Learning-Based Predictive Virtual Inertia Control for Frequency Regulation in Low Inertia Power Grids

The increasing integration of renewable energy generators into the conventional power grid has lowered the overall power system inertia. Low inertia may adversely impact the grid’s stability and resilience, leading to unintended grid failure and total system collapse upon severe contingencies. Current methods of emulating power grid inertia to enhance stability under low inertia conditions have some drawbacks. For example, the proportional-integral (PI) control method suffers from high overshoot and settling time, while the performance of the conventional model predictive controller (MPC) is highly dependent on the accuracy of the system model. Consequently, a new learning-based predictive virtual inertia control (VIC) is proposed. This strategy incorporates a physics-informed neural network (PINN) predictive model with an optimization-based safety filter to mitigate the impact of increasing integration of renewable energy generators through a virtual inertia emulation strategy. Our proposed strategy leverages advances in deep learning by implementing a PINN-based model to predict the system’s future behavior. The proposed method is compared to the MPC and PI controller. The results demonstrate the superior dynamic performance of the proposed method. Our implemented technique achieved the least frequency deviations upon contingencies in the IEEE 34 bus and the IEEE 39 bus studied systems with integrated inverter-based generators.