Non-Intrusive Load Management Under Forecast Uncertainty in Energy Constrained Microgrids

Abstract This paper addresses the problem of managing load under energy scarcity in islanded microgrids with multiple customers and distributed solar generation and battery storage. We explore an understudied, practical approach of scheduling customer-specific load limits that does not require direct control of appliances or a market environment. We frame this as a stochastic, model-predictive control problem with forecasts of solar resource and electricity demand, and develop alternative solutions with two-stage stochastic programming and approximate dynamic programming. We test the efficacy of the alternative solutions against heuristic and deterministic controllers in an environment simulating the customers’ responses to load limits. We show that using forecasts to schedule limits can significantly improve power availability and the customers’ benefits from consumption, even without the controller having a full model of the customers’ responses.