基于SVM模型的山区高速公路多车事故影响因素分析

Influencing Factors Analysis of Multiple Vehicle Accidents in Mountainous Expressway Based on SVM Model

为了分析山区高速公路多车事故的影响因素,收集了2012—2017年昌金与泰赣两条典型山区高速的1351条多车事故数据.考虑利用传统离散选择模型分析山区多车事故影响因素,结果发现该数据违背了离散选择模型假定条件,故而采用支持向量机模型进一步分析数据.考虑线性核函数、非齐次多项式核函数、高斯径向基核函数构建支持向量机模型,利用网格搜索法对模型参数进行了寻优,并通过敏感性分析评估了事故潜在风险因素对事故严重程度发生概率的影响.研究表明:高斯径向基核函数条件下的支持向量机模型预测山区多车事故结果最佳,平均预测精度0.733,表现出良好的分类识别效果和泛化能力,说明在传统离散选择模型应用不佳情况下,可利用支持向量机模型进行事故数据的挖掘分析.

In order to analyze the influencing factors of multi-vehicle accidents on mountain expressways,the data of 1351 multi-vehicle accidents on two typical mountain expressways,Changjin and Taigan,from 2012 to 2017 were collected.Considering the traditional discrete selection model to analyze the influencing factors of multi-vehicle accidents in mountainous areas,it is found that the data violates the assumptions of the discrete selection model,so the support vector machine model was used to further analyze the data.Considering linear kernel function,non-homogeneous polynomial kernel function and Gaussian radial basis kernel function,the support vector machine model was constructed.The model parameters were optimized by grid search method,and the influence of potential risk factors on the probability of accident severity was evaluated by sensitivity analysis.The research shows that the support vector machine model under the condition of Gaussian radial basis function has the best prediction result for multi-vehicle accidents in mountainous areas,with an average prediction accuracy of 0.733,showing good classification and recognition effect and generalization ability.It shows that the support vector machine model can be used to mine and analyze accident data when the traditional discrete selection model is not well applied.