Physics-Informed Neural Network for Microgrid Forward/Inverse Ordinary Differential Equations

Abstract

The increasing penetration of distributed generations is fundamentally reshaping the dynamics and stability characteristics of microgrids in power distribution systems. This shift complicates the model development and due to the lack of complete parameters of microgrid components, especially when vendors do not offer transparent information about their devices. To address the issue of unclear parameters and ambiguous mechanisms bringing by this situation, we introduce physics-informed neural network (PINN) into the parameter estimation and dynamic simulation processes of microgrids. By solving ordinary differential equations constructed for the electrical equipment within microgrids, PINN is able to provide dynamic time-domain predictions and estimating unknown parameters. This approach harnesses the strengths of neural networks while also complying with the laws of physics. It has been validated through modelling and verification in the real-time digital simulation system, demonstrating its effectiveness. This work has also shown the potential of extending, the introduced method to other microgrid application in which traditional numerical methods may encounter difficulties.