CALIBRATION-FREE AND ADAPTIVE DRIFT COMPENSATION IN SENSOR NETWORKS FOR MONITORING AEROSOL PARTICLES

Abstract

This article discusses methods for compensating for errors in distributed sensor networks of inexpensive PM2,5 sensors designed to monitor air quality in indoor and urban environments. The main objective of the study is to develop and experimentally test reference-free and adaptive calibration approaches aimed at reducing inter-node systematic shifts, correcting the nonlinearity of transmission characteristics, and compensating for temporal drift in conditions of limited availability of reference measurements. Experimental studies were conducted on a network of twelve PMS5003 sensor nodes in a controlled laboratory environment with variable aerosol concentrations and stabilized temperature and humidity parameters.

The article presents spatial maps of PM2,5 concentration distribution and time series of measurements before and after the application of various compensation methods. Analysis of the results showed that adaptive neural network approaches and self-calibration methods provide a more stable reduction in error and smoothing of network artifacts compared to classical linear and polynomial models, especially in conditions of drift and changes in the operating environment.

The proposed methodology contributes to improving the reliability of aggregated sensor network indicators and can be used to support engineering decisions in environmental monitoring systems and smart buildings.