Machine Learning and Deep Learning Models Applied to Photovoltaic Production Forecasting

The increasing trend in energy demand is higher than the one from renewable generation, coming years. One of greatest sources consumption are buildings. management a building by means production photovoltaic situ common alternative to improve sustainability this sector. An efficient trade-off source fields Zero Energy Buildings (ZEB), nearly (nZEB) or MicroGrids (MG) requires an accurate forecast production. These systems constantly generate data that not used. Artificial Intelligence methods can take advantage missing information and provide forecasts real time. Thus, manuscript comparative analysis carried out determine most appropriate On hand, Machine Learning considered Random Forest (RF), Extreme Gradient Boost (XGBoost), Support Vector Regressor (SVR). other Deep techniques used Standard Neural Network (SNN), Recurrent (RNN), Convolutional (CNN). models checked with building. validated using normalized Mean Bias Error (nMBE), Root Squared (nRMSE), coefficient variation (R2). deviation also conjunction these metrics. results show test set errors less 2.00% (nMBE) 7.50% (nRMSE) case considering nights, 4.00% 11.50% if nights considered. In both situations, R2 greater 0.85 all models.