Combined Machine Learning and weather models for photovoltaic production forecasting in microgrid systems

Given the high stochasticity of solar irradiance, the accurate forecasts of photovoltaic plants production can contribute to correctly manage power systems resources. In microgrids and grids, the availability of thorough forecasts allow to suitably operate generation, load and storage resources, avoiding grid imbalance conditions. In this work authors investigate the possibility to employ a Machine Learning approach that leverages the PV power forecast obtained by weather model and past observed generated power data. The proposed approach permits to obtain at least 3.7% improvement in photovoltaic production forecasting, demonstrating the methodology effectiveness.