Renewable Energy Forecasting: A Functional Data Analysis Approach with Climatic Variables

Given the almost irreversible state of climate change, now more than ever it is necessary to adopt habits that have the lowest possible impact on the natural environment. Most energy suppliers base their production on fossil fuels, which have severe negative consequences on the global ecosystem. A promising alternative to these resources is renewable energy sources: these derive from natural resources and, for this reason, are considered regenerative, free, and zero-impact. However, they are unfortunately irregular and discontinuous due to the influence of climate and the surroundings where the energy plants are located; therefore, the output power is intermittent by nature. So, to better organize and manage energy dispatching, it is important that suppliers rely on models for forecasting energy production to cope with unforeseen events and power drops. This case study aims to predict the output power of a photovoltaic plant through Functional Data Analysis, a branch of statistics that treats observations as functions. Indeed, the data that led to this application are curves, representing the daily trend of certain climatic variables, such as temperature and solar irradiance. The outcome of the case study highlights how functional models successfully capture data variability while maintaining high generalization capacity, which is reflected in the excellent evaluation metrics. A comparison was made with benchmark models, namely generalized additive models, from which the superiority of functional models can be inferred in terms of forecasting accuracy and percentage of explained variability. A potential future development of this application is the implementation of models that allow the prediction of climatic variables, which are essential to conducting this case study.