Data analysis and interpretable machine learning for HVAC predictive control: A case-study based implementation

Jianqiao Mao*, Ryan Grammenos, Konstantinos Karagiannis

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

62 Downloads (Pure)

Abstract

Energy efficiency and thermal comfort levels are key attributes to be considered in the design and implementation of a Heating, Ventilation and Air Conditioning (HVAC) system. With the increased availability of Internet of Things (IoT) devices, it is now possible to continuously monitor multiple variables that influence a user’s thermal comfort and the system’s energy efficiency, thus acting preemptively to optimize these factors. To this end, this paper reports on a case study with a two-fold aim; first, to analyze the performance of a conventional HVAC system through data analytics; secondly, to explore the use of interpretable machine learning techniques for HVAC predictive control. A new Interpretable Machine Learning (IML) algorithm called Permutation Feature-based Frequency Response Analysis (PF-FRA) is also proposed. Results demonstrate that the proposed model can generate accurate forecasts of Room Temperature (RT) levels by taking into account historical RT information, as well as additional environmental and time-series features. Our proposed model achieves 0.4017 °C and 0.9417 °C of Mean Absolute Error (MAE) for 1-h and 8-h ahead RT prediction, respectively. Tools such as surrogate models and Shapley graphs are employed to interpret the model’s global and local behaviors with the aim of increasing trust in the model.
Original languageEnglish
Pages (from-to)1-21
JournalScience and Technology for the Built Environment
Early online date24 Aug 2023
DOIs
Publication statusE-pub ahead of print - 24 Aug 2023

Keywords

  • HVAC system
  • Indoor Temperature Prediction
  • Interpretable machine learning
  • Thermal Comfort
  • energy efficiency

Fingerprint

Dive into the research topics of 'Data analysis and interpretable machine learning for HVAC predictive control: A case-study based implementation'. Together they form a unique fingerprint.

Cite this