AC/DC hybrid distribution network reconfiguration with microgrid formation using multi-agent soft actor-critic

Recent extreme events trigger tremendous concerns on distribution system resilience. Meanwhile, high penetration of inverter-interfaced distributed generators (DGs) and diversified source and load mix facilitate the development and implementation of hybrid AC and DC distribution networks (HDNs). This paper proposes a deep reinforcement learning-based (DRL) approach for distribution network reconfiguration with microgrid formation in face of extreme events. The proposed optimization model facilitates critical service restoration by forming isolated sections nested inside the HDNs when severe power outages occur (e.g., disconnection from the main grid). The operational characteristics of isolated HDNs (e.g., droop-controlled nodes in AC and DC sections, lack of slack buses in autonomous operation, etc.) are considered. To reduce the computational burden, a multi agent soft actor-critic (MA-SAC) approach is developed to solve the proposed reconfiguration problem, where multiple agents coordinately control circuit breakers to sectionalize the HDNs and can cater for different system states and scales. Simulation tests are conducted in two test systems to verify the validity of the proposed approach.

Automated summary

This paper proposes a deep reinforcement learning-based approach for distribution network reconfiguration with microgrid formation in response to extreme events. The proposed method restores critical services by forming isolated sections, taking into account the operational characteristics of the isolated sections, and adopts a multi-agent approach to reduce the computational burden.