【作者】 杨少增；

【导师】 张建华；

【作者基本信息】 华东理工大学 ， 控制科学与工程， 2014， 博士

【摘要】 从技术可行性、经济性和安全性的角度出发,人类已经意识到,以彻底取代人类操作员为目的的完全自动化的实现正变得越来越困难,人类操作员仍将继续长期存在于各种系统中。因此,对人机交互系统的研究成为了自动化技术发展的另一个分支。在高安全性要求的人机交互系统中,微小的事故往往可能造成巨大的损失。其中,由操作员功能状态(Operator Functional State, OFS)失效造成的其所承担的任务无法有效完成是导致各种事故发生的重要原因。为此,一些学者提出了适应性自动化(Adaptive Automation, AA)的概念。在AA系统中,通过对OFS进行估计和预测,一旦发现操作员出现高风险状态,立刻对其任务负荷进行调整或提醒操作员采取一定措施,以使操作员所承担的任务要求与其当前的状态两者相匹配。在AA系统的构建中,建立可以对OFS进行精确估计和预测的模型是一个关键问题。本文通过对操作员电生理数据进行采集和分析,使用模糊建模方法建立了基于操作员电生理数据的OFS估计和预测模型,完成了如下的研究工作：(1)使用aCAMS (automation-enhanced Cabin Air Management System)软件对5名操作员被试进行了多任务负荷仿真实验,并采集了被试在不同任务负荷下的电生理数据及其任务性能数据。对电生理数据进行滤波、功率谱分析、数据平滑等预处理,通过相关性分析,得到了3个EEG特征作为OFS模型的输入,使用被试的任务性能数据作为OFS的量化指标和模型的输出,为后续的模糊建模工作提供了数据集；(2)使用粒子群优化(Particle Swarm Optimization, PSO)算法对OFS模糊模型的参数进行估计。在该过程中,比较了PSO算法与增量型PID控制器的内在联系,将二者结合提出了一种新的搜索方式,开发了一种增量型PID控制的PSO算法(IPID-PSO)。为了检验该算法的有效性,首先在7个基准函数的优化问题中进行了测试,发现对于多峰函数,IPID-PSO算法在优化效果上有优于其它3种PSO算法的表现。接着,将IPID-PSO算法应用于OFS模糊模型的参数估计中,所得的模糊模型实现了对OFS的良好估计；(3)使用Wang-Mendel (WM)方法进行OFS模糊建模。在基于WM方法进行模糊模型设计时,分析了高斯隶属函数的宽度参数σ对模糊模型抗噪声能力的影响。为了确定最优的σ值,在使用聚类法进行论域划分时,设计了一种混合高斯隶属函数,将对σ的确定转化为对相邻隶属函数重叠度δ值的确定。为了得到最优的δ值,首先比较了使用不同δ值的模糊模型在4个数据集预测中的性能,得到了适用于不同含噪水平数据的最优δ值,说明了进行最优δ值选取的普遍意义。接着,将同样的比较应用于OFS模糊建模中,取得了类似的结论,并实现了对OFS的良好估计。同时,比较结果显示,使用聚类方法加混合高斯隶属函数的论域划分形式在OFS模糊建模中表现出了优于传统均匀论域划分形式的性能；(4)为了实现AA系统的功能,即对操作员高风险状态的预防,使用了OFS预测的概念,据此建立了OFS动态预测模型,并进行了仿真验证。对OFS预测模型的结构进行了估计,结果显示,采用WM方法的一阶模糊模型可以获得最优性能。为了提高对高风险OFS的有效预测率,用多模型策略代替了单模型策略,并建立了多个WM模型用于OFS预测。为了验证该预测模型的有效性,设计了一种自适应任务分配策略,对基于该自适应任务分配策略的人机交互控制系统进行了仿真。仿真结果显示,在该人机交互系统中,OFS得到了有效的调节,操作员的任务性能水平得到了显著改善,同时,操作员高风险状态出现的次数大大减少,从而大幅度提高了人机系统的安全性。更多还原

