Non-intrusive load monitoring based on switching voltage transients and wavelet transforms

Abstract

Continuous Wavelet Transform (CWT) analysis to find feature vectors for switching voltage transients for Non-Intrusive Load Monitoring (NILM) is presented and discussed, and compared with the previously used short time Fourier transform (STFT). The feature vectors computed from both CWT and STFT were used to train Support Vector Machines (SVMs) that identify the connection or disconnection of appliances for a NILM system. Experimental results show that the CWT analysis based on the complex Morlet wavelet improves classification accuracy as compared to the analysis based on STFT. More importantly, a 20× reduction of the vector size requirement is shown, thus greatly lowering computational requirements. It can be expected that commercial transient-based NILM will be based upon the CWT methods shown here.