IoT for fresh water quality monitoring

Abstract: Water is one of the most important resources in the world. It has direct impact on the daily life ofmankind and sustainable development of society. Water quality affects biological life and has to obeystrict regulations. Traditional water quality assurance methods, used today, involve manual samplingfollowed by laboratory analysis. This process is expensive due to high labour costs for sampling andlaboratory work. Moreover, it lacks real time analysis which is essential to minimise contamination.This thesis aims to find a solution to this problem using IoT sensors and Machine Learning techniquesto detect anomalies in the water quality. The spatial scalability is key requirement when selecting transmissionprotocols, as sensors could be spread around the water network. We consider solutions readilyavailable or soon to be in the market. The key LPWAN technologies studied are: SigFox, LoRaWANand NB-IoT. In general these protocols have many characteristics essential for fresh water monitoring,like long lasting battery life and long range, however, they have many limitations in terms of transmissiondata rates and duty cycles. It is therefore essential to find a solution that would correctly find anomaliesin the water quality but at the same time comply with limited transmission and processing capabilities ofthe node sensors and above mentioned protocols.A trial sensor is already in place in lake M¨alaren and its readings are used for this study. Supervisedmachine learning algorithms such as Logistic Regression, Artificial Neural Network, Decision Tree, OneClass K-NN and Support Vector Machine (SVM) are studied and discussed regarding the data available.SVM is then selected, implemented and optimised to comply with the limitations of IoT. The trade offbetween false anomalies and false normal readings was also discussed.