面向多尺度数据的高速公路事故发生机理与风险模型移植研究

高速公路交通事故频发且后果严重，严峻的安全问题亟待解决。针对当下数据挖掘技术与道路交通安全问题兼容性差、实时风险模型无法有效辨识高速公路多场景风险因子的异质性、有限数据无法支撑实时风险建模且难以指导工程实践等问题。本项目将考虑高速公路在时空维度的场景差异，通过优化既有数据挖掘技术，提出一种兼容性好、精确度高的改进关联规则挖掘算法，揭示高速公路交通事故内在机理；基于统计学理论，设计面向多场景高速公路的事故风险识别模型，构建实时风险场景靶向预测体系；应用机器学习手段，探究模型移植方法，研究基于多尺度数据的实时风险建模方法；通过算法嵌套，完成基于交通安全大数据平台的实例验证分析。项目旨在阐明不同场景高速公路的交通事故发生规律、实现对多场景高速公路的风险靶向预警、并突破国内相关数据精度不足的技术瓶颈。相关成果对丰富高速公路动态安全控制理论方法，提升我国高速公路风险检测技术水平具有重要意义。

Freeway traffic crashes are common occurrences, with serious consequences and safety problems that need to be solved urgently. In view of the current incompatibility of data mining technology and road safety problems, real-time crash risk models cannot effectively identify the heterogeneity of multi-scene risk factors of the freeway, and limited data cannot support real-time risk modeling and guide engineering practice. This project will consider the difference of scene in Spatio-temporal dimension of the freeway and propose an improved association rule mining algorithm with good compatibility and high accuracy by optimizing existing data mining technology and will reveal the internal mechanism of freeway crashes. Based on statistics theory, a multi-scene freeway crash risk identification model will be designed, and a real-time risk scenario targeted prediction system will be constructed. Machine learning methods will be applied to explore model transplantation methods, and real-time risk modeling methods based on multiscale data will be studied; Through algorithm nesting, a case application analysis based on traffic safety big data platform will be completed. This project aims to clarify the occurrence rules of freeway crashes in different scenes, realize the risk targeted prediction and warning for multi-scene freeway, and break through the technical bottleneck of insufficient accuracy of China’s relevant data. The related results and achievements will be of great significance to enrich the theory and method of freeway dynamic safety control and improve the technical level of China’s freeway risk detection.