基于BP神经网络的高速公路事故预测模型

An Accident Prediction Model for Expressway Based on BP Neural Network

依据收集到的山岭区和丘陵区高速公路的几何线形、交通量和事故数据,建立高速公路路段基于几何线形指标和交通量的事故预测模型。利用平均影响值法对几何线形指标进行相关性检验,满足与事故率相关性条件的线形指标包括直线段长度、平曲线半径、偏角、竖曲线半径和纵坡坡度,根据检验结果划分事故预测单元。建立基于BP神经网络的事故预测模型,验证了模型在具体路段上的适用性。对模型进行灵敏度分析,确定各线形指标对事故率发生的影响程度。验证结果表明:模型准确度达到88%,该模型对山岭区或丘陵区高速公路基本路段具有广泛的适用性,对具体路段的预测具有较高的精度。

According to the geometric alignment,traffic volume and accident data of expressway in mountain and hilly areas in China,an accident prediction model is developed based on geometric alignment and traffic volume for expressway segments.A correlation test for geometric features is undertaken by mean impact value.The results show that the features are found to be significant over the test including tangent length,horizontal curve radius,deflection angle,vertical curve radius and longitudinal gradient.Accident factors are divided according to the test results.An accident prediction model is developed based on BP neural network.The accuracy of this model reaches 88%,and its applicability on specific expressway segments is verified.The relationship between alignment features and accident rate is verified with sensitivity analysis.The results of model validation show that this model can be applied to all types of expressway segments in mountain and hilly areas with a high accuracy.