【作者】 王娆芬；

【导师】 王行愚；

【作者基本信息】 华东理工大学 ， 模式识别与智能系统， 2012， 博士

【摘要】 近几十年来,随着自动控制技术的蓬勃发展,人类逐渐成为自动控制系统的监督者和决策者。相应地,人类操作员拥有了更高的权限并担负了更重的责任。在高安全性要求的自动控制系统中,特别是交通运输(民航、铁路、船运等)和过程工业(核电厂、化工厂等),操作员的失误可能会引起严重的事故。为了避免此类安全事故,研究者们提出了根据操作员功能状态(OFS)调整系统控制策略的方法,即自适应自动控制。通过操作员功能状态估计,可以预测操作员的脆弱或有风险的工作时刻,从而利用自适应辅助系统提醒系统或操作员采取相应措施,避免事故发生。因此,准确的估计操作员功能状态是成功运用自适应自动控制的关键。本文针对操作员功能状态估计问题,首先设计了过程控制实验环境,并采集了单任务多级任务负荷状态下的操作员生理电信号；针对操作员脑电和心电信号的特点,进行了去干扰处理并提取了特征信号,找出了与操作员功能状态显著相关的特征；最后采用智能建模与优化方法,建立了操作员功能状态模型。本文的主要研究成果如下：(1)目前对操作员功能状态的研究主要集中于航空、航海、汽车驾驶等应用领域,对过程控制操作员的功能状态研究较少。本文采用密闭舱空气管理系统软件模拟过程控制环境,设计了单一任务类型,任务难度变化的操作员功能状态的研究实验。通过日本光电公司EEG1100型生物电信号采集仪,采集了实验过程中被试的脑电、心电和眼电信号。采用视觉模拟尺度方法,记录了被试的疲劳程度、紧张程度和努力程度等主观脑力负荷评价。而被试任务性能数据由密闭舱空气管理系统软件自动记录。(2)脑电信号是反映操作员功能状态变化最重要的生理电信号。脑电信号很容易受到各种噪声和伪迹信号的干扰,其中眼电伪迹的干扰尤为突出。因此,本文提出了一种自动眼电去除方法。首先采用独立成份分析分离原始脑电信号,计算了独立成份的多种特征,并运用模糊c均值识别脑电成份和眼电成份,最终得到了去眼电干扰后的脑电信号。结果表明,该方法能够消除个体差异性,正确去除了每位被试脑电信号中的眼电伪迹,是一种有效的自动眼电伪迹去除方法。(3)研究者们对脑电信号的研究主要集中于频域分析,通常计算脑电信号的功率谱,然后采用传统的频域设置划分四个基本节律。但这种频域划分是人为设定的,并不符合基本节律的频率分布情况。因此,本文采用经验模态分解方法,将脑电信号按照不同的时间特征尺度从高频到低频逐渐分离成与基本节律频域相对应的本征模态分量。随后,对本征模态分量进行功率谱分析,计算出本征模态分量的频率和能量特征。最后,定量分析了这些特征与操作员功能状态之间的关系,选取出了反映OFS变化的显著特征。分析结果表明,基于经验模态分解的脑电信号分析方法是行之有效的。(4)采用小波包分解方法,对多任务OFS实验中过程控制操作员的心率变异性(HRV)进行分析,研究了操作员HRV的低频、中频、高频能量、中高频能量比和小波包熵等五项特征的变化,分析了这些特征与操作员任务性能评价、主观测量评价以及任务负荷之间的相关性,从而找出了能够反映被试功能状态的HRV特征。研究结果表明,由于个体差异性,不同被试的5项HRV特征中与操作员的主、客观性能评价之间具有显著相关性的特征不同,可针对不同被试选择相应的特征作为描述操作员功能状态的指标。(5)为了准确估计操作员功能状态,采用自适应模糊神经网络模型(ANFIS)建立了操作员功能状态模型,并提出了一种交叉粒子群优化算法(PSOC)来优化ANFIS的参数。在PSOC算法中,原种群在每次迭代结束后与一个增加的辅助种群进行交叉操作,以维持原种群的多样性,从而提高了算法的寻优能力,避免了早熟收敛。实验结果表明,基于PSOC-ANFIS的操作员功能状态模型能够较好的描述操作员生理变量与功能状态之间的复杂非线性关系。(6)在基本差分进化算法中引入蚁群觅食过程中正反馈的思想,提出了一种控制参数自适应选择的差分进化算法—蚁群搜索差分进化算法(DEACS)。在DEACS算法中,每个个体变异、交叉时的控制参数都是由上一次迭代的寻优性能决定。典型测试函数结果表明,提出的DEACS算法寻优性能优于其它3种方法,其收敛速度更快。将提出的DEACS算法用于OFS模型的参数优化,由仿真结果可知,DEACS能够有效降低算法控制参数的设置对不同被试模型精度的影响,并且所建立的OFS模型能够很好的反映过程操作员的心理负荷,适用于操作员功能状态的实时在线估计。更多还原

