Analysis of Artificial Neural Networks Applied in Energy Consumption of Single-Family Houses

The rapid population growth increased electricity consumption in households, along with sustainable development goals (SDGs) to ensure sustainable management of energy with minimal energy waste, have given fundamental value to accurately predicting energy consumption for better planning, production, and conservation. Artificial neural networks (ANNs) have proven to be ideal for energy consumption forecasting. In this study, a deep and up-to-date literature review has been conducted. Additionally, through an experimental design, 136 models have been configured and trained, considering different hyperparameters with four different dataset resolutions (15, 30, 45, and 60 minutes) for the Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM) neural networks, and a variant of LSTM, the LSTM Encoder-Decoder, known as Sequence to Sequence (Seq2Seq). Finally, forecast values for energy consumption are obtained for each of the mentioned resolutions, the employed neural techniques, and their minimal errors. The LSTM recurrent neural networks achieved better error metrics compared to the Multilayer Perceptron technique. Likewise, it maintained minimal differences with the LSTM Encoder-Decoder.