Automatic Power Signature Analysis using Prony's Method and Machine Learning-Based Classifiers

The selection of the feature extraction method that defines the electric power signature of an appliance is a fundamental step for the Non-Intrusive Load Monitoring (NILM) problem. In general, harmonic content, damping and transient switching features are normally used to classify different loads connected or disconnected in a consumer unit. In this sense, this paper introduces the use of frequency, phase, amplitude, and damping factor of current signals estimated by the Prony's Method. Using these components as features for different machine learning classifiers, such as k-Nearest Neighbors (k-NN), Support Vector Machine (SVM) and Ensemble, one can obtain an accuracy of 92% with the Ensemble classifier for different waveforms from a publicly available database. Additionally, we show that the proposed method presents comparable results with state-of-the-art methods for power signature classification, indicating that the Prony's Method as a feature extractor is a promising and still underexplored alternative in the related literature.