Time-frequency analysis techniques for non-intrusive load monitoring

Abstract

The work in this thesis examines time-frequency analysis techniques and in particular the wavelet transform to extract the features contained within the electrical load signals. A novel approach that is based on wavelet design was utilized to generate a wavelet library which was used to match each load signal to a specific wavelet using Procrustes and covariance analysis. In order to automate the load identification process, two machine learning classifiers representing an eager learner and a lazy learner were used in this work. The proposed wavelet design concept has been verified experimentally, and the results of implementing the proposed load detection and classification approach shows significant improvement in the classification accuracy compared to other existing detection approaches reaching an overall accuracy of 98%.