Bi-level Sparsity Learning Decision-Making for Disabling Primary Frequency Regulation to Suppress Frequency Oscillations

A bi-level sparsity learning decision-making approach is proposed for designing the schemes of disabling primary frequency regulation (DPFR) to suppress frequency oscillations. It addresses the inefficiency of classical methods that rely on multi-cycle measurements and heuristic rules. To quickly identify candidate units for DPFR, this paper develops a method to estimate the damping torque coefficient from limited measurements. It uses nonlinear least squares extrapolation, reducing the required data window to within a single oscillation cycle. Subsequently, two robust damping constraints are derived for DPFR decision-making. These constraints are designed to prevent DPFR from introducing additional negative damping. They also ensure a robust improvement in system damping under uncertainty in the shifted oscillation frequency. Furthermore, a DPFR decision-making model is formulated by integrating a primary frequency regulation metric with robust constraints. The model functions to suppress oscillation while maximizing available regulating power. Finally, a bi-level sparsity learning method is presented to solve the model. It efficiently identifies high-quality DPFR schemes. The approach is validated on a hardware-in-the-loop platform with two hydropower-dominated power systems. The validation results indicate that the proposed method reduces the data window required for accurate damping torque coefficient identification from two to three oscillation cycles to less than one cycle. The DPFR schemes generated by bi-level sparsity learning achieve the targeted oscillation suppression while preserving more primary frequency regulation capability. These results demonstrate that the proposed approach improves the efficiency of frequency oscillation suppression.