NUST Institutions Library Catalogue

The aim of this study was to explore the use of data-driven approaches for designing net zero
energy buildings using bifacial solar panels and deep learning models. The study utilized
historical data on solar irradiance to forecast future irradiation levels, which were then used to
optimize the tilt angle of bifacial solar panels for maximum energy collection. Two deep learning
models, namely RNN-LSTM and transformers, were evaluated using various evaluation metrics,
including MAE, RMSE, SMAPE, and R2. The results showed that both models were effective in
forecasting irradiations with varying forecasting horizons, with the transformers model
outperforming the RNN-LSTM model in terms of MAE values. The study provides insights into
the use of data-driven approaches for designing net zero energy buildings, highlighting the
potential of deep learning models in optimizing the use of renewable energy sources for
sustainable development. The results show that both models were effective in forecasting
irradiance, with R2 values of 0.927 for RNN-LSTM and 0.894 for Transformers. Additionally,
the Transformers model outperformed the RNN-LSTM model with a lower range of MAE values
from 0.05 to 0.03 across the same horizons. These findings suggest that deep learning models
can effectively forecast solar irradiance and aid in optimizing the performance of net-zero energy
buildings.