Study of Roadside Accident Prediction of a Multibridge and Multitunnel Section on an Expressway in a Mountainous Area

Abstract

In order to ensure the roadside traffic safety of a multibridge and multitunnel section on an expressway in mountainous areas, optimize the roadside traffic facilities, and reduce the road traffic accidents caused by the overly frequent changes of the driving environment among the bridge group and tunnel group and their intervals, combined with the driving rules, the prediction model of the numbers of roadside accidents, passenger and freight vehicle accidents in a multibridge and multitunnel section on an expressway in mountainous areas is established using statistics, machine learning, and other related theories. To analyze the influence of the driving environment of the expressway on drivers’ visual, psychological, and operational characteristics, ten prediction indicators are selected from the aspects of road alignment, traffic structure, traffic environment, and weather conditions. The action mechanism between roadside accidents and the ten predictive indicators is explained by employing Spearman correlation analysis. The roadside accident prediction model based on BPNN, GA-BPNN, and PSO-BPNN are established. MAE, RMSE, and MAPE are used as the model evaluation indicators to select the optimal model. The roadside accident data including rollover, side collision, and collision with fixed objects are verified with examples using the accident patterns of Chongqing–Hunan Expressway in the past five years. The results show that (1) the roadside accidents of a multibridge and multitunnel section on an expressway in a mountainous area are comprehensively affected by ten prediction indicators, which are positively correlated with the length of road section, the proportion of curved roads, and the proportion of bridges, and the influence of the length of the road section is the greatest indicator; (2) compared with BPNN and GA-BPNN prediction models, the errors of MAE, RMSE, and MAPE of PSO-BPNN are reduced by 18.5%, 17.65%, and 24.16% on average, and the model prediction error is smaller and the accuracy is higher; and (3) the accurate prediction of the number of roadside accidents and the number of passenger and freight vehicle accidents can provide effective decision-making support for the optimization design of roadside facilities.