山东交通学院学报

为分析城镇土地利用及路网形态对过境公路交通事故的影响程度，以北方某市G503、G302、G334、G203、S514等5条公路沿线的33个城镇为研究对象，通过对城镇路网建模获取拓扑特征参数，采用兴趣点(point of interest, POI)计算土地利用强度与土地混合利用程度，采用随机森林模型分析自变量对过境公路交通事故的影响程度，并通过负二项回归分析自变量与过境公路交通事故的相关性。结果表明：随机森林模型中的城镇POI数、过境公路介数中心性、过境公路节度、城镇节点数、土地混合利用程度的特征权重分别为0.62、0.14、0.11、0.07、0.06,城镇POI数对过境公路交通事故的影响最大，过境公路介数中心性与节点数对过境公路交通事故有一定影响，城镇节点数、土地混合利用程度对过境公路交通事故的影响较弱；在负二项回归方法下，过境公路介数中心性、城镇POI数与过境公路交通事故显著正相关，城镇节点数、过境公路节度和土地混合利用程度与过境公路交通事故的相关性不显著。随机森林模型结论与负二项回归分析结论接近，随机森林模型对自变量的非加和性与非线性特点适应较好，适用于交通事故致因分析。控制城镇规模、合理规划城镇路网对减少过境公路交通事故风险和提高交通安全有重要意义。

In order to analyze the extent of the influence of urban land use and road network morphology on interregional highway traffic accidents, 33 towns along five highways(G503, G302, G334, G203, S514) in anorthern city are selected as research subjects. The topology characteristic parameters are obtained by modeling the urban road network, and the intensity of land use and the degree of mixed land use are calculated using points of interest(POI). The random forest model is employed to analyze the influence of independent variables on interregional highway traffic accidents, and the correlation between independent variables and interregional highway traffic accidents is analyzed using negative binomial regression. The results indicate that the feature importances for the number of POIs, betweenness centrality of the interregional highways, degree of the interregional highways, number of urban nodes, and degree of mixed land use in the random forest model are 0.62, 0.14, 0.11, 0.07, and 0.06, respectively. The number of POIs has the greatest influence on interregional highway traffic accidents, while betweenness centrality and the number of nodes of the interregional highways have some influence. The number of urban nodes and the degree of mixed land use have weaker influence on interregional highway traffic accidents. In the negative binomial regression method, the betweenness centrality of the interregional highways and the number of POIs are significantly and positively associated with interregional highway traffic accidents, while the number of urban nodes, degree of the interregional highways, and degree of mixed land use are not significantly correlated. The conclusions of the random forest model are close to those of the negative binomial regression analysis. The random forest model adapts well to the non-additive and non-linear characteristics of the independent variables and is suitable in analyzing the causes of traffic accidents. Controlling urban scale and reasonably planning urban road networks are of great significance in reducing the risk of interregional highway traffic accidents and improving traffic safety.