Machine learning strategies assessment for energy commodities forecasting

In recent years, the liberalization of the electricity market and the increasing penetration of distributed energy generation based on renewable sources have contributed to expand the uncertainty in the energy commodities trend. Effective methods able to make accurate predictions about electricity price and demand, as well as forecasting of the amount of power produced by photovoltaic systems and wind generators, would have a beneficial impact on the system management and would maximize profits for the actors involved in the energy exchange market. This work aims to test and compare different Machine Learning methods (Linear Regression, Support Vector Regression and Neural Networks) to obtain predictions of two energy commodities: zonal electricity price and PV power production. The algorithm that performed better in this sense (SVR) was then applied to a real life like case study to understand what kind of results can be obtained. Data that have been fed to algorithms are referred to Northern Italy. The total load curve comes from Terna, the Italian transmission system operator, while the zonal price (PZ) history has been provided by GME, the exchange for electricity and natural gas spot trading in Italy. The RES power production forecast was related to a PV field located in Fossano, Italy, whose production history was provided by EGEA SPA, owner of the system.