Research on Visualizing a Plant Operator's Concepts and Supporting Distributed Decisionmaking

在大型复杂工厂的运行中，操作所需的专业知识通常分布在多个人类操作员之间，他们必须相互协作，分享和/或交换个体决策以及他们关于工厂现状的背景概念。在本研究项目的第一年，我们试图通过影响图来阐明专家操作员对该工厂的认知图，并开发了一套合理设计的控制规则的推导方法，以建立多个过程的协调控制。最后，将所推导出的控制规则应用于实际污水处理厂，给出了控制结果。然后，我们尝试设计一个界面系统，该系统能够像人类操作员从…中判断植物状态一样，形成关于植物异常行为的层次化组织概念更多的测量数据流。采用人工智能归纳学习方法中的概念形成方法，设计了支持人机协同工作环境的人性化界面系统；第二年，在前一年研究成果的基础上，提出了一种界面智能体的概念形成方法，通过观察用户与系统之间的交互来捕捉用户的内部概念结构。我们提出的算法包括两个阶段：第一个阶段是从一个给定的属性集中选择基本属性，这些属性最初可能包含相关的和不相关的属性，第二个阶段是使用遗传算法来构造新的属性，该算法应用于一组逻辑上以析取范式表示的基本特征。将我们的方法应用于一个人工数据集，并讨论了它在学徒学习中的实际应用。然后，我们制定了一个界面代理来做出高风险、时间关键的决策。通过应用概念形成方法，针对可用于界面代理的植物异常的层次组织概念，提出了该公式。较少

In an operation of a large-scaled and complex plant, it is usual that the expertise needed for the operation is distributed among a number of human operators and they have to collaborate with each other sharing and/or exchanging the individual decisions as well as their background concepts on the current status of the plant.In the first year of this research project, we attemp to explicate the expert operator's cognitive map concerning with this plant by influence diagrams and developed a method for deriving a set of control rules that is rationally designed for establishing the coordinate control of multiple processes. Finally, the derived control rules were applied to an actual plant for sewage disposal and the control result was shown. Then, we attempted to design an interface system that has a capabilities of forming the hierarchically organized concepts on the anomalous behaviors of plants just as the human operators have when they make judgments on the plant status from the incom … More ing stream of measuremnt data. We adopted a concept formation method that is one of the inductive learning methods of artificial intelligence, and design a human-friendly interface system that can support a man-machine collaborative work environment.In the second year, based on what is attained in the previous year we presented a method for concept formation for an interface agent that affempts to capture user's internal conceptual structure by observing interactions between an user and a system. Our proposing algorithm consisted of two phases : the one for selecting the essential attributes out of a provided set of attributes that may initially contain both relevant and irrelevant ones, and the other for constructing new attributes using genetic algorithms applied to a set of elementary features logically represented in a disjunctive normal form. Our method was applied to an artificial data set and its practical usage for apprentice learning was discussed. Then, we formulated an interface agent's making high-stakes, time-critical decisions. The formulation was made with respect to the hierarchically organized concepts of plant anomalies that are availble to an interface agent by applying a concept formation method. Less