Distribution Grid Critical Load Restoration under Uncertain Topology Changes via a Hierarchical Multi-Agent Reinforcement Learning Approach

Extreme weather events and/or cyber-attacks can significantly disrupt the power generation of a power grid, leading to catastrophic consequences. In this paper, the critical load restoration (CLR) problem in the community distribution grid is addressed. Existing approaches for CLR rely on the assumption that the grid topology does not change during the restoration period. However, it is highly likely that, under major disruptions, some of the buses/lines can get disconnected and change the underlying topology of the grid. These uncertain topology changes could lead to a different CLR problem formulation as well as a restoration strategy. To this end, we propose a hierarchical multi-agent reinforcement learning (HMARL) framework for CLR, which uses topology-dependent action masks (TDAM) to handle the changing topology. The main idea is to divide the distribution grid into multiple cells capable of independent control and a coordinating agent to allow power transfer between different cells under topological variations during restoration. Moreover, TDAM helps identify the actions that are unavailable after the topology change. We demonstrate the effectiveness of the proposed method on a modified IEEE-123 bus system, showing that it achieves robust load restoration despite fluctuating topology.