Examination of Deep Knowledge in Knowledge Compilers

A new method is proposed to examine a deep knowledge in knowledge compilers based on qualitative reasoning(QR)[1]. It is profitable if a knowledge compiler can examine a deep knowledge base to judge whether a thorough shallow knowledge can be generated or not, and can request the knowledge engineer to add some knowledge to the deep knowledge base if necessary, just like ordinary compilers check program source lists before generating executable objects and let the programmer know about defects of the source lists. Deep knowledge in QR-based knowledge compilers can be represented as simultaneous qualitative equations(SQEs). And the examination method reported in this paper is based on a structure analysis of a set of SQEs, and enables compilers to suggest necessary additional qualitative equation(QE)s. The way how this examination function works is explained with a sample model of a heat transport system of a nuclear power plant. Background In the conventional numerical calculations, the device under consideration is analyzed using numerical models, and analysis accuracy can be improved as the models become sophisticated. However, such a method doesn’t show us how the device behaviors are determined by its descriptions. For example, we cannot know “why this temperature increases when this pump stops”. And we need another whole analysis to know “how this behaves if this pipe were 1cm shorter than it is”. In contrast with numerical analyses, qualitative reasoning can be viewed as aiming at “to make it clear how the device behaviors are determined by its descriptions, in exchange for a compromise on the behavior accuracy.” Advantages of qualitative reasoningE2lE3]E41 has been claimed that reasoning can be started from incomplete information, and all possible behaviors can be derived. And explanation generation capacity has been recognized as one of the greatest advantages. However, combinatorial explosion is the heaviest drawback and makes crucial tradeoffs with the advantages described above. There are many efforts to achieve disambiguation and to maintain explanation capability and simplicity of qualitative reasoning, mostly by combining qualitative reasoning/simulation with numerical processing. Within most of these researches, disambiguation is done by choosing one among all qualitative solutions by matching with numericaI~yobtained result or human intuitive knowIedge.R~lIt is undoubtedly profitable if disambiguation process is explained also in a qualitative manner. In other words, qualitative reasoning will be much more effective if it shows why the other solutions should be denied as spurious ones, even when the conflicting change propagations have comparable elfect[6l. For this reason, some systems employ additional qualitative constraints for disambiguation. However, identification of the additional constraints is done manually in an ad hoc manner. And, in case that qualitative reasoning is done using insufficient information, users may need to know what information can determine the system behavior uniquely. If a deep knowledge base is viewed as a set of qualitative constraint, there should be a clear condition to attain disambiguation. However, researches to identify this condition and to utilize this condition for examining deep knowledge are hardly foundE7l. What the authors want to do in this research is to enable the system to examine initially given set of qualitative knowledge base and to suggest additional qualitative constraints useful for disambiguation of the solution. In this paper, an attempt to realize this suggesting function is presented. The proposed method deals with equilibrium equations. First, a qualitative model of a heat transport system of nuclear plant is shown, which has plural feedback loops, to emphasize the importance of additional qualitative constraints. Then, mechanism of ambiguity is analyzed. Finally, the method to check the initial deep knowledge and to suggest additional knowledge effective for disambiguation is explained and demonstrated. A Qualitative Model of Heat Transport System of Nuclear Power Plant A qualitative model of a heat transport system in a nuclear power plant is shown below. Dump Heat Reactor ~ntermediate [Exoha er Heat Transfer resistance Rdhx Fig.1 One ioop Heat Transport System Model of A Nuclear