Date of Defense
11-2024 2:30 PM
Location
F1-1164
Document Type
Thesis Defense
Degree Name
Master of Electrical Engineering (MEE)
College
COE
Department
Electrical and Communication Engineering
First Advisor
Prof. Hussain Shareef
Keywords
Non-Intrusive Load Monitoring (NILM), Phase-Space Reconstruction (PSR), Two-Dimensional Fourier Series, Feature Extraction, Energy Monitoring, Classification, Load Disaggregation, Load Signature.
Abstract
The growing need for energy and efficient energy control has emphasized the importance of tracking appliance-level energy usage. The capability of Non-Intrusive Load Monitoring (NILM) to separate energy usage data per appliance from a single measurement provides a practical solution. This thesis explores developing a novel NILM method using Phase-Space Reconstruction (PSR) and 2-D Fourier Series to enhance feature extraction from the steady-state current waveforms. Existing NILM techniques frequently encounter accuracy challenges caused by overlapping power signatures of appliances and complex operational states. The proposed method is designed to efficiently capture the steady-state characteristics of electrical appliances. It is evaluated on the COOLL dataset, which includes high resolution current and voltage measurements for a variety of appliances, using basic machine learning algorithms like K-Nearest Neighbors (KNN) and Decision Trees (DT). The findings show that the use of a combination of PSR and 2-D Fourier series to extract appliances features has enhanced the classification accuracy, achieving a peak of 99.88% when the number of harmonics and the time delay for PSR are optimal. The results suggest that a relatively large time delay produces best representation for the current waveforms and the initial few harmonics are sufficient to efficiently capture load signatures. Future studies should explore the method using various datasets, apply feature selection techniques to reduce dimensionality and implement a real-time system to confirm the method effectiveness.
DEVELOPMENT OF A NON-INTRUSIVE LOAD MONITORING TECHNIQUE USING PHASE-SPACE-RECONSTRUCTION AND 2-D FOURIER SERIES CURRENT WAVEFORM FEATURES
F1-1164
The growing need for energy and efficient energy control has emphasized the importance of tracking appliance-level energy usage. The capability of Non-Intrusive Load Monitoring (NILM) to separate energy usage data per appliance from a single measurement provides a practical solution. This thesis explores developing a novel NILM method using Phase-Space Reconstruction (PSR) and 2-D Fourier Series to enhance feature extraction from the steady-state current waveforms. Existing NILM techniques frequently encounter accuracy challenges caused by overlapping power signatures of appliances and complex operational states. The proposed method is designed to efficiently capture the steady-state characteristics of electrical appliances. It is evaluated on the COOLL dataset, which includes high resolution current and voltage measurements for a variety of appliances, using basic machine learning algorithms like K-Nearest Neighbors (KNN) and Decision Trees (DT). The findings show that the use of a combination of PSR and 2-D Fourier series to extract appliances features has enhanced the classification accuracy, achieving a peak of 99.88% when the number of harmonics and the time delay for PSR are optimal. The results suggest that a relatively large time delay produces best representation for the current waveforms and the initial few harmonics are sufficient to efficiently capture load signatures. Future studies should explore the method using various datasets, apply feature selection techniques to reduce dimensionality and implement a real-time system to confirm the method effectiveness.
