A Transient and Steady-State Power Signature Feature Extraction Using Different Prony's Methods

Abstract

The extraction of features that defines the electric power signature of a load or an appliance is one of the most relevant stages 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 work compares five different approaches of the Prony's method (Polynomial, Least Squares, Total Least Squares, Matrix Pencil, and IIR-based) with the aim of simplifying the order of the model and provide an analytic solution to the problem of estimating the harmonic content (frequency, phase, and amplitude) with the exponential damping factor for current signals. Using the components from Prony's method as steady-state and transient features, an Ensemble classifier is used to perform the classification of the different waveforms of two publicly available databases (COOLL and LIT). With that, we show that the proposed method achieves 98% for the COOLL dataset and 97% for the LIT database, using the Matrix Pencil approach.