Physics-Informed Neural Network-Based Predictive Model for Virtual Synchronous Machines in Inverter-Based Microgrids

Virtual synchronous machines (VSMs) have recently emerged as a practical solution for incorporating virtual mechanical inertia into power systems using power electronic converters to add frequency support to the grid to handle the increase in the rate of change of frequency (RoCoF) that may lead to stability problems. Analyzing this concept compared to frequency-droop-based control schemes provides valuable physics-based insights into the tuning and operation of VSM controllers. However, the design and control of VSMs necessitate accurate predictive models that can adapt to microgrids' dynamic and complex nature. Physics-informed neural network (PINN) is an emerging discipline that combines principles from physics and neural networks to enhance the accuracy and efficiency of predictive models. By incorporating the fundamental laws and equations that govern a system, PINNs guide the learning process of neural networks and improve the interpretability of the resulting models. This study proposes PINNs trained to predict the VSMs model while ensuring compliance with the fundamental principles of physics described by the conventional swing equation governing synchronous machines (SMs). The effectiveness of the predicted model has been verified via simulations carried out using MATLAB software, demonstrating evidence of its potential to improve microgrid stability and power quality.