Towards a data-driven and scalable approach for window operation detection in multi-family residential buildings

TL;DR

This paper introduces an unsupervised, sensor-based method for detecting window operations in residential buildings, improving scalability and robustness over traditional models, and supporting natural cooling strategies.

Contribution

The paper presents a novel unsupervised approach using off-the-shelf sensors to detect window operations, outperforming SVM models in key scenarios and enhancing scalability.

Findings

Outperforms SVM in key indicators

Robust performance with indoor temperature data

Potential to enable scalable natural cooling solutions

Abstract

Natural cooling, utilizing non-mechanical cooling, presents a low-carbon and low-cost way to provide thermal comfort in residential buildings. However, designing naturally cooled buildings requires a clear understanding of how opening and closing windows affect occupants' comfort. Predicting when and why occupants open windows is a challenging task, often relying on specialized sensors and building-specific training data. This limits the scalability of natural cooling solutions. Here, we, propose a novel unsupervised method that utilizes easily deployable off-the-shelf temperature and humidity sensors to detect window operations. The effectiveness of our approach is evaluated using an empirical dataset and compared with a state-of-the-art support vector machine (SVM) model. The results demonstrate that our proposed method outperforms the SVM on key indicators, except when indoor and outdoor temperatures have small differences. Unlike the SVM's sensitivity to time series characteristics, our proposed method relies solely on indoor temperature and exhibits robust performance in pilot studies, making it a promising candidate for developing a highly scalable and generalizable window operation detection model This work demonstrates the potential of unsupervised data-driven methods for understanding window operations in residential buildings. By enabling more accurate modeling of naturally cooled buildings, our work aims to facilitate the widespread adoption of this low-cost and low-carbon technology.