A Reinforced Learning Approach to Dispatch Distributed Generators for Enhanced Resilience

This paper proposes a reinforced learning-based approach for dispatching distributed generators to enhance operational resilience of electric distribution systems against hurricanes. Existing resilience enhancement approaches rely on solving large-scale optimization problems that are computationally expensive and time demanding, which are not suitable for real-time applications. In this paper, a multi-agent framework is developed using a Soft Actor Critic algorithm to dispatch distributed generators for resilience enhancement. The proposed approach provides a fast-acting control algorithm that determines the size and the location of distributed generators to reduce the amount of load curtailment during hurricanes. The problem is formulated as a Markov decision process that consists of system states, an action space, and a reward scheme. A system state represents the system topology and characteristics upon which an action is taken and a reward value is calculated. An iterative Markov decision process is used to train the proposed Soft Actor Critic algorithm using multiple line outages generated from a hurricane fragility model. The trained network dispatches distributed generators whenever there are islanded grids and load curtailments. The proposed method is demonstrated on the IEEE 33-node distribution feeder system. The results show the capability of the proposed algorithm to determine optimal sizes and locations of distributed generators for resilience enhancement.