On AutoMLs for Short-Term Solar Radiation Forecasting in Brazilian Northeast

Within the context of clean energy generation, solar radiation forecast is applied for photovoltaic plants to increase maintainability and reliability. Machine learning models applied to solar radiation time series helps to improve forecast results. AutoMLs are becoming very popular for industrial application, because it simplifies some the challenge of finding the best suited model. This work presents three different optimizations of known models applied for solar radiation forecasting, ACOLSTM, ACO-CLSTM and AGMMFF. A methodology was used to obtain the results, with Brazilian INMET hourly scaled data, which were compared to the open sources AutoMLs TPOT, HPSKLEARN and AutoKeras. The obtained results for the presented models is promising for use in automatic solar radiation forecasting systems since ACOLSTM was found to outperform the compared models.