Modeling Chemical Reaction ODEs using PINNs

Abstract

In recent years, physics-informed neural networks(PINN) are based on deep learning and have been widely used in scientific and engineering problems, where differential equations serve as the foundation. In our case, we are focusing on the field of plasma, specifically the chlorine discharge global model. The model is composed of a set of stiffness ordinary differential equations obtained from various chemical reaction equations. However, the naive-PINNs has limitations in addressing stiffness. Therefore, in this paper, we aim to address stiffness by combining selected methodologies from the fields of both deep learning that enhances the efficiency of learning(log transformation, standardization) and various PINNs methodologies(Fourier embedding, self-adaptive learning). We demonstrate the effectiveness of the integrated methodology through the ROBER problem, a simple argon stiff ordinary differential equation, and ultimately, the chlorine discharge global model which is of our interest.