基于机器学习的车辆作用下桥梁局部结构响应预测理论

The bridge local structure, which is typically represented by bridge deck, is mainly analyzed and designed by vehicle load model (defined by standard axle load). While in practice, the bridge local structural response has complex characteristics, showing remarkable sensitivity to the temporal and spatial distribution of vehicle load, which results in significant variability of response. It is difficult to cover such variability with the vehicle load model adopted by the current design code. Although the random traffic simulation technology has gained substantial achievements in the recent year, the accuracy of the existing methods can still not meet the analysis of the bridge local structural response. Therefore, based on the artificial intelligence method, this research explores the data-driven intelligent analysis theory of structural response, in a bid to solve this problem. Firstly, behavior rules of the vehicle in various driving environments are studied. On the basis of the multi-agent modeling, a micro-scale traffic flow simulation method, which considers complex driving characteristics and short time variability, is proposed. Secondly, with the help of accurate FEM of local structure, we can obtain adequate data of load distribution and it corresponding local response. Furthermore, research on machine learning model construction and framework optimization is executed to implement the structured output of multiple local structural responses. Finally, with reference to the traffic data from the monitoring section, the micro-scale traffic flow is simulated corresponding to the concerned sections. Such traffic flow is further utilized as the estimator of the value range of local structural responses. On the foundation of the above research, quick warning and assessment of multiple responses in multi-zone scenario are sought to be realized in engineering applications. The achievements in this research will form instrumental theories and approaches for prediction of bridge local structural response, and provide fundamental data for the further revealing of the mechanism of local damage.

以桥面结构为代表的桥梁局部结构在设计评估等过程中需重点考虑车辆荷载的作用。实际运营中此类结构对车辆时空分布敏感，且车辆轴载变异显著，造成响应异常复杂。目前规范中的车辆荷载模型（即轴载模型）难以有效考虑这些因素；现有车流模拟技术也无法满足分析的精度要求。本项目基于人工智能方法，探索数据驱动的荷载响应智能分析理论，解决桥梁局部结构响应预测中的基础科学问题。首先，研究车辆驾驶行为规律，基于多智能体建模提出考虑复杂行驶特性和短时变异性的微观车流模拟方法。其次，基于局部结构的精确有限元模型，获取荷载分布及与其相匹配的结构响应，开展机器学习模型构建及其架构优化，实现多响应的结构化输出。最后，基于荷载监测断面的车辆数据，产生响应关注断面的微观车流，预测局部响应的量值范围，实现多区域、多响应的快速预警和评估。研究成果将形成桥梁局部结构响应预测的基础理论和方法，也为进一步揭示局部损伤机理提供数据支撑。

桥梁汽车荷载具有显著的地区特异性和高度的时空变异性；目前基于简化抽象的汽车荷载模式仍存在明显的应用局限性，其重要原因在于难以充分描述海量交通荷载数据中蕴含的汽车荷载演化特征；完全基于物理模型的理论方法在揭示加载车流演化特性方面有一定难度。..研究结合计算机视觉中的目标识别和目标追踪技术，构建了融合称重数据和轨迹数据的桥梁加载车流监控系统，用于实时汽车荷载分布的精确重现；提出了桥梁加载车流的流量-密度-速度-荷载的理论模型，在微观交通流模型中引入荷载信息；考虑到未来交通模式的改变，整合人类驾驶车辆和自动驾驶车辆的移动规则，实现了混合加载车流的模拟。数据采集技术和模拟方法为后续桥梁加载车流数据的大量获取和响应计算提供了基础工具。..研究借助大数据聚类技术，基于通过断面的车辆时间序列及其相关统计参数，对桥面的交通荷载状态进行分类，揭示了车流数据在荷载影响下的交通模式特征；考虑交通与荷载演化的多个维度，进行差异化的加载车流模拟，实现了复杂车辆荷载作用下的桥梁结构响应分析；分析了狮子洋大桥、张靖皋长江大桥和燕矶长江大桥等超大跨径悬索桥的汽车荷载响应，明确了规范模型的适用性和保守程度，为汽车荷载设计标准的确定提供理论支撑。..研究利用深度学习方法，构建历史荷载响应记录或断面车流统计数据与未来结构响应的映射关系，直接采用局部荷载响应信息或断面车流信息对桥梁汽车荷载响应进行快速预测，为未来桥梁运营中基于断面车辆监测信息的荷载响应水平预测和预警提供了模型和方法支持。..研究成果实现了基于机器学习方法构建车辆荷载作用下桥梁结构响应预测理论的总体目标，为桥梁长期运营中汽车荷载的响应预测、模型设计和交通控制提供了关键的基础模型和科学方法。