Enhancing Thermal Perception Predictions for Energy-Efficient Buildings with Machine Learning

Introduction

Maintaining a comfortable indoor climate in large buildings is paramount for occupant well-being and energy efficiency. However, achieving this balance often poses challenges for heating, ventilation, and air conditioning (HVAC) systems. Machine learning (ML) models offer a promising approach to predicting occupant thermal perceptions and improving HVAC efficiency. However, these models can be affected by biased human perception data and uncertainties, leading to inaccurate predictions and inefficient control.

Study Overview

Researchers from Carnegie Mellon University’s Civil and Environmental Engineering Department have proposed a method that combines data and models to address the limitations of current ML approaches. They employed Multidimensional Association Rule Mining (M-ARM) to identify and correct biases in human responses to temperature, thereby improving the accuracy of thermal perception predictions. This study was published in the journal Building and Environment.

Methodology

The researchers analyzed seven ML models and leveraged conflicting information provided by building occupants when answering multiple related questions about their thermal comfort. By utilizing M-ARM, they identified potential instances of subjective data biases and miscalibration issues associated with current methods. This approach enabled them to estimate the true “comfort zone” for most people in a building.

Results

The study found that the proposed method significantly improves the prediction reliability and reduces errors in existing ML models. The authors investigated the impact of various factors, including dataset size, classifier types, and calibration methods, to optimize the performance of the models.

Significance

The findings of this study provide valuable insights into advancing ML-based strategies for thermal perception predictions. The improved accuracy of these predictions can lead to better strategies for controlling temperature in buildings, resulting in increased occupant comfort and reduced energy consumption.

Implications

* Energy Savings: The proposed method has the potential to contribute to energy conservation by reducing excess energy consumption caused by defective data sets.

* Holistic Comfort Assessment: The study highlights the importance of considering multiple factors, such as humidity, temperature, and clothing, when evaluating human comfort.

* Reliable Predictions: By addressing subjective data biases and model predictive uncertainties, the method enhances the reliability of thermal perception predictions.

* Improved Control Strategies: The improved predictions enable the development of more effective strategies for controlling temperature in buildings, leading to enhanced occupant comfort and energy efficiency.

Conclusion

This research demonstrates the effectiveness of combining data and models to improve thermal perception predictions in large buildings. The proposed method, which utilizes M-ARM to correct biases in human responses, significantly enhances the accuracy of ML models. This work paves the way for developing more reliable ML-based strategies for controlling temperature in buildings, ultimately resulting in increased occupant comfort and reduced energy consumption.