1992 Volume 7 Issue 5 Pages 837-849
This paper proposes a method of qualitative simulation with association mechanism to quantitative analysis. There exist two difficult problems when predicting behaviors of physical systems with qualitative models ; one is how to construct models, and the other is how to solve ambiguity during reasoning. The proposed method here is based on two associating operations between quantitative and qualitative analysis. As for the construction of qualitative models, the method uses the dependency structure among parameters of quantitative models. Then a certain subset of whole parameters is selected by a user's viewpoints. The method extracts a reduced dependency structure for such selected parameters. Consequently qualitative models are constructed by assigning qualitative relations among selected parameters based on the reduced structure. As for pruning undesirable behaviors during reasoning, it is done by comparing qualitative behaviors with quantitative data, where we introduce converted data and interpretative time-points. Converted data and interpretative time-points are generated as follows. First, each quantitative data is converted into qualitative data from the standpoint of parameters' changing states. Second, interpretative time-points are derived from merging each duration time that is described in each of converted data. Thus the method compares the inferred qualitative behaviors with converted data along interpretative time-points. These converted data are regarded as constraint during reasoning. Third, if one inferred qualitative behavior satisfies such constraint, this method goes ahead with interpretative time-points and continues reasoning. On the other hand, if no inferred state satisfies it, it indicates modification of qualitative models or ignores some constraints by setting those constraints as qualitative states of parameters. This duration-independent characteristic of reasoning process plays an important role in applying this method to real problems. Causal explanation is also generated through the propagation process in reasoning. We apply this method to large-scale power plants and extract causal explanation.