Skip to content
Open access

Hierarchical Short-Horizon Model Predictive Control With User Adaptation for Microcontroller-Based Balcony Photovoltaic Energy Management

2026 · IEEE Access · Vol 14, pp. 152848-152867 · 0 citations · 26 references

Abstract

Energy management in balcony photovoltaic (PV) systems is becoming increasingly complex with the integration of energy storage, flexible loads, and dynamic electricity prices under time-varying, user-dependent generation and demand. Model predictive control (MPC) can coordinate these flexibilities using forecast information while accounting for system dynamics and operational constraints. However, its computational requirements, particularly with increasing prediction horizons, are relevant for low-cost embedded hardware. This work investigates how an energy management system (EMS) can provide predictive operation under limited computational resources. The examined balcony PV system combines direct-current (DC) and alternating-current (AC) battery storage with thermal energy storage and is evaluated using two residential datasets. Alternative objective formulations, quadratic programming (QP) and mixed-integer linear programming (MILP) structures, and prediction horizons from one to four hours are analyzed. The resulting MPC is extended with replay-based user adaptation to account for individual preferences and enable on-device adaptation. Furthermore, a supervisory strategy incorporates remaining-day forecasts to complement the limited MPC prediction horizon. Results indicate that a two-hour multi-objective QP achieves performance close to the corresponding MILP while avoiding binary decision variables to facilitate embedded implementation. Preference adaptation improves the profile-specific evaluation score in 26 of 32 independent contexts while preserving the dominant user priorities. In extended closed-loop evaluations, strategic supervision reduces PV curtailment by 61.9 % and grid import by 8.8 %, while increasing PV self-consumption by 13.9 and self-sufficiency by 7.0 percentage points. The resulting QP-based controller was successfully implemented and executed on an ESP32-S3, demonstrating feasibility for resource-constrained embedded operation.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.