Distributed Machine Learning for Resilient Operation of Electric Systems

Abstract

Power system resilience is crucial to ensure secure energy delivery to electricity consumers. Power system outages lead to economical and societal burdens for the society and industries. To mitigate the socio-economical impacts of a power outage, we need to develop efficient algorithms to ensure resilient operation of the power system. In this paper, we first explain the notion of data-driven resilience. Then, we present a pathway of leveraging edge intelligence to improve resilience. To this end, we propose a novel distributed machine learning paradigm. Our proposed structure relies on local Resilience Management Systems (RMS) that serve as intelligent decision-making entities in each area, e.g. an autonomous micro-grid or a smart home can act as RMS. The RMS agents, which are available in different areas, can share their local data (i.e., a microgrid's operational data) with their neighboring RMS to coordinate their decisions in a distributed fashion. This will provide two major advantages: 1) distributed intelligence replaces centralized decision-making leading to robust decision-making and enhanced resilience; 2) since local data are locally shared among all entities within an RMS, if one of the RMS agents fails to communicate with the rest of network, we still can maintain a feasible solution (which is not necessarily optimal). Finally, we presents different scenarios in the simulation results section that showcases the system performance for two buildings under various outage scenarios.