Spatiotemporal characteristics and severity modelling of electric vehicle-pedestrian collision accidents

To investigate the spatiotemporal distribution pattern of electric vehicle-pedestrian collision accidents, the indicators of clustering strength and accident severity were incorporated into the traditional kernel density estimation method. A spatiotemporal kernel density estimation approach for traffic accident-prone locations was built. The random parameter Logit model based on mean heterogeneity was developed to model the severity of electric vehicle-pedestrian collision accident by considering the potential random variables and heterogeneity in accident variables. Key factors affecting the severity of electric vehicle-pedestrian collision accidents were identified, and the marginal utility of each significant variable was calculated to quantify its impact on the severity of electric vehicle-pedestrian collision accidents. Research results show that the electric/fuel vehicle-pedestrian collision accidents are predominantly located in urban center areas, whereas the distribution of electric vehicle-pedestrian collision accidents is relatively more concentrated. Moreover, the number and severity of electric vehicle-pedestrian collision accidents are higher on weekdays and daytime. Electric vehicle driver age ranging from 40 to 60 years old and accident occurring time between 18:00 and 6:00 the next day are identified as random variables. The utility value of driver age follows a normal distribution with standard deviation of -0.582 and variance of 0.926², and has mean heterogeneity in variables of road function level (trunk road) and vehicle type (truck). The utility value of accident occurring time follows a normal distribution with standard deviation of -0.313 and variance of 0.518², and there is mean heterogeneity in variable of weather (haze). Additionally, the factors such as male drivers, middle-aged and young drivers, and weekday travel are found to contribute to the varying degrees of severity in electric vehicle-pedestrain collision accidents. The Akaike information criterion of the random parameter Logit model considering mean heterogeneity decreases by 26 compared with the random parameter Logit model, and the logarithmic likelihood value increases by 94.272. The introduction of mean heterogeneity improves the fit goodness of severity model for electric vehicle-pedestrian collision accidents based on the random parameter Logit model, providing a more scientific and comprehensive explanation of causal principles behind electric vehicle-pedestrian collision accidents.