Day-ahead Irradiance and Electrical Demand Prediction for Sustainable Energy Communities

Recent progress in Machine Learning has led to innovative strategies in data-driven energy management systems, addressing challenges posed by intermittent renewable energy sources and fluctuating electric demand. However, accurate forecasting is crucial for scheduling algorithms of sustainable energy communities. This study introduces a specialised forecasting model for Data-driven Energy Management Systems, tailored to effectively operate in the highly unpredictable meteorological conditions along the southwestern Atlantic coastline in Ireland. Leveraging both raw and engineered data features of a real-world campus-based Microgrid system from July 2021 to February 2022, this research evaluates the performance of the long short-term memory algorithm with dropout regularisation for day-ahead forecasts of irradiance and electrical consumption. The performance of the regularisation-based long short-term memory model is compared to state-of-the-art Machine Learning and Deep Learning models. The simulation results show that the regularisation-based long short-term memory is the best prediction model for both solar irradiance and electrical demand.