Machine Learning for Predicting Indoor Air Temperature in Smart Buildings
Author Information
Author(s): Shakhovska Nataliya, Mochurad Lesia, Caro Rosana, Argyroudis Sotirios
Primary Institution: Lviv Polytechnic National University
Hypothesis
Can advanced machine learning techniques improve the accuracy of indoor air temperature predictions in smart buildings?
Conclusion
The proposed LSTM model with Rolling Window Cross-Validation effectively predicts indoor air temperature with high accuracy and low loss values.
Supporting Evidence
- The LSTM model demonstrated robust generalization capabilities with loss values ranging from 0.0004709 to 0.02819861.
- Adaboost and Gradient Boosting models outperformed linear regression in predicting indoor air temperature.
- The proposed approach allows for real-time monitoring and model adjustment to maintain predictive accuracy over time.
- Rolling Window Cross-Validation enhances the model's adaptability to evolving data trends.
Takeaway
This study shows how computers can learn to predict indoor temperatures better, helping to keep buildings comfortable and save energy.
Methodology
The study used LSTM networks with Rolling Window Cross-Validation for time-series modeling and evaluated model performance using metrics like MSE and R².
Potential Biases
Potential biases may arise from the dataset's limited diversity and the specific building types analyzed.
Limitations
The model may struggle with sparse or non-representative data, particularly in extreme climate conditions.
Participant Demographics
Data collected from two buildings in different seasons, with measurements taken from various floors.
Statistical Information
P-Value
0.03367
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website