Improving Semi-Quantitative Reasoning by Landmark Approximation

Many extensions to existing qualitative simulation packages allow combined use of available numeric and qualitative information . Some provide numeric integration across two qualitative states but none of them highlights numerically or semi-quantitatively the qualitatively predicted time points . Moreover, high accuracy at time points is only achieved by a fine level of granularity across the entire time interval . This paper presents landmark approximation as a new methodology to improve semi-quantitative predictions in qualitative simulation . We also present a transition table and a method to propagate semiquantitative values that is suitable for landmark approximation . These techniques have been implemented in a simulation engine that is based on qualitative reasoning techniques with a semi-quantitative extension . We demonstrate that a semi-quantitative approximation to landmarks improves the predictive power of the simulation tool even with a reduced number of interpolated states .