Unsupervised NILM Implementation Using Odd Harmonic Currents

In this paper, an unsupervised non-intrusive load monitoring approach is proposed in order to encounter the disaggregation problem for Non-Intrusive Load Monitoring (NILM) methodologies, using odd harmonic current amplitudes. The problem has been contemplated as a multi-class multi-label one and for the combinations examined the number of appliances operating simultaneously varies between from one to up to three appliances. K-means has been utilized to cluster the different combinations, using additionally the elbow technique, in order to obtain the most suitable number of clusters that should be created. The results indicate that the proposed technique performs efficient load identification even with few samples at the training stage especially under the consideration of the third and fifth odd harmonic currents.