The advancements of modern technology have motivated researchers, business and governmental stakeholders to envision the idea of a unified network-based platform, known as the Internet of Things, that can interconnect devices and allow the bi-directional communication between relevant parties. The information exchanged through the Internet of Things can originate from a variety of fields, namely healthcare, environmental and infrastructure monitoring, transportation and logistics, smart grid and smart houses, which can produce a vast amount of diverse data in real time.;One of the most important challenges, emerging from the implementation of the Internet of things, is the acquisition of meaningful information using the data derived from the various smart devices and sensors. Therefore, it is essential to identify suitable data processing and analysis approaches, able to adapt depending on the application-specific requirements, and consequently transform the available data into useful information that can be utilised by any various stakeholders.;The main focus of this research thesis was in the field of Non-Intrusive Load Monitoring (NILM) in order to provide solutions suitable for low resolution smart metering data, similar to the specifications of the smart meters selected for deployment from most utilities and governmental stakeholders. Through an extensive and up-to-date review of the NILM field presented in this thesis, it has been identified that only recently, researchers have focused on disaggregating using only active power aggregate data for feature extraction at low sampling rates.;Therefore three unsupervised NILM methods were proposed as an outcome of this research, one using solely Dynamic Time Warping (DTW), a signal processing-based method, while the other two methods propose a combination of DTW and k-means, namely DTW+kM and kDTW, in order to address the computation complexity observed using the DTW-based method. The DTW+kM approach performs DTW for creating a library of appliance signatures and classification via clustering using k-means, and the kDTW is incorporating a DTW refinement post processing step in order to optimise the performance of the initial implementation.;The proposed methods were evaluated and benchmarked against various state-of-the-art NILM methods using the publicly available REDD  and REFIT [2, 3] datasets, and reported good performance.;Furthermore, the research presented in this thesis has investigated two other heterogeneous applications of the Internet of Things with respect to the emerging data challenge, and have proposed customised monitoring systems, and a variety of signal processing and machine learning approaches for analysing the corresponding data. More specifically, a prototype monitoring system was proposed for monitoring earthwork assets and preliminary findings reported in this thesis. In the context of visual content interaction using a video based eye tracking device, applicable in healthcare, computing,and even advertisement, the user's attention was evaluating using various statistical and signal processing methods, with the wavelet-based analysis being the best contestant for identifying features for extraction using pupil dilation and gaze fixation.
|Date of Award||28 May 2020|
- University Of Strathclyde
|Sponsors||EPSRC (Engineering and Physical Sciences Research Council)|
|Supervisor||Lina Stankovic (Supervisor) & Stuart Galloway (Supervisor)|