Physics-based symbolic regression for power flow modeling and analysis

Anonymous

Keywords: Interpretable AI, Physics-informed neural network, Distribution networks

Abstract: Symbolic regression searches the space of mathematical expressions to find the model that best fits a given dataset. Therefore, it can successfully integrate mathematical expressions and underlying physical laws to improve data-driven power flow modeling and analysis. We introduce physics-based symbolic regression for power flow modeling and analysis by taking inspiration from ’AI Feynman’ (Udrescu & Tegmark, 2020). A physics-informed neural network is integrated into the proposed symbolic regression algorithm in Udrescu & Tegmark (2020). The results show that, for power flow analysis and modeling of a low-voltage distribution network, the physics-based symbolic regression outperforms the original algorithm, where a black-box neural network is used as a part of the algorithm. The contribution of this paper is, therefore to introduce the idea of integrating physical laws and constraints into the physics-inspired symbolic regression algorithm using physics-informed neural networks.