【作者】 张翼；

【导师】 王学民；

【作者基本信息】 天津大学 ， 生物医学工程， 2012， 硕士

【摘要】 随着生活节奏的加快、社会竞争的加剧，许多特殊行业由于工作压力过大而导致睡眠缺失，甚至是失眠，进而导致：觉醒程度降低，执行能力降低；注意力降低，工作效率下降；应急反应能力降低，对危险信号的觉察判断能力下降，事故发生率提高。因而实时检测睡眠缺失过程中的警觉度变化规律，进而动态调整工作任务和工作负荷，减小事故发生率，变得尤为重要。例如宇航员长期在太空活动，昼夜节律改变引起的生理节律失调，太空失重隔离幽闭环境引起的时间感缺失，超负荷工作强度引起的压力过大，都会导致睡眠缺失和失眠现象。因而通过宇航员睡眠缺失情况判断警觉度状态，进而动态调整其任务强度，可以保证任务顺利完成。本文在以往研究的基础上，设计36h睡眠剥夺实验，记录受试者睡眠缺失过程中的自发脑电、诱发脑电、主观向量表评测、按键反应等信号，并利用主观向量表、人工分期和睡眠剥夺时长进行信号标注；分析自发脑电中时域、频域、非线性特征，诱发脑电中CNV、P300、MMN特征及行为学特征，观测其在睡眠缺失下的变化规律，探索其生理意义；利用不同模式分类算法对脑电数据进行模式分类，实现警觉度自动分阶，并引入不同的特征降维方法，降低数据冗余度，提高运算速度。本文得到的结论主要有：在无工作负荷的前提下，睡眠缺失前12h，警觉度状态基本保持不变；睡眠缺失12h-24h，脑力资源不足，导致警觉度状态急剧下降，且枕部变化更为明显；睡眠剥夺24-36h，脑力资源影响与生物钟效应相互抵消，受试者警觉度状态保持基本稳定，处于第二平台期；随着睡眠剥夺时间的增加，个体间差异性逐渐增大；BP算法分类正确率为95.67%，易受局部最优解的影响，SVM算法分类正确率为99.17%，效果更佳；FDR导联优化可在保证分类正确率的前提下，降低特征维数，PCA特征优化特征维数降低幅度更大，但分类正确率不高。更多还原

【Abstract】 With the acceleration of pace of life and the intensification of social competition, people whowork in specific industry have too much work pressure to sleep well, which leads to induction ofarousal level, executive capacity，attention，work efficiency，emergency response capacity,judgment of danger signal and increase of accident rate. Therefore, it is important to detect alertlevel in time and actively adjust tasks and workload. For example, astronauts usually work inspace for a long time. So, disorders in astronauts’ circadian rhythm caused by special changes inday and night of space and the loss sense of time cause by weightless environment of space alsoand the excessive pressure caused by overload work intensity will lead to astronauts’ sleepdeficiency and serious insomnia. Thus, it is important to actively adjust task intensity to ensuresuccessful completion by estimating astronauts’ alertness level.In this paper, on the basis of previous studies, first, we designed a36hour sleep deprivationexperiment. Second, we collected and analyzed spontaneous EEG, evoked potential, subjectivescale, and keystroke response. And we marked signals with artificial staging by sleep experts,subjective scale and length of sleep deprivation time of. Third, we perform the comprehensiveanalysis of spontaneous EEG in temporal, frequency, spatial domains and the nonlinear features,combine with analysis of P300, MMN, features of evoked potential land the behavior featureswe explore the variation regularity as the increase time of sleep deprivation and the physiologicalsignificance of the features. Last, multiplicate classification algorithms were used to realize theauto stage of alertness. And multi-modes character optimizations were used to achieve dimensionreduction and improve the speed of classification.The conclusions we get are as follows: if there has no working load, the alertness statusremains unchanged in the first12hours of sleep deprivation; among12to24hours of sleepdeprivation, the mental resource is not enough, which leads to alertness level dramaticallydeclined., and occipital area changed more apparently; among24to36hours of sleep deprivation,mental resource and biological clock effect offset, and alertness status maintains in secondplatform stability; BP algorithm is susceptible to optimal solution and its correct rate is95.67%, Incontrast, SVM algorithm has better effect with correct rate99.17%; FDR could reduce thedimension of features with high accuracy rate. PCA could reduce the dimension of features, butthe correct rate is lower.更多还原