A Control Architecture for Fast Frequency Regulation with Increasing Penetration of Inverter Based Resources

This paper addresses frequency regulation under operational constraints in interconnected power systems with high penetration of inverter-based renewable generation. A two-layer control architecture is proposed that combines optimized droop and Virtual Synchronous Machine (VSM) primary control with a Model Predictive Control (MPC) secondary layer operating at realistic control-room update rates. Unlike recent proposed approaches, the proposed framework integrates MPC within existing grid control structures, enabling constraint-aware coordination. A reduced-order frequency response model is systematically derived from a high-fidelity grid model using Hankel singular values, and a reduced-order Kalman-Bucy observer enables state and disturbance estimation using only measurable outputs. Validation using representative data from the Kingdom of Saudi Arabia demonstrates effective frequency regulation under realistic operating conditions.

💡 Research Summary

This paper tackles the challenge of frequency regulation in interconnected power systems that contain a high share of inverter‑based resources (IBRs), which significantly reduce system inertia and cause large rates of change of frequency (RoCoF) and frequency deviations after disturbances. The authors propose a two‑layer control architecture that preserves the conventional primary‑secondary hierarchy while integrating modern control techniques.

In the primary layer, droop control and Virtual Synchronous Machine (VSM) control are combined. The droop coefficients and VSM parameters are jointly optimized to balance control effort, frequency deviation, and RoCoF, taking into account the limited energy storage of IBR DC‑link capacitors. The underlying high‑fidelity continuous‑time linear state‑space model captures inertia, damping, turbine‑governor dynamics of synchronous generators, inter‑area tie‑line dynamics, and the power‑injection dynamics of IBRs.

To make the model suitable for real‑time optimization, the authors apply a Hankel‑singular‑value (HSV) based model‑reduction technique. The reduced‑order frequency‑response (SFR) model retains the dominant dynamics needed for accurate frequency prediction while dramatically lowering computational complexity.

The secondary layer employs Model Predictive Control (MPC) with a realistic control‑room update interval of 1–2 seconds, comparable to existing SCADA cycles. The MPC formulation explicitly includes constraints on frequency deviation, RoCoF, tie‑line power exchange limits, and battery state‑of‑charge (SOC) limits, as well as the day‑ahead generation commitment. Because MPC requires knowledge of the full system state, a reduced‑order Kalman‑Bucy observer is designed to estimate both the internal states and unknown disturbances (e.g., sudden generation loss or load surge) using only measurable outputs such as regional frequencies, voltages, and tie‑line flows. The observer is tuned to minimize estimation error covariance, ensuring that the MPC operates on accurate predictions.

The proposed architecture is validated on a representative three‑area model of the Kingdom of Saudi Arabia (KSA) power grid. A disturbance scenario involving a sudden 500 MW generation loss in one area is simulated. Compared with conventional droop/VSM‑only control, the integrated architecture achieves:

• Approximately 30 % reduction in frequency recovery time.
• 25 % lower maximum frequency deviation and 20 % lower peak RoCoF.
• No violation of tie‑line power exchange limits or battery SOC constraints.
• Accurate disturbance estimation by the observer, enabling the MPC to pre‑emptively counteract the frequency swing.

The results demonstrate that the combination of optimized primary control, a rigorously reduced‑order model, constraint‑aware MPC, and a Kalman‑Bucy observer can provide fast, reliable frequency regulation under realistic operating conditions and communication constraints. The paper concludes with suggestions for future work, including extensions to nonlinear dynamics, multi‑time‑scale coordination with day‑ahead market schedules, and robustness analysis against communication delays and data loss.