Implementation of predictive modelling-empowered HVAC-driven PV self-consumption Enhancement Strategies in a Warehouse

The current work proposes a machine learning (ML)-based HVAC management strategy that shifts heating load to periods of high photovoltaic (PV) output, using demand and generation forecasts, aiming to maximize self-consumption. In this context, the case study of a conditioned warehouse with installed onsite PV panels and dedicated office and storage areas is considered. EnergyPlus is used to simulate building baseline behavior under varying conditions, generating data to train physics-informed ML models that would predict the hour-ahead load of the building. Additionally, PV generation forecasts are obtained from an open-access dataset. The strategy agent receives real-time and short-term forecasts for both building load and PV generation, continuously monitoring their balance. Based on this information, the agent applies slight heating setpoint adjustments, relaxing the setpoints or preheating the indoor environment by one degree, to shift the heating load away from grid reliance and toward periods of high PV availability. The results show that implementing the proposed interventions increases the building's self-consumption rate by 9% during the targeted intervention periods and by 7.47% over the entire heating season. Likewise, the self-sufficiency rate improves, indicating that a larger share of the building's energy demand is met by the PV system.