Presentation Details
| A Deep Reinforcement Learning Framework for Resilient Multi-Period Distribution Network Reconfiguration Ricardo Calloquispe-Huallpa1, Rachid Darbali-Zamora2, Anny Huaman-Rivera1, Erick E.Aponte-Bezares1. 1University of Puerto Rico-Mayagüez, Mayagüez, PR, USA.2Sandia National Laboratories, Albuquerque, NM, USA |
Abstract
This paper proposes a deep reinforcement learning (DRL)-based framework for multi-period distribution network reconfiguration aimed at maximizing the delivery of critical services under varying operating conditions. The problem is formulated as a Markov decision process (MDP), where the agent learns to determine network switching actions over time while satisfying physical and topological constraints through an action masking mechanism. To quantify system performance, a weighted service index (WSI) is introduced, incorporating a nonlinear operability function that captures both the importance of facilities and their accumulated service over the time horizon. This formulation promotes a balanced allocation of resources, prioritizing critical facilities while ensuring broader service coverage. The proposed approach is evaluated on a distribution network representing a rural community in Puerto Rico using real load and photovoltaic (PV) generation data. The results demonstrate that the learned policy effectively adapts to different operating conditions, including distributed energy resource (DER) and line outages, achieving high levels of service delivery. A comparative analysis against a genetic algorithm (GA) shows that the proposed method attains superior performance in terms of service quality while reducing computational time by several orders of magnitude. These results highlight the potential of the proposed framework for real-time resilient operation of distribution networks.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
Content Locked. Log into a registered attendee account to access this presentation.