PIUL-DERs: Physics-informed pseudo-labeling unsupervised learning for smart homes with distributed energy resources

This article proposes a novel algorithm PIUL-DERs, physics-informed pseudo-labeling unsupervised learning for the residential distributed energy resources (DERs) management of smart homes with electric vehicles (EVs) and photovoltaic (PV) systems. A self-optimizing pseudo-labeling unsupervised learning framework is developed to predict EV usage and PV generation, where a physics-informed model specifically captures PV generation uncertainties by embedding solar irradiance model, weather characteristics, and thermal efficiency. Then, based on residential DERs predictions, the carbon-aware centralized energy management scheme (CEMS) addresses the challenges of energy storage coordination in smart homes, considering dynamic power grid carbon intensity to reduce grid dependence of individual smart homes and thus facilitate real-time carbon emissions reductions. The simulation results validate the reliability and efficiency of the PIUL-DERs. The proposed PIUL-DERs accurately predicts PV/EV profiles with 73 % less complexity than that of the existing method and realizes carbon-aware CEMS between smart homes, reducing carbon emissions by 46 %. • Propose a physics-informed pseudo-labeling unsupervised learning method to predict residential distributed energy resources. • Exploit the physical laws of EVs and PVs to generate pseudo-labels and augment the dataset. • Introduce carbon-aware CEMS to facilitate real-time carbon emissions reduction. • Outperform state-of-the-art solutions with 73 % faster training efficiency. • Reduce grid dependence by 42 % and carbon emissions by 46 %.