【Abstract】 During the booming development of automation technique in decades, human becomes the monitor and decision maker in the automatic system. Consequently, it brings higher authority and broader ranges of responsibilities to individuals. The operator’s performance degradation may be the reason of some serious disasters, particularly in safety-critical applications such as public transportation (aviation, railway, shipping, etc) and manufacturing industries (nuclear and chemical plants). In order to solve the above problems, researchers proposed adaptive automation in which the control tasks can be relocated according to the operator functional state (OFS). The assessment of OFS would help predict the operator’s periods of high operational risk. And then the adaptive aiding is applied so as to avoid disasters which are caused by operator performance degradation. Therefore, the accurate assessement of OFS is the key to apply adaptive automation successfully. In the dissertation, the experimental environment of process control is designed, and the psychophysiological measurements of operators under multi-level of task loads are sampled. According to the characteristics of electroencephalogram (EEG) and electrocardiogram (ECG), the author eliminates the artifacts of EEG, extracts EEG and ECG features, and indicates the salience features significantly correlating to OFS. Subsequently, intelligent modeling and optimization methods are used to establish the corresponding OFS models. The main results in this dissertation can be summarized as follows:(1) Currently, the operator functional state is studied in the applications cases, such as the aviation, navigation, driving, etc, but rarely in the process control. An Automation-enhanced Cabin Air Management System (AutoCAMS) is used to simulate the process control environment, and the multi-level taskload experiment with single task is designed on this environment. The EEG, ECG and electrooculagram (EOG) signals of the subjects are recorded by EEG 1100. The visual analogue scale method is adopted to measure the subjective evaluations on fatigue, anxiety and effort while the task performance data are automatically recorded by AutoCAMS.(2) EEG is the most important psychophysiological signal that reflects the operator functional state. But EEG signal is easily interfered by noises and artifacts, especially by ocular artifact. An automatic ocular artifact suppression method is proposed. Independent component analysis is used to separate the original EEG signals first. Then, five features of independent components are calculated. EEG and ocular artifact components are recognized through fuzzy c-means clustering and finally clean EEG are obtained. The result shows that the method can remove the ocular artifact from EEG signal effectively and is a reliable ocular artifact suppression method.(3) The empirical mode decomposition (EMD) which is developed specially for analyzing nonlinear and nonstationary signals is employed for EEG signal analysis. The segmented EEG data are analyzed via EMD. The Welch method is applied to estimate power spectrum density of intrinsic mode functions (IMFs), whose features including peak power, peak frequency, gravity frequency, absolute power and relative power are calculated. Then the correlations between EEG features and OFS are analyzed. Finally, the salience features for each subject are obtained.(4) Wavelet packet transform is introduced to analyze the heart rate variability (HRV) of process control operator in the multi-task OFS experiment. Five features, including the energy of low frequency (LF), mid-frequency (MF), high frequency (HF), MF/HF ratio and wavelet packet entropy, are investigated. Through the study of their correlations with the operator’s primary task performance, subjective measurement and task load, the HRV features that can be used to assess the operator’s mental load are obtained. The results indicate that some of the five features have significant correlations with the operator functional state and can be adopted as important indices for representing OFS.(5) In order to estimate and predict OFS, adaptive network based fuzzy inference system (ANFIS) is employed to build up the operator functional state model, whose parameters are optimized by using a proposed particle swarm optimization with crossover (PSOC). In the PSOC algorithm, the original swarm implements the crossover with an introduced auxiliary population after each iteration so as to preserve the diversity of the original swarm. The operation improves the optimizing capability and avoids premature convergence. The empirical results illustrate that PSOC-ANFIS based OFS model can describe the complicated nolinear relationship between the psychophysiological variables and the functional state of operators.(6) Based on the idea of positive feedback in the foraging process of ant colony, a novel differential evolution algorithm with ant colony search (DEACS) is proposed, whose control parameters are selected adaptively. When mutation and crossover, these control parameters for each individual are determined according to the optimization performance at the last iteration. The results of typical benchmark functions show that the proposed DEACS algorithm outperforms three other algorithms and converges more rapidly. The DEACS is applied in optimizing the parameters of OFS model. It is indicated by the simulation results that DEACS can reduce the influence of the parameter settings on the model accuracy of different subjects. The OFS model can provide good representation of the mental load and is applicable of realtime online assessment of OFS.更多还原