近年来，区域交通事故建模成为道路交通安全的热点研究之一，旨在将影响交通小区安全水平的各因素纳入交通规划阶段考虑，实现从源头上降低交通事故风险。已有研究对区域交通曝光变量的度量不够准确，对影响变量的复杂时空耦合关系考虑不充分，导致现有模型的预测精度和可移植性较差，无法深入揭示区域事故风险的影响机理。对此，本项目拟从大数据中提取区域交通曝光信息，归纳大数据驱动的出行特征辨识和融合方法；构建基于深度学习的区域交通事故模型，揭示区域事故风险的时空演化机理和传播规律；解析区域事故风险时空演化的分层致因，归纳区域交通事故风险的前兆特征；建立路网特征–交通出行特征–区域事故风险之间的关联机制，解析不同交通出行特征和道路网络设计参数对区域事故风险的影响机理，为复杂交通环境下的区域安全调控和道路优化设计提供理论基础。通过研究，形成大数据驱动的区域交通事故建模理论和方法，为保障道路安全、高效运行提供科学支撑。

Recently, zonal traffic crash modeling has become one of the hot research topics in the field of road traffic safety. Zonal traffic crash modeling aims to incorporate various factors of zonal safety levels into traffic planning stage, which can potentially reduce crash risk from the source. Existing studies did not evaluate zonal traffic exposure accurately and neglected the complex spatiotemporal correlations among influencing variables, resulting in lower model prediction accuracy and poor transferability. Thus, previous zonal crash models cannot better reveal the influence mechanism of zonal crash risk. More specifically, the proposed project will study the following research tasks: (1) to extract zonal traffic exposure information from big data sources, and to explore big data-driven methods for zonal travel pattern mining and fusion; (2) to develop zonal crash risk prediction models on the basis of deep learning methods to reveal the spatiotemporal evolution and propagation mechanisms of zonal crash risk; (3) to explore the multi-level contributing factors of zonal crash risk, and to identify the crash precursors of zonal traffic crashes; (4) to establish the relationship between the road network pattern, the travel pattern and zonal crash risk. Particular attention will be given to understand to what extent different travel patterns and road network characteristics affect zonal crash risk. The related findings will provide the fundamental basis for zonal safety regulation and optimized road network design under complex traffic environment. Based on the research results of this project, methods and guidelines will be developed to build big data-driven zonal crash models to improve the operation and safety performance of road network.