考虑事故数据零截尾分布和异质性特征的交通安全分析方法研究

我国交通安全数据相对不完备，零值缺失现象普遍存在，将这类数据应用于基于完备交通数据的传统事故模型难以准确预测交通安全水平，数据异质性也会影响模型的可靠性，却易被忽略。基于国内零截尾分布的交通安全数据，并考虑其异质性的事故预测与路网风险评估方法研究鲜有报道。本项目针对国内安全数据现状，首先构建零值缺失的交通安全数据库，识别影响交通事故的主要因素；其次，深入分析基于零值缺失数据的分布规律，并考虑数据异质性分别构建交叉口和路段的零截尾事故预测模型；进而，与传统模型对比以验证模型有效性，并分析其适用范围；最后，提出基于考虑数据异质性的零截尾事故预测模型的路网风险评估方法。项目基于零值缺失的有限事故数据样本，考虑其异质性构建的事故预测模型能有效提高预测精度，可建立一套适用于我国数据特性的交通安全分析方法，为主动交通安全管理提供理论依据和技术支撑，对提升我国交通安全管理水平具有重要理论意义和应用价值。

Traffic safety data are not very well prepared in China compared to some other countries. It is quite common that zero crash counts are missing in the data (i.e., zero-truncated data). Crash prediction models developed using conventional methods, which require complete safety data, with the zero-truncated data are not reliable in predicting crashes. In addition, the performance of the models may be worse if the heterogeneity issue is not addressed. Very limited studies have developed crash prediction models and roadway risk assessment approaches considering the truncated distribution and heterogeneity in the crash data in our country. .In this project, the Research Team will first develop a safety database based on the specific situations of traffic crash records in our country, and identify important factors that contribute to the occurrence of crashes. Then, analyze the characteristics and distributions of crash count data with missing zeroes, and develop advanced intersection and segment, separately, zero-truncated crash prediction models with consideration of heterogeneity. Followed that, the team will conduct comparative analysis between the proposed model and the conventional model to verify the results of the proposed models, and further identify situations when the proposed models are applicable. Finally, the Research Team will develop methods to assess roadway network crash risk with the developed models. .This project proposes crash prediction models based on limited zero-truncated safety data with consideration of heterogeneity, which improves the prediction accuracy, and establishes risk assessment methods that are applicable to the crash characteristics in China. The results will provide support for proactive safety, effectively improve the level of safety management in our country, and reduce the occurrence of traffic crashes. The implementation of this project is significant for theoretical and practical roadway safety management.