Primary Submission Category: Machine Learning and Causal Inference
Applying causal inference in architectural engineering – thermal comfort model and control
Authors: Ruiji Sun, Stefano Schiavon, Hui Zhang, Lei Shi, Thomas Parkinson,
Presenting Author: Ruiji Sun*
A rapidly growing volume of data from continuous building monitoring has been available due to their electrification and digitalization. The architectural and engineering community widely uses machine learning algorithms to transform high-dimensional data into models with high prediction accuracy. However, the main shortcoming of traditional machine learning techniques is their black-box nature which limits causal effect understanding. Understanding causality from continuous indoor environmental quality (IEQ) monitoring data would help create a sustainable, resilient, and healthy built environment. We propose a causal inference framework to study the long-term effect of IEQ on occupants’ comfort, health, and performance. We first developed a collaborative causal graph across multi-disciplines to summarize exited knowledge of IEQ. We investigated one aspect of IEQ called thermal comfort, which is the subjective perception of the thermal environment. The IEQ causal graph shows the drawbacks of various classical thermal comfort models. We identify the indoor thermal environment’s heterogeneous effects on long-term thermal comfort using d-separation. The structural causal model can be implemented in building mechanical systems to control the indoor environment within a comfortable temperature range.
Â