Large-scale smart energy metering deployment worldwide and the integration of smart meters within the smart grid are enabling two-way communication between the consumer and energy network, thus ensuring an improved response to demand. Energy disaggregation or non-intrusive load monitoring (NILM), namely disaggregation of the total metered electricity consumption down to individual appliances using purely algorithmic tools, is gaining popularity as an added-value that makes the most of meter data.In this thesis, the first contribution tackles low-rate NILM problem by proposing an approach based on graph signal processing (GSP) that does not require any training.Note that Low-rate NILM refers to NILM of active power measurements only, at rates from 1 second to 1 minute. Adaptive thresholding, signal clustering and pattern matching are implemented via GSP concepts and applied to the NILM problem. Then for further demonstration of GSP potential, GSP concepts are applied at both, physical signal level via graph-based filtering and data level, via effective semi-supervised GSP-based feature matching. The proposed GSP-based NILM-improving methods are generic and can be used to improve the results of various event-based NILM approaches. NILM solutions for very low data rates (15-60 min) cannot leverage on low to highrates NILM approaches. Therefore, the third contribution of this thesis comprises three very low-rate load disaggregation solutions, based on supervised (i) K-nearest neighbours relying on features such as statistical measures of the energy signal, time usage profile of appliances and reactive power consumption (if available); unsupervised(ii) optimisation performing minimisation of error between aggregate and the sum of estimated individual loads, where energy consumed by always-on load is heuristically estimated prior to further disaggregation and appliance models are built only by manufacturer information; and (iii) GSP as a variant of aforementioned GSP-based solution proposed for low-rate load disaggregation, with an additional graph of time-of-day information.
|Date of Award||19 May 2020|
- University Of Strathclyde
|Sponsors||University of Strathclyde|
|Supervisor||Lina Stankovic (Supervisor) & Vladimir Stankovic (Supervisor)|