Dispatch of Virtual Inertia and Damping via Deep Reinforcement Learning for Improving System Frequency Stability

The optimal dispatch of virtual inertia and damping (DID) of virtual synchronous generator (VSG) inverters plays a key role in improving system frequency stability. However, DID is mostly modeled based on linearized models due to the limited ability of mathematical programming to solve non-linear problems, which cannot reflect the system dynamics after contingency. To this end, this paper aims to employ deep reinforcement learning (DRL) to solve the DID problem based on a non-linear detailed system model. The formulated system model contains detailed models for grid-forming (GFM) and grid-following (GFL) inverters and is simulated in the time domain. Moreover, the soft actor-critic (SAC) algorithm is employed to solve the DID and trained by the system metrics from time domain simulations. Finally, a case study based on a three-area system is presented to demonstrate the effectiveness of the proposed DRL approach and the performance of DID to enhance power system frequency stability.