Exploring Gaussian Process Regression for indoor environmental quality: Spatiotemporal thermal and air quality modeling with mobile sensing

Abstract

Measuring and monitoring indoor environmental quality (IEQ) is essential to ensure occupants’ health, well-being, and productivity in the built environment. Conventional IEQ assessment approaches, including stationary sensor networks and human-propelled field surveys, are limited in scalability in both spatial and temporal dimensions. To address this, researchers have developed mobile sensing platforms for more agile measuring and monitoring. However, a systematic modeling approach for reconstructing high-resolution spatiotemporal maps of IEQ variables from the sparse data collected by these mobile platforms is still lacking. This paper introduces a Gaussian Process Regression (GPR) framework designed to interpolate and process raw data from robotic mobile sensing platforms and IoT devices, thereby reconstructing spatiotemporal representations of various IEQ parameters from both short-term and long-term monitoring. The results demonstrate that for short-term spatiotemporal indoor air quality distribution reconstruction, GPR outperforms K-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Regression (SVR) when using different cross-validation strategies. In addition to a high correlation with the measurements and small normalized mean squared error, GPR models show a minimal bias in variance compared to the measured data. Secondly, when applied to long-term autoregressive modeling of an air temperature sensor network, GPR exhibits fast convergence and maintains a root-mean-square error of 0.21 °C over long-term predictions. The results suggest that GPR can capture the temporal and spatial variations in air temperature data. The two case studies demonstrate that the proposed approach holds the potential to serve as a robust and generalizable modeling framewoek to assist IEQ monitoring and assessments.