Intelligent Online Scheduling of Residential Microgrids based on Decision-Focused Learning

In the online scheduling of renewable-rich residen-tial microgrids, machine learning models are often applied to predict system uncertainty. However, the traditional predict-then-optimize modeling, where machine learning is used as a black-box tool, has been shown to be inefficient, as more accurate prediction often does not lead to improved decision-making. To address this issue, we propose to solve the online scheduling problem based on the Decision Focused Learning (DFL) framework, which integrates both prediction and decision-making through an end-to-end model training process. However, the scheduling problem in residential microgrids involves a highly nonlinear integer program that cannot be efficiently solved by standard DFL methods. To overcome this limitation, we propose a novel decomposition algorithm for the bilevel formulation of DFL, which transforms the original problem into a series of tractable sub-problems. We test our method on the modified IEEE 123-node test system and evaluate its performance through cost and prediction analysis. We demonstrate the effectiveness of our approach in managing microgrid operations under uncertainty and highlight the benefits of integrating learning into the scheduling process.