Machine learning-driven photovoltaic generation forecasting for prosumer decision support

The problem of forecasting electricity generation is key to enabling decision-making support at the level of individual prosumers in the power grid and efficient prosumer integration into the grid. This study investigates the application of machine learning (ML) approaches to photovoltaic generation forecasting, aiming to provide general practical recommendations for the development of applied forecasting solutions. To this end, a specific use case was considered in the context of a private household with a photovoltaic installation, where data was gathered for several years. Based on the experimental results, a set of recommendations for applying ML models to photovoltaic generation forecasting tasks was formulated in the context of prosumer decision support. These recommendations address key aspects such as training and test data sizes used in the model creation process, and prediction horizon size used in the prediction process. In addition, guidelines on model file size were developed from the perspective of practical model utilization in specific use cases. This research demonstrates that establishing universal guidelines for ML model utilization in the Power System (PS) domain is both beneficial and achievable. It also highlights opportunities for further research on developing solutions for automated recommendations for required training data sizes and prediction horizons