Predicting Zone Air Temperature in Smart Buildings Using LightGBM Models

Heming Zhang

Keywords: Smart Buildings, Temperature Prediction, LightGBM, Time Series Forecasting, HVAC, Sensor Data, Machine Learning, Exogenous Variables

TL;DR: This paper uses LightGBM models to accurately predict zone air temperatures in smart buildings with an average MAE of only 1.093°F, demonstrating the effectiveness of machine learning for HVAC optimization.

Abstract:

Efficient temperature monitoring and accurate prediction significantly enhance the management of smart building systems by optimizing energy consumption and improving occupant comfort. This study presents a systematic approach to predicting zone air temperatures in smart buildings using advanced machine learning techniques, with a focus on the LightGBM algorithm. Leveraging the Smart Buildings Dataset, we trained individual prediction models for zone-specific sensors, utilizing historical data and exogenous variables. The models exhibited exceptional predictive accuracy, achieving an average Mean Absolute Error (MAE) of 1.093°F, Root Mean Squared Error (RMSE) of 1.586°F, and an R² value of 0.994. This research underscores the applicability of machine learning for smart building systems and introduces a reproducible pipeline tailored for sensor-specific temperature prediction.
The source code has been uploaded to GitHub: https://github.com/BUSHIWOKALE/LightGBM.git

Submission Number: 33