Economic feedback-guided forecasting and scheduling for microgrids using rolling model predictive control

Li, Yang, Li, Yang, You, Jiuzhou, Li, Yuanzheng, Li, Yuanzheng, Long, Chao, Zhou, Yue ORCID: https://orcid.org/0000-0002-6698-4714 and Shahidehpour, Mohammad 2027. Economic feedback-guided forecasting and scheduling for microgrids using rolling model predictive control. Applied Energy 427 , 128778. 10.1016/j.apenergy.2026.128778

Full text not available from this repository.

Abstract

Microgrid rolling energy management often follows a decoupled forecast-then-optimize workflow, where forecasts are trained separately from downstream rolling model predictive control (MPC) scheduling. However, lower statistical forecasting errors do not necessarily reduce operating costs because conventional forecasting objectives are not directly aligned with downstream scheduling outcomes. To address this issue, this paper proposes an economic feedback-guided decision-aware forecasting and scheduling framework for microgrids using MPC. In the proposed framework, the implemented-action economic cost from rolling MPC is used to guide forecasting model training, while the MPC scheduler and solver remain unchanged. A modified encoder-only Informer is developed as a multi-task day-ahead forecasting module for wind power, photovoltaic generation, and load demand. Its forecasts are evaluated through a rolling MPC framework. A joint training objective combining supervised forecasting loss and implemented-action economic feedback is then constructed to learn predictions that are more useful for practical rolling scheduling. To handle the non-differentiability of rolling MPC caused by mixed-integer decisions and hard constraints, simultaneous perturbation stochastic approximation is used to estimate the economic-feedback gradient. In this way, the implemented-action economic feedback can be used to update the forecasting model parameters without modifying the existing MPC scheduler. In addition, an aggregated electric vehicle model and a dual-inventory demand response model for time-shiftable electric load are established to support coordinated rolling scheduling. Case studies show that the proposed method reduces implemented-action costs while maintaining comparable forecasting accuracy. These results demonstrate that MPC-based economic feedback can improve the practical value of forecasts for microgrid scheduling.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Engineering
Publisher:	Elsevier
ISSN:	0306-2619
Date of Acceptance:	29 August 2026
Last Modified:	12 Oct 2026 15:30
URI:	https://orca.cardiff.ac.uk/id/eprint/190176

Actions (repository staff only)

Edit Item