基于脑电功率谱密度的作业人员脑力负荷评估方法

Method for operator mental workload assessment based on power spectral density of EEG

脑力负荷状态的准确识别对减少因作业人员无效脑力负荷导致的人因事故具有重要意义。针对人-机系统中作业人员脑力负荷客观评估问题开展了基于MATB-Ⅱ平台的3种不同脑力负荷水平下的航空情境实验,记录16名被试的NASA任务负荷指数(NASA-TLX)量表数据和脑电(EEG)信号,提出了一种基于脑电功率谱密度(PSD)和支持向量机(SVM)的个体脑力负荷评估方法。结果表明:随着实验设计脑力负荷水平增加,被试的主观脑力负荷得分显著提高(p<0.001),这表明该实验任务设计较好地诱发了低负荷、中负荷和高负荷情境。在此基础上,通过网格搜索法确定个体脑力负荷评估模型的统一优化参数,惩罚系数取3000,核函数参数取0.0001,模型测试正确率达到0.9665±0.0298,宏平均的受试者工作特征曲线下的面积(Macro-AUC)达到0.9910±0.0114。本文为作业人员脑力负荷状态的客观和准确评估提供了一种新的办法,为后期作业人员脑力负荷状态的实时判别提供模型基础。

The accurate recognition of mental workload levels is of great significance to reduce human accidents caused by operators with invalid mental workload.This paper focuses on the objective mental workload assessment of operators in human-machine system.An aviation situational experiment based on MATB-Ⅱwas carried out at three levels of mental workload.Sixteen subjects were asked to fill in the NASA-Task Load Index(NASA-TLX)scale and the Electroencephalogram(EEG)results during the experiment were recorded.By analyzing the collected subjective and physiological data,a subjectspecified mental workload assessment method was proposed using the Power Spectral Density(PSD)of EEG and the Support Vector Machine(SVM).The results show that the subjective mental workload scores increase significantly(p<0.001)with the increase of designed mental workload levels,indicating that the experimental design successfully induces different mental workload scenarios.Based on the rationality of the experimental design,the subject-specified mental workload assessment models are established,and the parameters of these models are optimized by grid search and then unified as the penalty parameter of 3000 and the kernel function parameter of 0.0001.The test accuracy reaches 0.9665±0.0298,and the area under Macro-Averaging receiver operating Characteristic curve(Macro-AUC)reaches 0.9910±0.0114.Thus,the models provide a new approach for the objective and accurate assessment of mental workload,providing a basis for the realtime discrimination of mental workload.