Adaptive Frequency Control for Inverter-Based Resources via Equivalent Inertia-Based Deep Reinforcement Learning

With inverter-based resources (IBRs) employing heterogeneous control strategies in modern power systems, optimizing their frequency control has become increasingly vital. Since IBR control strategies can often be switched through simple software updates, the same IBR may operate under different control strategies. This typically requires retraining of any deep reinforcement learning (DRL)-based frequency control agent. This paper presents a DRL-based framework that reduces the need for retraining when control strategies differ by leveraging equivalent inertia as a unifying metric. A DRL agent is trained only on virtual synchronous generators to provide frequency control during disturbances. These actions are then mapped online to droop-based control parameters using a precomputed matching table. This approach decouples the learning process from a specific IBR control strategy, allowing the same policy to operate across heterogeneous controls. The method is evaluated on the IEEE 14-bus test system using time-domain simulations. Results show that the mapped strategy delivers frequency performance on par with an jointly trained agent, while avoiding the overhead of repeated training for different IBR controls.