Applications of Machine Learning Algorithms in Wireless Sensor Networks

Abstract

As a key infrastructure of constructing the Internet of Things, wireless sensor networks (WSNs) have attracted plenty of research interest. It is expected to be widely applied in almost every aspect of future life. Nowadays, some preliminary applications and prototypes have emerged in various fields from military applications to health and environmental applications, etc. However, many conceptual and practical problems are still required to be solved. In WSNs, some common research problems include sensor routing and clustering, data fusion, sensor localization, communication signal processing, intelligent event detection and decision making, etc. Recently, with the booming of cloud computing, Machine Learning (ML) based methods have arisen to provide many novel and effective solutions for a variety of problems in WSNs. The advantages of ML based methods are promising and can significantly boost the application and development of WSNs. In this thesis, three topics in WSNs are mainly studied. Firstly, the problem of redundant transmission reduction is studied and ML methods have been applied in the proposed prediction-based data fusion to reduce the number of wireless transmission. Secondly, the problem of communication channel equalization is studied, ML based equalization methods have been proposed and discussed. Finally, the problem of intelligent event detection is studied in the WSN-based fluid pipeline leak monitoring applications where Deep Learning (DL) and enhanced model based leak detection and localization methods are proposed and discussed.