Optimizing energy efficiency and occupant comfort in smart buildings using SMOTE-augmented deep learning approaches

Area of Science:

• Building Science
• Artificial Intelligence
• Sustainable Energy

Background:

• The construction sector significantly impacts energy consumption and greenhouse gas emissions.
• Intelligent, energy-efficient, and sustainable building systems are crucial.
• Deep learning shows promise for building energy management but faces limitations like class imbalance and limited model evaluation.

Purpose of the Study:

• Develop a robust deep learning framework for optimizing building energy efficiency and occupant thermal comfort.
• Address class imbalance issues in thermal comfort datasets.
• Improve the prediction accuracy of deep learning models for thermal comfort.

Main Methods:

• Implemented a unified framework including data preprocessing, Synthetic Minority Over-sampling Technique (SMOTE), and feature normalization.
• Comparatively evaluated multiple deep learning architectures: Deep Neural Networks (DNN), Deep Flatten DNN, Bi-LSTM, and Attention-based LSTM.
• Integrated SMOTE for class balancing with a comparative analysis of advanced deep learning models.

Main Results:

• The Attention-based LSTM model achieved the highest prediction accuracy (91%), surpassing Bi-LSTM (89%), Deep Flatten DNN (88%), and DNN (87%).
• Demonstrated superior Precision, Recall, and F1-score compared to other models.
• Achieved an 8-percentage-point improvement over a previously reported GNN model on the same dataset.

Conclusions:

• The proposed SMOTE-augmented framework enhances the accuracy and robustness of thermal comfort prediction.
• Offers an effective solution for intelligent, energy-efficient, and sustainable building energy management.
• Shows strong potential for practical application in real-world building systems.