中文摘要：

为识别高速公路路侧事故的主要诱导因素,分析路侧事故起数与道路线形和交通条件之间的关系,以广珠（广州-珠海）东线高速公路3年中发生的178起路侧事故为基础,分别采用定长法和不定长法划分路段单元,从道路线形和交通条件中选取11个自变量,采用零堆积负二项回归模型建立路侧事故起数预测模型。选择Vuong检验统计量、对数似然值和AIC信息准则3个指标进行模型的拟合优度检验,选择相对误差和累积残差2个指标进行模型的拟合准确性检验;通过对比分析负二项回归模型和零堆积负二项回归模型的拟合优度和拟合准确性检验结果判断其优劣性,并采用弹性分析确定较优模型中显著自变量对因变量的影响程度。研究结果表明：无论采用定长法划分还是不定长法划分路段单元,零堆积负二项回归模型构建的路侧事故起数预测模型明显优于负二项回归模型;采用零堆积负二项回归模型构建的路侧事故起数预测模型,其定长法划分的路段单元模型的拟合准确性优于不定长法;对于定长法划分的路段单元,基于零堆积负二项回归模型的路侧事故起数预测模型有5个自变量对路侧事故起数均有显著影响,影响程度大小依次为车道数、曲率变化率、曲线比例、曲度和平均纵坡坡度。

英文摘要：

To identify the main induction factors of roadside accident and analyze the relationship between roadside accident frequency and main influence factors such as road geometry,traffic conditions of expressway,based on 178 cases of roadside accidents occurred in Eastern Guangzhou to Zhuhai Expressway in three years,this paper used the fixed-length method and homogeneous longitudinal grade method to divide the study section,selected eleven independent variables from the aspects of road geometry and traffic conditions,and adopted zero-inflated negative binomial regression model to establish roadside accident frequency prediction model.Three indicators including Vuong test statistics,log likelihood value,and Akaike＇s informationcriterion were used to test the goodness-of-fit of the model.Two indicators including relative error and cumulative residual were used to test the prediction accuracy of the model.To further judge the better model,negative binomial regression model and zero-inflated negative binomial regression model were compared by the goodness-of-fit and prediction accuracy.Elastic analysis was used to determine the influence degree of the independent variables on the dependent variable in the better model.The results show that the roadside accident frequency prediction model based on zero-inflated negative binomial regression model is better than that of negative binomial regression model both for fixed-length segmentation method and homogeneous longitudinal grade segmentation methods.For roadside accident frequency prediction model based on zero-inflated negative binomial regression model,the prediction accuracy of fixed-length segmentation method is better than that of homogeneous longitudinal grade segmentation method.For fixed-length segmentation method,five independent variables in roadside accident frequency prediction model based on zero-inflated negative binomial regression model have significant impact on roadside accident frequency,and the influence degree in descending order is number of lane,cu