基于机器学习的飞行员脑力负荷评估研究进展

Research Progress of Mental Workload Assessment Based on Machine Learning in Pilots

基于机器学习实现对飞行员脑力负荷的实时准确评估,对减少人为飞行事故具有重要意义。对飞行员脑力负荷评估的生理测评法进行综述,重点介绍了不同任务下基于脑电、心电或多生理信息应用机器学习对飞行员脑力负荷进行评估的方法。通过对比分析发现,在空间、成本或其他资源受限的情况下,脑电是飞行员脑力负荷监测的首选生理信号;当数据量足够大时,相比传统机器学习,构建适合生理信号特点的深度学习模型对脑力负荷进行评估精度更高;基于迁移学习方法构建脑力负荷评估模型,可以弱化生理数据的日间变化以及受试者间生理数据的差异对模型的负面影响,提高脑力负荷评估模型泛化性和准确率。

Real-time and accurate assessment of mental workload of pilots based on machine learning is of great significance for reducing flight accidents caused by human errors.In this paper,the electroencephalogram(EEG),the electrooculogram(EOG)and the electrocardiography(ECG)indexes in physiological measurement methods for mental workload of pilots were reviewed.The application of machine learning based on EEG,ECG or multi-physiological indexes to evaluate pilots'mental workload in different tasks was introduced in detail.The comparative analysis showed that EEG was the preferred physiological signal for monitoring pilots'mental workload under limited resources.In the case of sufficient data,the deep learning model,which was suitable for physiological signal characteristics,could provide higher accuracy than the traditional machine learning in mental workload assessment.The mental workload assessment model based on the transfer learning method could weaken the negative impact of the day-to-day variability and the inter-subject differences of physiological data.This method could improve the generalization performance and accuracy of the assessment model.