An integrated low-cost air quality sensor and a multi-task calibration framework for particulate matter

The adverse effects of air pollution on human health are significant, and low-cost sensors (LCSs) have gained attention as an effective supplementary monitoring approach. However, most existing LCSs require external power sources, which restricts their deployment flexibility. Due to cost limitations, the data quality from LCSs is often lower and necessitates calibration. Previous calibration studies have primarily focused on incorporating meteorological variables into the models, with limited attention to individual sensor differences and spatial positioning. In this study, an integrated wireless LCS was developed, capable of monitoring multiple indicators without dependence on external power. A multi-task calibration framework based on TabNet was proposed to calibrate fine particulate matter (PM_2.5) and inhalable particulate matter (PM₁₀) simultaneously. Compared with using only meteorological variables, the inclusion of sensor difference (SD), spatial autocorrelation (SA), and solar zenith angle (SZA) factors improved the average R-squared (R²) value for PM_2.5 from 0.919 to 0.942, and for PM₁₀ from 0.909 to 0.938. The models exhibited stable performance under ten-fold cross-validation and generalized well in field tests. Results from various generalization strategies indicated that closer proximity does not necessarily lead to better generalization performance. This study provides valuable insights for urban air quality monitoring and research related to human health.