Application of wavelet-based classification in non-intrusive load monitoring

In this paper, non-intrusive load monitoring using a single point sensing and wavelet-based classification is presented and applied to a test system feeding two dynamic and two static three-phase loads. The features in the three-phase voltage and current signals are extracted by wavelet transform to decompose the original signals. The energy of the obtained wavelet coefficients at the detail levels constitute a feature set for classification. Decision trees representing ensemble classifiers are then developed using the wavelet-based features and the performance of each ensemble classifier is then evaluated in the presence of several power quality disturbances resulting from applying several faults types and at different locations. The results have shown that higher order Daubechies wavelets, and in particular Daubechies of order 5, as well as more decision trees in the ensemble classifier both contribute to more accurate classification.