Abstract: Aiming at the injury severity of motorcycle crashes involving non-helmeted riders, various kinds of Bayesian spatiotemporal logistic models were established to systematically evaluate the comprehensive impact of driver, vehicle, road, and environmental characteristics. Based on relevant data from 5,447 crashes from 2 015 to 2 019, five kinds of models incorporating spatial, temporal, and spatiotemporal interaction effects were established and the Markov chain Monte Carlo method was used to carry out parameter estimation. Results show that the two-component mixed model comprehensively considering Leroux CAR spatial prior, temporal random walk, and spatio-temporal interaction effect achieves the best performance, achieving a classification accuracy of 86.74% and a 3% reduction in DIC value, which significantly outperforms other models and identifies rainfall as a significant risk factor at the first time. Further analysis reveals that older age, distracted driving, drug driving, high-speed driving, nighttime travel, and complex road conditions all significantly increase the severity of accidents, while low speed, urban roads, and certain distracted environments can mitigate risks to some extent. Research indicates that breaking through the assumption of spatial and temporal independence and introducing spatiotemporal interaction effects are of great significance for revealing complex risk patterns. The proposed model can offer a refined risk assessment tool for traffic management and provide a theoretical basis for formulating targeted safety strategies, such as strengthening education for elderly drivers and enforcing helmet wearing.

Key words: traffic and transportation engineering; accident severity analysis; motorcycle accidents; spatiotemporal comprehensive effect; Bayesian inference

摘要： 针对未佩戴头盔摩托车驾驶员的事故严重程度，构建了多种贝叶斯时空Logistic模型，以系统评估驾驶员、车辆、道路及环境特征的综合影响。基于2015—2019年5 447起相关事故数据，建立了包含空间、时间以及时空交互项的5类模型，并采用马尔可夫链蒙特卡罗方法进行参数估计。结果表明，综合考虑Leroux CAR空间先验、时间随机游走及时空交互效应的双分量混合模型表现最佳，其分类准确率达到86.74%，DIC值降低3%，显著优于其他模型，并首次识别降雨为显著风险因素。进一步分析发现，年龄较大、分心驾驶、毒驾、高速行驶、夜间出行以及复杂道路环境均显著增加事故严重性，而低速、城镇道路及部分分心环境则在一定程度上缓解风险。研究表明，突破空间与时间独立性假设并引入时空交互效应对于揭示复杂风险模式具有重要意义。模型可为交通管理部门提供精细化的风险评估工具，并为制定针对性安全策略（如加强老年驾驶员教育与头盔佩戴执法）提供理论依据。

关键词: 交通运输工程；事故严重程度分析；摩托车事故；时空综合效应；贝叶斯推断