Air quality measurement and prediction system using artificial neural networks (#1938)

Authors

Astocondor-Villar, Jacob

Canales Escalante, Carlos Andres

Vilcahuaman-Sanabria, Raul

Solis-Farfan, Roberto

Benites-Gutierrez, Miguel

Ipince-Antunez, Daniel

Gomero-Ostos, Nestor

Abstract

This research project is carried out to measure the contaminated air and the concentration of suspended particles ranging between 2.5μg and 10μg also known as PM10 and suspended particles smaller than 2.5μg known as PM2.5, in the, district of Ventanilla and Mi Peru, in Peru. The work consists of measuring CO2 and PM2.5 and PM10 pollution to prevent the health of the inhabitants of the area under study. The implementation of a system to measure the CO2 contaminated air and the concentration of PM10 and PM2.5 pollutants is carried out. The measurement system consists of a dust and CO2 sensor, the system also includes an ambient temperature and humidity sensor, a DHT11 sensor for the measurement of ambient temperature and humidity, and an ESP8266 module for wireless recording and cloud recording. The sensor values are processed by an arduino uno R3 card and ESP8266 via wifi. A cloud computing PaaS service offered by Google and its registry a Google Sheets spreadsheet. An ANN was chosen because they have been shown to be effective when applied to air quality predictions. Compared to other similar work, only one network was realized, but several prototypes were developed and evaluated to avoid arbitrariness in design decisions. Three particular aspects of NR design were experimented: data normalization, architecture selection and activation function selection. Finally, the prediction of PM10 and PM2.5 particulate matter concentrations is performed using artificial Neural Networks. In the present project, the structure of a multilayer ANN consisting of an input layer, an intermediate layer and an output layer (8 - 16 - 1) is used. The programming was done in the Matlab Neural Networks toolbox.