基于ST-ANKDE的电动汽车交通事故时空多发点鉴别OA

Against the backdrop of rapid electric vehicle development,traffic safety issues have become increasingly prominent.Compared with conventional fuel-powered vehicles,electric vehicle accident data exhibit significant heterogeneity and sparsity,making it difficult for existing fixed-bandwidth kernel density estimation methods to effectively characterize their spatiotemporal distribution features,thereby constraining the accurate identification of high-risk areas.To improve the accuracy of electric vehicle accident hotspot identification,this study introduces a temporal angle mapping mechanism to capture the periodic evolution patterns of accidents and proposes a spatiotemporal adaptive network kernel density estimation method(ST-ANKDE)that integrates road network structure with temporal periodicity.By employing adaptive bandwidth adjustment,the method improves the estimation accuracy of local kernel functions,thereby enabling refined modeling of the spatiotemporal distribution of accidents.On this basis,a multi-hotspot identification framework that integrates accident density and severity is further developed.An EPDO(Equivalent Property Damage Only)weighting mechanism is introduced to quantify the influence of accident severity on identification results.Combined with hotspot analysis,a three-dimensional statistic is constructed to identify spatiotemporal hotspots from the ST-ANKDE kernel density estimation results,effectively screening out statistically significant high-accident areas.Empirical results based on nearly 50,000 electric vehicle accidents in a certain city show that high-accident road segments account for only 0.62%of all road segments but they cover 31.76%of total accidents,with a PAI(Predictive Accuracy Index)value as high as 51.22,confirming the accuracy and efficiency of the proposed method.In terms of spatial distribution,electric vehicle accident-prone areas are mainly concentrated at arterial road intersections,ramps,and commercial core areas.From the perspective of collision type,electric vehicle-motor vehicle collisions show the strongest road dependence;electric vehicle-pedestrians/bicycle collisions are clustered in densely populated areas,and collisions with fixed objects are the most dispersed,mainly occurring on urban fringe and suburban areas.The findings provide a basis for dynamic risk identification and targeted management of electric vehicle accidents and offer quantitative support for precise prevention and control of urban traffic accidents.