Thesis

Low-rate non-intrusive load monitoring approaches via graph signal processing

Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2020
Thesis identifier
  • T16108
Person Identifier (Local)
  • 201569486
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • The large-scale roll-out of smart metering worldwide brings many new application possibilities. One promising application is appliance-level energy feedback based on identifying individual loads from aggregate measurements. Driven by high application potentials, the research in this area has intensified. In particular, non-intrusive load monitoring (NILM), that is, estimating appliance load consumption from aggregate readings, using software means only, has attracted a lot of attention, since it does not require any additional hardware to be installed. This thesis first proposes two Graph Signal Processing (GSP)-based approaches for disaggregation of total energy consumption down to individual appliances used. The first approach uses the Graph Laplacian Regularisation (GLR) minimiser results as a starting point, adding further refinement via Simulated Annealing (SA). The second approach applies data segmentation and associates data segments with graph nodes. A Dynamic Time Warping (DTW) distance is applied for evaluating weights between graph nodes. GLR minimiser is again used for clustering. Finally, a generic optimisation based approach is proposed for improving the accuracy of existing NILM by minimising the difference between the measured aggregate load and the sum of estimated individual loads with the difference from original NILM approaches' results as regularisation. For all proposed methods, the competitive performance are demonstrated in terms of both accuracy and effciency compared to state-of-the-art approaches, using the public Personalised Retrofit Decision Support Tools For UK Homes Using Smart Home Technology (REFIT) dataset and Reference Energy Disaggregation Dataset (REDD) electrical load datasets.
Advisor / supervisor
  • Stankovic, Lina
  • Stankovic, Vladimir
Resource Type
DOI

Relations

Items