Development of expert system tools with knowledge compilation function based on deep knowledge

本研究的主要目标是开发创新的知识处理技术，以帮助克服现有技术的局限性，如脆性和缺乏适应性。除了用于解决普通问题的启发式知识(浅层知识)外，人类专家还拥有基础知识和领域原理(深层知识)。当他们被赋予不能用启发式方法解决的陌生问题时，他们会试图用上面提到的深层知识来解决它。基于这一观察，研究者开展了基于深层知识的专家系统和浅层知识的知识编译的研究。其中一个主要成果是建立了一个基于五种深层知识的诊断专家系统知识编译系统。一个原型系统已经在SparkStation上的Common ESP上成功实现。这项研究与基于模型的推理有着密切的联系，基于模型的推理是对定性建模的动态系统的行为环境的研究。定性推理的一个缺点是推理的模糊性，这阻碍了该方法在实际问题中的应用。本研究的另一个目标是开发一种新的定性推理方法来减少歧义。设计了一种核电站冷却子系统的建模方法和推理方法。本研究的第三个目标是提出一种建立可复用知识库的方法。虽然尚未完成，但已经为变电站恢复知识库设计了该方法的第一个版本。

The major objective of this research is to develop innovative knowledge processing techniques which contribute to overcoming the limitation of the current technology such as brittleness and lack of adaptability. Human experts have fundamental knowledge and domain principles(deep knowledge) in addition to heuristics(shallow knowledge) which is used in solving ordinary problems. When they are given unfamiliar problems they cannot solve by using heuristics, they try to solve it by using above mentioned deep knowledge. Based on this observation, the investigator conducted the research on deep knowledge-based expert system and knowledge compilation of shallow knowledge. One of the main results obtained is a knowledge compilation system based on five kinds of deep knowledge for diagnostic expert systems. A prototype systems has been successfully implemented in Common ESP on Sparkstation.This research has a tight relation to model-based reasoning which is a research on environment of the behavior of qualitatively modeled dynamic systems. One of the shortcomings of the qualitative reasoning is ambiguity of the inference which prevents the method to be used in practical problems. Another objective of this research is to develop a new qualitative reasoning method for decreasing the ambiguity. A modeling method and inference method have been designed for cooling subsystem of a nuclear power plant.The third objective of this research is to develop a methodology for building reusable knowledge base. Although not completed yet, a first version of the methodology has been designed for the knowledge base for restoration of substations.