Analysis of factors affecting crash under risk scenarios based on driver homogenous clustering.

Research on road safety has focused on analyzing the factors that affect crashes. However, previous studies have often neglected differences in crash causation among heterogeneous clusters of drivers. In particular, the differences in the combined effect mechanisms of the factors in the risk scenarios have not been completely explained. Therefore, this study used the K-means algorithm to perform multidimensional feature homogeneous clustering for drivers involved in crashes and near-crashes. Structural equation modeling involving mediating effects was introduced to explore the direct and indirect effects of each influencing factor on vehicle crashes under risk scenarios and compare the differences in crash causation among different driver clusters. The results indicate that the drivers who experienced the risk scenarios can be classified into two homogeneous driver clusters. Significant differences exist in the demographic characteristics, intrinsic driving characteristics, and crash rates between them. In the risk scenario, traffic factors, distraction state, crash avoidance reaction, and maneuver judgment directly affect the crash outcomes of the two cluster drivers. Demographic characteristics and environmental factors have fewer direct influence on the crash outcomes of two-cluster drivers, but produce more complex mediating effects. Analysis of the differences in the influence of factors between clusters indicates that the fundamental cause of crashes for cluster 1 drivers includes poor driving skills. In contrast, cluster 2 drivers' crashes were more influenced by traffic conditions and their safety awareness. The analysis method of this study can be used to develop more targeted road safety policies to reduce the occurrence of vehicle crashes.

A Recognition Method of Truck Drivers' Braking Patterns Based on FCM-LDA2vec.

Taking truck drivers' braking patterns as the research objects, this study used a large amount of truck running data. A recognition method of truck drivers' braking patterns was proposed to determine the distribution of braking patterns during the operation of trucks. First, the segmented data of braking behaviors were collected in order to extract 25 characteristic parameters. Additionally, seven main correlation factors were obtained by dimensionality reduction. The FCM clustering algorithm and CH scores were used to identify nine categories of truck drivers' braking behaviors. Then the LDA2vec model was used to identify the distribution of different braking behavior words in braking patterns, and three categories of truck drivers' braking patterns were identified. The test results showed that the accuracy of the truck drivers' braking pattern recognition model based on LDA2vec was higher than 85%, and braking patterns of drivers in the daily operation process could be mined from vehicle operation data. Furthermore, through the monitoring and pre-warning of the braking patterns and targeted training of drivers, traffic accidents could be avoided. At the same time, this paper's results can be used to protect human life and health and reduce environmental pollution caused by traffic congestion or traffic accidents.