基于车辆行驶状态的险态驾驶动态感知及协同诊断模型研究

With the increasing number of automobile, nowadays the transportation safety becomes a extremely important issue for scientific research. The status of driver plays the most significant role in the safety of road traffic due to driver is the core factor in road traffic system. Thus, study of risk driving behavior of the driver becomes a vital issue for road traffic safety. It is selected as the main context of this project which focuses on the development of a dynamic diagnosis model for risk driving behavior under zero distraction condition. The risk driving model whose input was the risk driving indexes was set up in complex road conditions. It reveal the mapping rules of the risk driving and vehicle running status; Decomposition of the high SNR road ,vehicle and driver status information form the vehicle running status, drawing on the instantaneous frequency analysis methods to extract frequency domain information, the time domain, spatial information to build the high-dimensional feature vector for risk driving, kernel principal component analysis for feature reduction in low-dimensional space-cycle assessment of the risk-state driving state; a dynamic risk driving synergistic diagnosis model is developed by coupling the forward method based on positive analysis of risk model and backward methods based on signal processing of vehicle running status. Following this, one novel risk driving model for positive analysis and a new verification model for driving characteristics signal process are developed to implement the risk driving synergistic diagnosis system. It is known that the driving characteristics during the risk driving are dynamic and the driver model is also time-varying under different stages in the close-loop system of driver, vehicle and road. It causes the time-varying problem in risk driving diagnosis. To tackle this problem, a coupled dynamic synergistic diagnosis model is developed based on the forward and backward methods used in diagnosis analysis. The Research projects provide a theoretical method for the risk driving degree of precision division and non-contact-line monitoring and enable technical support to promote the process of practical the risk driving diagnostic techniques.

驾驶行为是影响交通安全最活跃的因素，如何对险态驾驶进行动态诊断已经成为世界各国科学家的研究热点。本项目拟以车辆行驶状态为对象，研究险态驾驶的动态感知及诊断方法。即在复杂路况条件下，建立以险态驾驶指标为输入的险态驾驶模型，揭示险态驾驶与车辆行驶状态的映射法则；从车辆行驶状态中分解出高信噪比的道路信息、车辆信息、驾驶员状态信息，借鉴瞬时频率分析方法提取频域信息，并与时域、相位和空间信息相结合，构建险态驾驶状态的高纬特征向量，采用核主成分分析方法进行特征约简，在低维空间中建立险态驾驶状态全周期评估方法；研究险态驾驶模型正向推演与车辆行驶状态反向分析协同耦合的险态行为动态诊断模型，解决险态驾驶演化中模型和特征均具有时变特性的难题，在对驾驶员"零干扰"的前提下，实现险态驾驶的动态诊断。项目研究为险态驾驶程度的精确划分和非接触式在线监测提供理论方法和使能技术支撑，推动险态驾驶诊断技术的实用化进程。

项目以车辆行驶状态信息为对象，针对险态驾驶研究中理论模型和信号特征时变特性，以闭环驾驶模型参数动态辨识为桥梁，基于“正向分析”与“反向分析”相结合的思想构建完成了模型动态推演信号表现/监测信息更新模型的险态驾驶协同耦合监测诊断模型。其主要研究成果包括：.1) 针对人—车—路闭环驾驶系统是一个复杂的系统，险态驾驶行为的发展是一个动态的过程，不同阶段描述其行为的数学模型和险态特征是变化的特点，提出了利用模型/信号协同耦合的险态驾驶方法。.2) 针对预瞄神经网络驾驶员模型没有考虑道路突变、驾驶员应急反应等极限工况引起的车辆动力学改变的缺点，提出并构建完成驾驶过程中的注意力转移模型和面向险态驾驶的神经网络自适应调节模型，以此为基础挖掘出险态驾驶指标与驾驶员模型参数之间的联系。.3) 针对传统遗传算法最优染色体种群信息丢失和早熟等问题，融合了量子编码与实数编码的优点，提出并完成了基于实数量子编码和混沌变异量子遗传的险态驾驶模型参数辨识方法，实现闭环驾驶员模型的参数辨识，为基于模型/信号耦合协同的险态驾驶监测诊断模型的动态更新提供了关键支撑条件。.4）针对驾驶行为、道路、车辆信息在车辆行驶状态数据中频带分布特性，结合闭环驾驶系统非平稳、非线性的特性，提出基于近似熵的有效IMF选取方法，以提取车辆行驶状态数据中蕴含驾驶行为信息的有效分量，消弱道路信息分量。 .5) 制定了一套以驾驶员反应时间和注意力指标为主，主观评测疲劳指标为辅的疲劳状态的客观评价方法。用正交实验设计了一套全面合理的疲劳驾驶实验。构建了疲劳驾驶计算样本的生成流程，并且对基于模型/信号协同耦合的疲劳驾驶行为诊断效果进行了分析验证，结果表明项目提出的协同耦合诊断模型具有动态、自适应跟随性的特点，能够从正常驾驶数据中通过模型推演出险态驾驶的部分特征，对丰富险态驾驶特征时空模式，减少对难以获取的疲劳驾驶样本的依赖有积极意义。

内容获取失败，请点击重试