Integrating domain knowledge in equation discovery

Abstract

In this chapter, we focus on the equation discovery task, i.e., the task of inducing models based on algebraic and ordinary differential equations from measured and observed data. We propose a methodology for integrating domain knowledge in the process of equation discovery. The proposed methodology transforms the available domain knowledge to a grammar specifying the space of candidate equation-based models. We show here how various aspects of knowledge about modeling dynamic systems in a particular domain of interest can be transformed to grammars. Thereafter, the equation discovery method LAGRAMGE can search through the space of models specified by the grammar and find ones that fit measured data well. We illustrate the utility of the proposed methodology on three modeling tasks from the domain of Environmental sciences. All three tasks involve establishing models of real-world systems from noisy measurement data. © Springer-Verlag Berlin Heidelberg 2007.