Deep Multi-Task Forecasting of Net-Load and EV Charging with a Residual-Normalised GRU in IoT-Enabled Microgrids
Muhammed Cavus, Jing Jiang, Adib Allahham
Abstract
The increasing penetration of electric vehicles (EVs) and rooftop photovoltaics (PV) is intensifying the variability and uncertainty of residential net demand, thereby challenging real-time operation in smart grids and microgrids. The purpose of this study is to develop and evaluate an accurate and operationally relevant short-term forecasting framework that jointly models household net demand and EV charging behaviour. To this end, a Residual-Normalised Multi-Task GRU (RN-MTGRU) architecture is proposed, enabling the simultaneous learning of shared temporal patterns across interdependent energy streams while maintaining robustness under highly non-stationary conditions. Using one-minute resolution measurements of household demand, PV generation, EV charging activity, and weather variables, the proposed model consistently outperforms benchmark forecasting approaches across 1–30 min horizons, with the largest performance gains observed during periods of rapid load variation. Beyond predictive accuracy, the relevance of the proposed approach is demonstrated through a demand response case study, where forecast-informed control leads to substantial reductions in daily peak demand on critical days and a measurable annual increase in PV self-consumption. These results highlight the practical significance of the RN-MTGRU as a scalable forecasting solution that enhances local flexibility, supports renewable integration, and strengthens real-time decision-making in residential smart grid environments.