Identifying spatio-temporal pattern of electric vehicles involved traffic accidents

Abstract

The increasing popularity of electric vehicles (EVs) has given rise to concerns regarding emerging safety challenges. To address this, an enhanced spatio-temporal kernel density estimation (STKDE) method is developed, which leverages GIS technologies to analyze the temporal evolution of EV accidents across different locations. Our approach introduces several innovations: (1) The incorporation of network distance to address overestimation associated with the conventional Euclidean distance; (2) Integration of a Severity Index (SI) to rationalize accident weighting; (3) Utilization of a likelihood cross-validation method to determine optimal bandwidths for simplified computation. Additionally, statistical significance tests and a ranking system based on cluster strength are implemented to enhance the accuracy and stability of hotspot identification compared to the traditional kernel density estimation (KDE) method. Empirical analyses reveal spatial trends, indicating a higher likelihood of EV accidents occurring near city administrative centers and along arterial roads. Temporally, our findings show increased daytime hotspot density with distinct morning and evening peaks compared to nighttime. Furthermore, a total of 209 hazardous locations, containing 907 accidents, representing 23.4% of all accidents are filtered. Our improved STKDE approach demonstrates a mean hit rate of 0.553 and a PAI of 5.573, which are 3.5% and 27.5% higher, respectively, than those achieved with SKDEKDE. These insights can assist transportation agencies in implementing targeted interventions and resource allocation strategies.