Regression Model for the Prediction of Pollutant Gas Concentrations with Temperature Modulated Gas Sensors

Air pollution presents an increasingly important issue on a global scale. Gas sensors based on Semiconducting Metal Oxides (SMOX), exhibit a high sensitivity and low limit of detection, making them an ideal candidate to detect pollutant gases to conform to guidelines published by the world health organization. In this work, we show that the temperature modulation of a single multi-pixel SMOX gas sensor is a cost and size efficient way to detect and quantify pollutant gases relevant for outdoor air quality in a broad range of concentrations. Roughly 1 700 hours of data were recorded and analyzed. A convolutional neural network regression model trained on 4 192 samples was able to predict pollutants in random gas mixtures with low mean relative errors (MRE): CO - 4.2 %, NO2 - 11.1 %, O3 - 13.6 %, and SO2 16.1 %. Relative humidity was predicted with an MRE of 2.5 %.

Automated summary

Air pollution is becoming an increasingly important issue globally. This study demonstrates that temperature modulation of a multi-pixel SMOX gas sensor is a cost-effective and efficient method for detecting and quantifying pollutant gases in a wide range of concentrations in outdoor air. A convolutional neural network regression model trained on a large dataset achieved low mean relative errors in predicting pollutants in random gas mixtures, as well as relative humidity.