【Abstract】 From the technical feasibility, economy and safety point of view, mankind has realized that the implementation of fully automation with the purpose of completely replacing the human operator is becoming increasingly difficult, thus human operators will continue to persist in various kinds of system. Therefore, the study of human-machine (HM) interaction system becomes another branch of automation technology development. In safety-critical HM systems, trivial accidents often may cause huge losses. Among them, operator’s assigned task failure caused by operator functional state (OFS) breakdown is found one of the main reasons of such accidents. To prevent accidents, the adaptive automation (AA) concept was proposed. In AA systems, the OFS are evaluated and predicted. Once a high-risk OFS is detected, either the operator’s task will be reallocated or the operator will be required to do some adjustments to make the two match each other. In the realization of AA systems, building the exact model used for OFS evaluation and prediction is a key problem. After collecting and analyzing operators’physiological data, in this paper, the fuzzy modeling method based on operators’ physiological data was employed as the main paradigm to derive the OFS evaluation and prediction model. The main contributions are as follows:(1) The aCAMS software was used to simulate the multi-task environment and5subjects participated in the experiments. Physiological data and the task performance data were collected while operators were working with different tasks. The raw physiological data were preprocessed with filtering, power spectrum analysis, data smoothing and so on. After that, by using the correlation analysis technique,3EEG features were selected as the inputs of the OFS model. The operator task performance data were used to quantify the OFS and were treated as the output of the OFS model. This work prepared the dataset used in the following OFS fuzzy modeling.(2) The particle swarm optimization (PSO) algorithm was employed to estimate the OFS fuzzy model’s parameters. During this process, the PSO algorithm and the incremental PID controller were compared and their intrinsic links were analyzed. The two were combined and a new search strategy was proposed and an incremental-PID-controlled PSO (IPID-PSO) algorithm was developed. To verify the usefulness of the algorithm, the IPID-PSO was firstly applied in the optimization of7benchmark functions. The results showed for multi-modal functions, IPID-PSO performed better than other3PSO variants on final results. Then, the IPID-PSO was applied to the estimation of the OFS fuzzy model parameters. The derived fuzzy model can well evaluate the OFS.(3) The Wang-Mendel (WM) method was employed for OFS fuzzy modeling. In the WM-based fuzzy model designing, the relationship between the Gaussian membership function’s a parameter and the fuzzy model anti-noise ability was analyzed. The clustering method was employed for the domain partition. To determine the best a value, by using an hybrid Gaussian membership function, the determination of a was transferred to the determination of adjacent membership functions’intersection point membership grade8which was called the overlap value. To derive the best δ value, firstly, different8values were adopted in fuzzy models which were used for the prediction of4datasets and the prediction performances were compared. Thus the best8value was derived for each dataset with different noise level and the generalization of8selectioin was demonstrated. Then, the same comparison was executed in the OFS fuzzy modeling by using WM method with different8values and the similar conclusion was derived. Meanwhile, comparison demonstrated that, domain partition based on clustering method and the hybrid Gaussian membership performed better than the traditional even domain partition in the OFS fuzzy modeling.(4) To realize the function of the AA system which means the prevention of high-risk OFS, the "OFS Prediction" concept was used. According to this concept, the OFS dynamic predictive model was constructed and verified. Firstly, the predictive model structure was identified. The results demonstrated the WM-based1st-order input-output model can get the best performance for OFS prediction. To improve the effective prediction rate of high-risk OFSs, the multi-model strategy was used instead of the single model, and multiple WM models were built for OFS prediction. To verify the usefulness of the predictive model, an adaptive task allocation strategy was designed and based on which an adaptive HM system was simulated. Simulation results showed the OFS can be effectively regulated, the operator task performance can be significantly improved and the number of high-risk OFSs can be greatly reduced in the adaptive HM system. Thus, the safety of HM system was substantially enhanced.更多还原