Using Domain Specific Knowledge for Automated Modeling

Abstract

In this paper, we present a knowledge based approach to automated modeling of dynamic systems based on equation discovery. The approach allows for integration of domain specific modeling knowledge in the process of modeling. A formalism for encoding knowledge is proposed that is accessible to mathematical modelers from the domain of use. Given a specification of an observed system, the encoded knowledge can be easily transformed into an operational form of grammar that specifies the space of candidate models of the observed system. Then, equation discovery method Lagramge is used to search the space of candidate models and find the one that fits measured data best. The use of the automated modeling framework is illustrated on the example domain of population dynamics. The performance and robustness of the framework is evaluated on the task of reconstructing known models of a simple aquatic ecosystem.