基于多维信息特征分析的驾驶人认知负荷研究

准确评估驾驶人认知负荷水平，对于深入研究驾驶人行为特性，改善驾驶安全性具有重要意义。现有的驾驶人认知负荷分类方法，大多基于心电、脑电等生理信息和车辆信息，由于特征选择上的单一性，导致驾驶人认知负荷分类模型的分类精度不高。设计基于跟驰场景的不同认知负荷N-back次任务试验，通过采集受试者的生理信号和车辆信号，结合NASA_TLX主观评分和机器学习算法，提出了基于多维信息特征融合的驾驶人认知负荷分类方法。研究表明：基于生理信息和车辆信息的多维信息特征认知负荷分类方法，其精度显著高于传统的基于生理信息的认知负荷分类方法，以多维信息特征为输入，随机森林法以其稳定性好、抗过拟合能力强的特点，表现出优异的分类效果，相比神经网络和支持向量机，具有最高的平均分类精度。

Driver Cognitive Loads Based on Multi-dimensional Information Feature Analysis

The accurate assessment of driver cognitive loads is of great significance for the study of driver behavioral characteristics and improving driving safety. Most existing driver cognitive load classification methods are based on physiological information, such as ECG and EEG signals, or vehicle information. The accuracy of driver cognitive load classification methods is poor based on the selection of individual features. This paper proposes an N-back secondary-task experiment with different cognitive loads based on a car-following scene. A method for driver cognitive load classification based on multi-dimensional information feature fusion was developed by collecting both the physiological signals of subjects and vehicle signals, and combining them using the NASA_TLX subjective scoring and machine learning algorithms. The results demonstrate that the proposed classification method for cognitive loads based on multi-dimensional information characteristics, including physiological information and vehicle information, significantly outperforms traditional classification method based on physiological information alone. Compared to neural networks and support vector machines, the random forest method achieved higher average classification accuracy based on its stability and ability to avoid overfitting when multi-dimensional information characteristics are used as inputs.