Power grids are changing as more electricity comes from inverter-based resources (IBRs) such as wind, solar and battery systems. These devices do not behave like traditional generators and can make the grid less stable during big shocks. This paper proposes a new, model-free way to tune the control settings of many IBRs together so the whole grid responds better to large disturbances.

Instead of writing a simplified math model of the whole grid, the researchers treat a high-fidelity power system simulator as a black box. They feed the simulator a candidate set of controller parameters and a disturbance, then measure simulation outputs such as the deepest frequency dip or peak (nadir/zenith), the rate of change of frequency (RoCoF), and the largest phase-angle swing. Those outputs are the only information their optimizer uses. The optimizer is a new algorithm the authors call PMZO-Adam — a projected, multi-point, zeroth-order method with adaptive moment estimation. In plain terms, it is a gradient-free search that keeps parameters inside safe ranges, uses several simulation runs per step to reduce noise, and borrows an adaptive update rule (Adam) to speed convergence. The authors implemented an electromagnetic-transient model of a modified IEEE 39-bus system with ten IBRs in Matlab Simulink for their tests.

Why use this approach? Traditional tuning methods often rely on simplified or linearized models and focus on small perturbations. Those models can miss the strongly nonlinear behavior that appears during large disturbances and they usually tune devices one at a time. The simulation-based, model-free method directly optimizes the grid’s transient behavior and coordinates settings across many devices. The paper discusses tuning parameters for both grid-following controllers (which track grid voltage and include a phase-locked loop, droop control and current loops) and grid-forming controllers (which set voltage and frequency, here represented by a virtual synchronous generator). Example tunable quantities include PLL gains, droop coefficients and current-controller gains.