An intelligent Internet of Things air pollution monitoring system: case of a local municipality in South Africa

Abstract

Air pollution is fast becoming a major challenge in the 21st century, with its impending threats to the atmosphere and societal health at large. Recently, there have been extensive studies on the monitoring of air pollution and air quality. This interest is owing to the numerous issues and dangers encountered globally because of air pollution and poor air quality. As a fast-developing country, South Africa has been faced with the problem of air pollution that needs to be addressed to avoid the devastating effects on its populace. This dissertation therefore presents the results of a study designed to develop an intelligent air monitoring technique to verify the quality of air pollution among the public. This development was driven by a detailed analysis of existing air quality monitoring methods with a focus on the use of artificial intelligence that has become a part of today’s society. The final product is an AI-based air quality surveillance system that is user-friendly for both society and consumers. The Northwest region of South Africa is known for its mining industries. This is where agriculture and the provincial primary economic activities take place. As a result, air pollution has been a growing concern, causing diseases like bronchitis, asthma exacerbation, cardiovascular diseases in underweight infants, and even mortality. The evidence of negative health impacts and mortality from these mining activities and other transport emissions cannot be disputed. The primary activity within the Northwest Province is confined to the main routes and mining industries. These include the Platinum Highway N4, N12 and N14, and Northwest province are focused on the Rustenburg and surrounding areas, which is over 90% of mining activities in the province. These activities have resulted in high levels of air pollution and the area is often tagged as an air pollution hot spot. The National Ambient Air Quality Standards (NAAQS) have sought to address the effects of air pollution on human health, but it is unclear what effects air quality has on terrestrial life. This research focuses on areas with high concentrations of air pollution, namely, particulate matter (PM), carbon dioxide (CO2), ozone (O3), nitrogen dioxide (NO2), and sulphur dioxide (SO2). Therefore, this study explains the need to implement easy, low-cost methods to monitor and predict early hazardous symptoms. By utilising machine learning methods, the internet of things, and artificial intelligence systems more frequently to reduce air pollution, we can help the public (both commercial and residential), who are worried about their welfare by taking preventative measures to avert potential dangers. These technological advances will be particularly useful in predicting the state of air quality to quickly identify the probable source of expected problems, such as machine learning and the Internet of Things (IoT). As part of this research, a cost-effective IoT-based system for real-time monitoring of air quality and control was designed. The designed hardware is equipped with five sensors, including a QM9, MQ2, Boost Voltage regulator, PM2.5 Air Quality Sensor, Adafruid Feather 32u4 FRM9x LoRa. A LoRa radio module is also added to facilitate radio transmission of sensor information from the Rustenburg Air Quality Station to the data collection facility. The innovation of technologies, such as the Internet of Things (IoT) and Machine Learning, has made it possible to monitor the quality of air contaminated in near real-time. This dissertation has also explored the use of Random Forest Regressor and Multi-layer perceptron Neural Networks for predicting and monitoring the quality of air pollutants. Furthermore, algorithms such as Support Vector Machine, K-means clustering, Decision Tree, Regression Tree, and Linear regression are discussed. The study sought to use Fourth Industrial Revolution (4IR) tools and technologies like IoT, Artificial, and computational intelligence to develop an intelligent air pollution monitoring system. Random Forest outperforms other models with a prediction error of 5% less than the next best-performing model for the first prediction, 12% less for the second prediction,7% less for the fourth prediction,34% less for the fifth, and 36% less for the sixth prediction. Although it came second in the Neural Network Time Series (NNTS) in the third prediction test with an error percentage of 10.41% as compared to NNTS an error percentage of 6,71% it had the most stable accuracy with a maximum prediction error of less than 15%.