Smart home Demand-Side Management Based on rooftop deep learning photovoltaic power forecasting

Abstract

Price-responsive consumers in smart homes can apply Demand-Side Management to controllable loads, based on the availability of energy produced from rooftop photovoltaic systems and on contractual tariffs, ultimately enhancing household energy efficiency and reducing operational costs. However, several key research gaps remain unaddressed: the limited integration of Photovoltaic power forecasting with optimal load scheduling, the underutilization of Numerical Weather Predictions to improve forecasting accuracy, and the lack of comprehensive scheduling strategies for heterogeneous loads, such as shiftable, curtailable, and sheddable appliances To bridge these gaps we propose an integrated methodology that predicts energy generation and optimizes the load allocation upon energetic and economic criteria. Since low forecast accuracy worsens load scheduling, forecasting systems based on deep learning-based architectures are developed and compared. The novel day-ahead scheduling module is formulated as a Mixed Integer Constrained Nonlinear Programming problem, solved with a Differential Evolution Algorithm. The proposed approach is applied to a real household equipped with a rooftop photovoltaic system. Results show that our method achieves up to 48% additional economic savings and reduces visual and thermal discomfort by six and one orders of magnitude, respectively, compared to an unscheduled operation.