Qualitative-Numeric Simulation with Q3

1 Overview A simulation contains a sequence of predicted states of a modeled system. A qualitative-numeric simulation is one containing both qualitative and quantitative state information, such that the qualitative information alone would be a qualitative simulation, and the quantitative information alone would be a numerical simulation. Qualitative-numeric simulation is thus a generalization of both qualitative simulation and numerical simulation, providing a framework for viewing historically disparate genres of simulation. Qualitative-numeric simulation also holds promise as an applied technique. Since it is a generalization of numerical simulation it has useful properties associated with numerical simulation that qualitative simulation does not have, like numerical predictions. And as a generalization of qualitative simulation, it has useful properties of qualitative simulation not present in numerical simulation, like dealing with weakly deened models, and automatically making qualitative distinctions among device behaviors and among model variable values. These points are demonstrated in an implemented qualitative-numeric sim-ulator, Q3. 1 2 Introduction Simulation of models to predict behavior of real systems has a long history and much practical use. Discrete event simulation, used e.g. in computer engineering , simulates models whose state variables change discontinuously. Simulation of continuous systems also has a history of practical use, usually coming under the heading of numerical methods. Combined simulation refers to simulations containing both discrete and continuous changes in state. Artiicial 1 Intelligence has relatively recently entered the simulation eld with qualitative simulation, e.g. de Kleer DEKL77;DEKL84], Forbus FORB84], Kuipers KUIP84;KUIP86], and others (see collections in Bobrow BOBR84] and Weld and de Kleer WEDE90]). Qualitative simulation derives predictions of behavior from non-numeric models. Qualitative models represent classes of systems in contrast to the complete system speciication typical of numerical simulation. Another reason for using qualitative modeling is that information precise enough for a numerical model may not be available. However, the predictions of a qualitative model are relatively weak and may lead rapidly to intractably many possible futures. Thus we would like a simulation technique which has the strengths of both qualitative and numerical simulation. Work on adding numerical information to qualitative models and simulations is reported by de These works succeeded in integrating quantitative and qualitative information to a signiicant degree. In most cases, however, qualitative and quantitative knowledge was maintained in distinct representations. The work presented here describes a system, Q3, that extends previous work, integrating the qualitative and the quantitative in a ne-grained way. Q3 produces qualitative-numeric …