Assessing the Impact of Weather Forecast Models Combination on the AMS Solar Energy Prediction

Nowadays, using variable energy resources, particularly renewable but intermittent energy resources such as solar and wind power, have caused an irregular electricity generation which presents technical and financial challenges in the power storage, use and sale. Therefore, solar power forecasting is attracting more and more attention from both industry and academia. It becomes highly relevant in providing the ability to accurately predict power fluctuations of solar energy resources in the power grid in order to maintain a balance between daily supply and demand and to offer an efficient/low-cost storage of energy. For this aim, the American Meteorological Society (AMS) organized a kaggle competition to find the best machine learning and statistical techniques used for solar power forecasting based on weather input data. In this work we focused on the influence of Numerical Weather Prediction (NWP) models on solar power forecast, for this reason six NWP models combinations were examined within seven regressors to ultimately get $2.01\mathrm{E} +6\mathrm{J}/ \mathrm{m}^{2}$ as MAE score in Kaggle leaderboard.