BP-Garson算法的高桥隧比路段交通事故预测研究

Research on traffic accident prediction of the freeway with high bridge and tunnel percentage based on BP Garson algorithm

基于R语言平台及BP神经网络算法对高桥隧比路段交通事故进行拟合建模,并采用Garson算法对其进行预测分析。以连续3年的交通事故数据建立预测模型,将该路段的养护数据作为路面路表性能变化的输入参数。对数据样本进行预处理分析,将路面行驶质量指数(RQI)、路面横向力系数(SFC)和路面车辙深度指数(RDI)作为路面路表性能变化的输入参数,以事故是否发生于桥隧段作为输出值,基于“路面性能—桥隧段发生事故”样本建立二值分类预测模型。ROC曲线和混淆矩阵的分析结果表明:神经元个数为8时的模型有较好的预测效果;Garson算法分析表明:用于表征抗滑能力的SFC是影响高桥隧比路段事故发生的主要因素。

To explore the law of traffic accidents of highway section with bridge-to-tunnel ratio in mountainous areas under variation on road surface performance, model building was conducted for fitting traffic accidents by using R language platform and BP neural network algorithm and analyzed using Garson algorithm.The prediction model is established based on the three-year traffic accident data, and the maintenance data of the road section are used as the input parameters of pavement surface performance change.Surface maintenance data were utilized as input parameters of variation on road surface performance.Preprocessing of data samples was prior to introducing of Riding Quality Index(RQI), Sideway Force Coefficient(SFC) and Rutting Depth Index(RDI), which were selected as input parameters of pavement performance.Whether accidents occurred in the bridge and tunnel section was used as the output parameter.A binary classification prediction model was established, which was based on the sample pair of pavement performance and accident happened in bridge and tunnel section.The results of ROC curve and confusion matrix show that the model with 8 neurons has better prediction effect.Analysis of Garson algorithm indicated that SFC, which was used to characterize the surface skid-resistance, was a main factor of accidents happened in highway section with high bridge and tunnel ratio.