Comparison of PV Power Generation Forecasting in a Residential Building using ANN and DNN
Inês Tavares, Ricardo Augusto Manfredini, José Almeida, João Soares, Sérgio Ramos, Zahra Foroozandeh, Zita Vale
Abstract
Due to the fast growth of energy consumption in buildings, it is crucial to ensure sustainability demands through the use of renewable energies. The solar energies have been standing out and, as a result, the forecast of photovoltaic (PV) production has received broad attention. However, the intermittent nature of the generated power brings some uncertainty. This paper presents two PV generation forecasting methodologies based on a multi-layer feed-forward Artificial Neural Networks (ANN) and a Deep Neural Networks (DNN) combined with a Convolution Neural Network layer and a Recurrent Neural Network layer. Both techniques were implemented based on a data set of a PV production from a panel installed at a residential building. The main objective of this paper is to analyze and compare the forecasting results precision of both techniques. The accuracy of both models was evaluated through the calculated errors. The comparative analysis between the two networks demonstrated that the ANN technique is capable of predicting the PV generation with low forecasting errors.