Abstract Global warming, mainly caused by greenhouse gas emissions, presents an emergency in which green energies are expected to play a significant role. While renewable technologies such as wind, solar, and hydropower dominate green energy generation, geothermal energy is gaining attention as a viable and reliable alternative. When combined with heat pumps, geothermal systems offer a consistent heat supply unaffected by weather variability, making them particularly suitable for replacing fossil fuels in applications like air conditioning and domestic hot water in residential and commercial buildings. Geothermal energy’s growing prominence highlights its potential as a critical component in addressing the global climate crisis. This study examines the application of Deep Learning (DL) models to forecast the temperature of the input collector in a heat pump fed by a geothermal installation in a bioclimatic residential building, aiming to optimize energy management and minimize environmental impacts. In addition to a statistical baseline model, ARIMAX (Autoregressive Integrated Moving Average with Exogenous Variables), five different DL model architectures were developed, tuned, and evaluated: Gated Recurrent Unit (GRU), Bidirectional Long Short-Term Memory (BiLSTM), Transformers, LSTM-Transformer, and a hybrid approach that combines LSTM and Convolutional Neural Network. Among the experiments, the GRU model performed better, achieving a Root Mean Square Error (RMSE) of 2. 19^ C, with the LSTM-Transformer model following closely with an RMSE of 2. 23^ C. These results highlight both the GRU’s efficiency and the hybrid approach’s potential in balancing forecasting capability and effectiveness. These findings emphasize the critical role of predictive analytics in optimizing energy systems in bioclimatic contexts.
Oliveira et al. (Mon,) studied this question.