Supervised Machine Learning Calibration of Low-Cost Air Quality Sensors for Long-Term Placement Using Multi-Sensor Fusion

Low-cost PM sensors have garnered interest for their ability to reduce the cost of investigating PM concentrations in both indoor and outdoor spaces. They perform well in high concentration lab testing with correlation coefficients greater than 0.9. In real-world applications, the correlation coefficients drop significantly because of sensing floors and adverse ambient conditions. There are plenty of supervised machine learning techniques that aim to correct the measurements ranging from linear regression to more advanced neural networks and random forests. This work aims to use those more complicated techniques to adjust the measurements using other data sets gathered by a sensor suite. The Minnesota Pollution Control Agency (MPCA) has deployed a network of 47 AQ-Mesh sensors around the Minneapolis-St. Paul Metro Area. The network was active for two years, with mass colocations at a regional federal sensing site before and after deployment. The sensor suite includes electrochemical sensors for nitrogen monoxide (NO), nitrogen dioxide (NO2), carbon monoxide (CO), and ozone (O3). The suite also has an Alphasense OPC-N2 particle counter for PM measurements, along with temperature, pressure, and relative humidity sensors. Using these sensors in combination with basic supervised machine learning regression and a large dataset spanning over two years predictors are trained, applied, and examined for stability. This is an online presentation for faculty, staff, students, and the community on April 4, 2022 at 2:30 PM via Zoom (https://minnstate.zoom.us/j/96158391512?pwd=UzBwZy9DUHRUc2htU1B6K0lxMng0QT09)