Application of Artificial Neural Networks to a Gas Sensor-Array Database for Environmental Monitoring

A sensors array based on two different types of chemical sensors such as tin dioxide commercial sensors and carbon nanotubes innovative sensors developed in the ENEA laboratories to monitor gases (e.g., CO, NO2, SO2, H2S and CO2) of relevance in polluted air has been analyzed. Measurements of chemical sensing of the sensors array have been performed in laboratory to create a database for applying artificial neural networks (ANNs) algorithms to quantify gas concentration of individual air pollutants and binary gas-mixture. A total number of 3,875 data-samples based on 413 distinct gas concentrations measured by 14 gas sensors has been used in the database. The ANN performance has been assessed for each targeted air-pollutant. The lowest normalized mean square error (NMSE) of 6%, 9% and 11% has been achieved for NO2, SO2 and CO2, respectively. In the contrast, NMSE as high as 28% and 39% has been measured for CO and H2S, respectively. The aim of this study is the selection of an optimal set of gas sensors in the array for enhanced environmental measurements of gas concentration in real-scenario.