Real-time inertia estimation in an inverter-dominated distribution grid using a physics-informed recurrent neural network

To ensure frequency stability, estimation of system inertia becomes essential in modern power grids. Growing shares of virtual inertia from inverter-coupled resources (ICRs) shift this task to the distribution grids (DGs) and introduce new challenges. Using a physics-informed neural network (PINN) to combine data-driven modelling with knowledge of system dynamics, this study presents an approach to real-time system inertia estimation in inverter-dominated DGs. Based on the PINN literature framework, a modified loss function (LF) with adaptive weighting is proposed for a recurrent PINN. The approach is evaluated on a 14-bus medium voltage (MV) DG model, featuring virtual inertia from distributed ICRs with characteristic nonlinearities.