A Two‐Level Hierarchical Sparse DRL Framework for Coordinated Wide‐Area and Local Damping in DFIG‐Rich Low‐Inertia Grids

This paper presents a two‐level hierarchical adaptive damping control strategy designed to robustly mitigate inter‐area low‐frequency oscillations (LFOs) in large‐scale power systems featuring high shares of doubly‐fed induction generator (DFIG) wind farms. The proposed structure integrates a central wide‐area damping controller (WADC), driven by real‐time phasor measurement unit (PMU) data, with multiple local power oscillation dampers (LPODs) implemented on carefully selected DFIG turbines. Geometric controllability and observability indices are employed to determine optimal DFIG sites, wide‐area and local feedback signals and effective PMU placements. A key contribution is the online parameter adaptation of both the WADC and the LPODs using a sparsity‐promoting deep deterministic policy gradient (DDPG) agent. After offline training, the agent enforces a coordinated sparse control policy, updating only the most influential gains to achieve superior damping with minimal control effort. Communication delays are explicitly represented via the Padé approximation and robustness against model uncertainties, operating variations and network nonlinearities is ensured through an inverse output multiplicative perturbation framework. The hierarchical controller is validated on a modified 16‐machine, 68‐bus system featuring large‐scale DFIG generation. Analyses confirm that the DDPG‐optimized WADC/LPOD scheme delivers exceptional damping, resilience and scalability for future renewable‐dominated power grids.