基于雪崩动力学模型的高解析度生理学驾驶疲劳检测方法研究

In order to ensure the safety of drivers, the uncertainty of external environment and the complexity of the fatigue evolution need to be considered to improve the accuracy and reliability of the fatigue detection method. In this project, the EEG-based dynamic analysis method that does not depend on initial conditions and external factors will be developed. Firstly, the dynamic analysis method based on functional brain networks is used to extract the physiological features from the EEG signals during driving. Secondly, based on the theory of self-organized criticality (SOC), a physiological model for snow avalanche dynamics will be constructed to investigate the SOC of fatigue evolution. Particularly, the uncertainty and complexity of avalanche behavior in critical state will be analyzed and the mechanism of avalanche dynamics adapted to the dynamic brain network analysis will be utilized to develop the critical point detection method that is robust. Moreover, during the evolution of the model, the irreversibility of the functional connectivity will be investigated to change the collapse rule to adjust the avalanche behavior (in the critical state) to ensure that the detected fatigue states are real fatigue state. Finally, a complete method that can detect different levels of fatigue more reliably will be proposed. The EEG signal processing method will assist to calibrate driving behavior data collected simultaneously to accomplish the feature classification to verify the validity and correctness of the fatigue criticality detection algorithm.

面对驾驶安全对疲劳检测准确性和可靠性的需求，外界因素的影响与内部疲劳演化模式的复杂性是需要考虑的问题。本项目拟借助基于脑电（EEG）处理的动力学演化方法实现不依赖于初始条件和外界参数控制的动态疲劳标定。首先，进行基于EEG功能脑网络的动态分析，提取驾驶过程中的生理学特征。然后，通过自组织临界系统的理论和方法，利用基于生理学特征的雪崩动力学模型，发现疲劳演化的自组织临界性。尤其研究适应于动态脑网络分析的雪崩动力学机制、临界状态雪崩行为的不确定性和复杂性分析，发展鲁棒的雪崩动力学临界点标定方法。更进一步，在模型的演化过程中，改变倒塌规则，探索功能连接的不可逆性对雪崩行为（临界点处）的修正，使标定的疲劳状态与真实的疲劳状态对应。最终实现一套能准确检测不同疲劳水平的方法。通过与驾驶行数据连用，指导其对不同疲劳特征类别作标签完成分类训练，以验证疲劳临界检测算法的有效性和正确性。

疲劳是一种主观的不适应感，容易导致完成原来所从事正常活动或工作能力的降低。对于长时间连续驾驶的人而言，疲劳程度不断积累，生理和心理机能都会产生某种程度的失调，容易引起交通事故。脑电（EEG）记录的是头皮上电极处脑区的电活动，这些电活动反映电极周围区域底层大量神经元电活动的总和。本项目利用EEG对脑部活动解析的毫秒级时间分辨率优势进行驾驶疲劳标定。在驾驶过程中建立了动态的功能脑网络，计算了基于网络特征的疲劳度量指标，发现疲劳过程中脑网络的小世界特性变差，疲劳度量指标呈阶段性变化。在疲劳度量指标上进行脑网络雪崩动力学推演，发现驾驶疲劳过程中存在自组织临界特性，通过雪崩行为的统计完成了疲劳临界点的标定，发现在疲劳的临界点附近，驾驶行为数据有异常的波动。证明了方法有助于驾驶疲劳的准确检测，防止疲劳导致的异常行为而引发的交通事故。