Investigation of indoor air quality in a low energy high school building combining micro gas sensors and unsupervised learning

Abstract

Because of their size and price, miniature gas sensors are good candidates for long-term, large-scale, continuous monitoring of the air quality in confined environments, even in the presence of occupants. In spite of their still somewhat limited metrological performances, these tools are able to provide relevant information on the pollutants spatial and temporal evolution. They can therefore be used to identify pollution sources and automatically control ventilation and remediation systems, provided they are associated with adequate data treatment procedures. In the present study, four sensors for the detection of CO2, NO, NO2 and O3 have been deployed, without previous calibration, inside a classroom of a low energy high school building, together with standard analytical instruments. The data are analyzed with a procedure based on the bisecting K-means algorithm. This unsupervised classification allows the identification of similar measurements, which can be merged into clusters. An excellent agreement has been found between the classification results provided by the analyzers and by the sensors, even if the latter were not calibrated before deployment. These results validate the data treatment methodology proposed in this work, and demonstrate the potential of using commercial micro sensors in real conditions.