基于驾驶风格主客观评估的车辆纵向动态规划及控制研究

Driving style evaluation and its recognition are an important part of the field of intelligent transportation, traffic safety, and automobile testing. This study is based on questionnaire surveys to obtain factors of personal traits such as demographics, personality, sensation-seeking, and the risk altitude. Meanwhile, driving style representative features including vehicle driving status indicators, vehicle driving stability indicators and driving operation indicators are obtained through driving simulation experiments. Structural Equation Model is then utilized to explore the relationship among driver personal traits, the driving environment, the driving style, and driving style representative features. Driving style representative features are further filtered based on the results of the Structural Equation Model, and the driving style recognition model is realized by the Gaussian Process Dynamic Model. Then the personalized constraints for vehicle states and control inputs are determined based on the driving style. With considerations of the driving comfort, energy saving and safety as the optimization goal, the Pontryagin's Minimum Principle is utilized to realize the optimal dynamic planning for the vehicle longitudinal velocity. In addition, the cooperative control strategy of the multiple control systems of the chassis is designed, which is based on the stochastic game theory, such that the overall dynamic performance of the vehicle would be maximized. This project aims to make a preliminary exploration of the human-centered design for intelligent driving systems, it also provides theoretical guidance for harmonious interactions with vehicle assistance driving or automatic driving.

驾驶风格评估和辨识是智能交通、交通安全和汽车测试领域普遍关注的重要内容，本研究基于问卷调查获取包括人口统计学特征、人格特征、感觉寻求倾向、风险态度的驾驶员个体特征变量，通过模拟驾驶实验获取包括车辆行驶状态指标、车辆行驶稳定性指标和驾驶操作指标在内的驾驶风格表征量，采用结构方程模型探究驾驶员个体特征因素、驾驶环境、驾驶风格以及驾驶风格表征量之间的关联关系，并对驾驶风格进行评估和分类；基于结构方程模型筛选驾驶风格表征量，采用高斯过程动态模型方法构建驾驶风格辨识模型。基于驾驶风格确定个性化的车辆状态约束和控制约束，以车辆乘坐舒适性、节能性和安全性为优化目标，采用庞特里亚金最小化原理实现车辆速度的最优动态规划，并基于随机博弈理论制定底盘多控制系统的协同控制策略，使车辆总体动力学性能最优。该研究将初步探索以人为中心的智能驾驶系统设计方法，为车辆辅助驾驶或自动驾驶下的人机和谐交互提供理论指导。

本课题基于自然驾驶大数据和驾驶员问卷调查，利用因子分析方法、结构方程模型等统计学方法，采用主客观综合评估方法研究驾驶员人口统计学特征、驾驶员心理因素与危险驾驶行为的关联关系，即驾驶经验对高风险驾驶行为的影响中，感觉寻求和风险认知起到的中介作用。基于主观问卷数据，结合客观碰撞发生率，对驾驶风格进行评估和分类，结果显示对高风险驾驶员的甄别率达到90%以上。针对高速公路稳定跟驰事件，从跟驰安全性和跟驰稳定性采用主成分分析方法筛选驾驶风格特征，并构建基于卷积神经网络和高斯过程分类方法识别驾驶风格，然后基于强化学习方法构建自适应驾驶风格的车辆纵向跟驰决策模型，模型结果显示：所提出的跟驰行为决策过程类似于人类驾驶员，且表现出了对应类别的驾驶风格特征。最后，面向道路纵坡场景，基于鲁棒反演滑模方法实现车辆纵向速度的鲁棒跟踪控制，仿真结果表明所提出的模型具有较好的抗干扰能力及优异的跟踪精度。此外，在本基金的资助下，还进一步探索了基于Koopman数据驱动的人机协同驾驶模型，以及抗干扰的鲁棒轨迹跟踪控制方法在其他类型车辆（如无人水面车辆）的应用。本课题的研究将初步探索“以人为中心”的智能驾驶系统设计方法，为设计“宜人性”的车辆辅助驾驶或自动驾驶下的人机和谐交互提供理论指导。

内容获取失败，请点击重试