively. The overall forecasting ability of each model was compared with the evaluation indexes such as confusion matrix, area under the receiver operating characteristic curve (AUC) and Kolmogorov-Smirnov (KS) test. The model with the best prediction accuracy was selected to analyse the influence of each characteristic parameter on the rear-end risk. The results show that the prediction accuracy of the RF model is 75%, which is 8% and 10% higher than that of the SVM model and the ANN model, respectively. The prediction accuracy, recall, AUC and KS values of the RF model are better than those of the SVM and ANN models. The three parameters of minimum headway, standard deviation of vehicle speed and standard deviation of acceleration have the highest influence on the risk of rear-end collision of trucks.

Key words: machine learning, freeway, truck, risk of rear-end collision, prediction model, time-to-collision(TTC)