By Nidhi DhullReviewed by Susha Cheriyedath, M.Sc.Sep 3 2024A recent article in Scientific Reports introduces an innovative predictive machine learning (ML) model, using active learning to design green buildings (GBs) with optimized energy efficiency and improved indoor sustainability.
Study: Active learning-based machine learning approach for enhancing environmental sustainability in green building energy consumption. Image Credit: Ground Picture/Shutterstock.com
Background
Green building (GB) techniques are vital for reducing energy waste in the construction sector, which is responsible for nearly 40 % of global energy consumption. Despite this, factors like occupant behavior and gaps in energy management can cause GBs to consume up to 2.5 times more energy than anticipated. To create a more sustainable and economically viable future, it is crucial to integrate innovative technologies with GB principles.
Artificial intelligence (AI) technologies, particularly machine learning (ML) and deep learning (DL), allow computers to learn from data and make predictions without explicit programming. These versatile tools are used across industries to drive innovation and improve efficiency. When applied to green buildings, ML and DL can significantly enhance energy efficiency and environmental sustainability.
Using predictive modeling and optimization algorithms, these AI techniques can optimize the use of renewable energy sources like solar and wind, reducing reliance on traditional power and promoting energy independence. In this study, active learning-based ML methods were employed to create a highly accurate model for predicting energy consumption and maintaining optimal indoor environments in GBs.
Methodological Framework
Developing an AL model for GBs involved several stages, including dataset visualization, preprocessing, model selection, and implementation using ML methods.
The dataset preprocessing was carried out through feature normalization to standardize the data. During model implementation, the dataset was split into 70 % training data and 30 % testing data. An AL-based ML model was then trained to accurately identify energy usage in GBs.
The AL strategies were implemented using an ActiveLearner object constructed with eight regressor models: random forest (RF), extreme gradient boosting (XGBoost), decision tree (DT), logistic regression (LR), gradient boosting regressor (GBR), K-nearest neighbor (KNN), CatBoost (CB), and Light GB machine (LGBM).
These regressor models iteratively analyzed the dataset to evaluate energy efficiency classes and train the ActiveLearner object. A three-iteration process was employed to label uncertain data samples. Subsequently, key metrics, including prediction accuracy (R2), mean squared error (MSE), mean absolute error (MAE), and relative absolute error (RAE), were calculated for each regressor. The performance of the regressors was then compared using an evaluation dataset.
Energy efficiency data, generated by Angeliki Xifara and analyzed by Athanasios Tsanas, was utilized to predict energy consumption in GBs. A total of 12 building shapes—varying in glazing area, glazing distribution, building orientation, and more—were simulated for energy analysis. In total, 768 building forms and 8 features were generated. For each feature, two states were forecasted using the regressors: heating load (Y1) and cooling load (Y2).
Results and Discussion
The statistical evaluation of the regressors revealed that the RF model achieved high prediction accuracy, with an R2 of 0.9950 in the first iteration. Its average errors were modest, with a MSE of 0.5445 and MAE of 0.3887, showing a low relative error to the true values. RF maintained excellent performance in the second and third iterations, with R2 values of 0.9947 and 0.9951, respectively, alongside slightly different error rates.
The GB model also demonstrated excellent R2 values, though with slightly higher error rates than RF in the first iteration. On the other hand, the KNR attained the best R2 (0.9700), MSE (3.2378), and MAE (1.1017), along with a RAE of 0.0484 across all three iterations. Meanwhile, the CB regressor performed exceptionally well, with an R2 of 0.9975, indicating a strong correlation between predicted and actual values. Its MSE (0.2667), MAE (0.2984), and RAE (0.0131) further reflected the low error between predicted and true values.
The prediction accuracy of the regressors varied between the Y1 and Y2 variables. The RF model achieved an R2 of 0.9951 for Y1 and 0.9765 for Y2, while the GB model showed consistent performance for both Y1 and Y2, with R² values of 0.9700 and 0.9776, respectively. CB performed the best overall, with an R2 of 0.9975 for Y1 and 0.9883 for Y2.
The XGBoost model achieved an R2 of 0.9941 for Y1 and 0.9798 for Y2, while LR performed moderately, with R2 values of 0.9666 for Y1 and 0.9584 for Y2. The Light GBM and DT models had higher accuracy for Y1 (0.9936 and 0.9948, respectively) than for Y2 (0.9756 and 0.9617, respectively).
Despite the promising performance of the proposed models, they had certain limitations. Primarily, their accuracy depended on data quality; incomplete or noisy data could impair performance. Additionally, intensive computation involved in AL-based methods is challenging for real-time applications or buildings with limited resources.
Conclusion
Overall, the researchers leveraged AI to predict and optimize GB energy consumption. The CB regression model demonstrated exceptional prediction accuracy of 0.9975 for cooling and 0.9883 for heating. The incorporated data visualization and preprocessing techniques strengthened the model’s reliability, highlighting its efficacy in advancing environmental sustainability within GB design.
The researchers suggest that the proposed AL method could have ripple effects beyond improving energy efficiency in sustainable architecture, including potential cost savings and a reduced carbon footprint.
Journal Reference
Mahmood, S., Sun, H., Alhussan, A., Iqbal, A., & El-kenawy, El-S. M. (2024). Active learning-based machine learning approach for enhancing environmental sustainability in green building energy consumption. Scientific Reports, 14(1), 1–18. DOI: 10.1038/s41598-024-70729-4, https://www.nature.com/articles/s41598-024-70729-4
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.