AI-DARWIN: A first principles-based model discovery engine using machine learning

Abstract

The limitations of AI-based black-box models regarding reliability and interpretability have long been a major concern for researchers who have been arguing the case for hybrid models that integrate first-principles with machine learning techniques as a remedy. Here we propose a machine learning system that automatically discovers such mechanism-based models from data for nonlinear parametric systems. This is an extension of our previous work on linear systems using genetic feature extraction. The approach works under conditions of unknown functional transformations relating the input to the output and results in functionally tractable and explainable model forms, which are mechanistically feasible.