Construction Of A Two-vehicle Crash Severity Assessment Model Based On The Pre-crash Vehicle State

Road traffic accidents cause significant losses to people’s lives and property,and among all accident types,two-vehicle collisions account for the highest proportion and cause the most serious casualty consequences.Road safety has focused on reducing the loss of two-vehicle collisions,especially the consequences of casualties.After long-term research,fruitful results have been achieved in accident severity evaluation and road accident reproduction techniques.However,since accident crash severity evaluation is a complex problem involving multiple disciplines,it requires continuous exploration and improvement,especially the introduction of pre-crash vehicle evaluation indexes to improve the accident severity evaluation model.Based on the point mass impulse-momentum law,this paper discusses the energy prediction method of deformation in a vehicle collision and derives the calculation formula of two-vehicle collision deformation energy prediction under different collision types with the precrash vehicle parameters and operating state parameters as independent variables,and verifies them by actual cases and PC-CRASH simulations.The validation shows that the prediction results are consistent with the actual case and simulation results,and the average error of the prediction results does not exceed 7% for both side and frontal and rear-end collisions.Therefore,the prediction accuracy of the method is high and can meet the accuracy requirements of crash severity assessment.This study deduces the formulae for calculating the crash severity indexes,such as the difference in velocity before and after the vehicle collision,energy equivalent velocity,crash momentum index,and crash energy index.Using 24,083 two-vehicle crash data from the FARS/CRSS in-depth database,a Bayesian ordered logit model was used to analyze the impact indexes of crash severity quantitatively.The results show that for each 1km/h increase in the speed difference and energy equivalent speed before and after the collision,the probability of serious injury accidents will increase by 0.49% and 0.22%,respectively.For each increase of0.1 in the crash momentum index and crash energy index,the probability of serious injury and fatal accidents will increase by 3.04% and 2.64%,respectively.The "Maximum Abbreviated Trauma Scale",widely used in medical and traffic fields,is the accident severity grading standard.The accident severity is classified into mild,moderate and severe levels according to the degree of human injury.Based on the machine learning method,the velocity difference before and after a collision,energy equivalent velocity,closure velocity,collision deformation energy,collision momentum index,and collision energy index were incorporated into the model assessment feature combinations to construct six two-vehicle crash severity prediction models.The results show that the Local cascading integration model considering crash severity has the best prediction performance among the six evaluation models.Through SHAP analysis and partial dependence analysis,the velocity difference before and after a collision,energy equivalent velocity,and crash momentum index is identified as the main decision factors for evaluating the models and revealing their effects on the prediction direction.Finally,this study combines the results of quantitative analysis to determine the index thresholds of pre-and post-crash velocity differences on three types of severity levels,which can be used as a reference for measuring accident severity.The research results provide new evidence to improve the traffic accident severity assessment model and provide reference for active vehicle safety prevention and control and risk avoidance strategies in the pre-crash stage.