Data-Driven Photovoltaic Power Production Nowcasting and Forecasting for Polygeneration Microgrids

In this paper, we deal with the problem of nowcasting and forecasting the photovoltaic power production (PvPP) on the basis of real data available for the Savona Campus and coming from the energy management systems (EMSs) of the smart polygeneration microgrid that feeds buildings in the University area. In this paper, we show how PvPP nowcast and forecast problems can be solved with the state-of-the-art data-driven techniques, which use the historical data collected by the EMS. In particular, we compare the performance of the kernelized regularized least squares, the extreme learning machines, and the random forests. In the machine learning field, these algorithms are the best choice in three different families of techniques: kernel methods, neural networks, and ensemble methods. Results show that our proposal can improve of almost one order of magnitude to the actual prediction system used in the EMS of the Savona Campus, which is based on the knowledge of the physical problem. Finally, by using the EMS installed at the Savona Campus, it has been possible to quantify the saving in costs and CO₂ emissions due to the new nowcasting and forecasting models.