Non-Intrusive Load Monitoring Using Current Shapelets

Abstract

Using a single-point sensor, non-intrusive load monitoring (NILM) discerns the individual electrical appliances of a residential or commercial building by disaggregating the accumulated energy consumption data without accessing to the individual components. To classify devices, potential features need to be extracted from the electrical signatures. In this article, a novel features extraction method based on current shapelets is proposed. Time-series current shapelets are determined from the normalized current data recorded from different devices. In general, shapelets can be defined as the subsequences constituting the most distinguished shapes of a time-series sequence from a particular class and can be used to discern the class among many subsequences from different classes. In this work, current envelopes are determined from the original current data by locating and connecting the peak points for each sample. Then, a unique approach is proposed to extract shapelets from the starting phase (device is turned on) of the time-series current envelopes. Subsequences windowed from the starting moment to a few seconds of stable device operation are taken into account. Based on these shapelets, a multi-class classification model consisting of five different supervised algorithms is developed. The performance evaluations corroborate the efficacy of the proposed framework.

Non-Intrusive Load Disaggregation by Convolutional Neural Network and Multilabel Classification

“…In [8], a characterization is proposed on the basis of the sampling rate of the features themselves, which are divided into very slow (less than 1 min), slow (between 1 min and 1 s), medium (higher than 1 Hz but slower than the fundamental frequency), high (up to 2 kHz) and very high (between 2 and 40 KHz). The features of the slow and very slow category can be used directly, as in [9,10], through statistical characterization of time-series sub-sequences as well as with signal processing [11][12][13]. Higher sampling rates allow for more detailed characterization of transients in the consumption of household appliances [14].…”

Comprehensive NILM Framework: Device Type Classification and Device Activity Status Monitoring Using Capsule Network

“…On the contrary, the more effective way for ALM is non-intrusive appliance load monitoring (NIALM or NILM), where cost-reasonable maintenance, accurate operation and prolonged sustainability are ensured with less measuring resources requirement. NILM emerges as an important constituent of smart energy metering techniques that determines individual energy consumption profile of different electrical appliances using a single measurement point [2].…”

“…Three primary approaches for analyzing and monitoring electrical energy data are reported in [1] - [5], which can be categorized as-steady-state analysis, transient-state analysis and non-traditional appliance features. The steady-state analysis detects the changes in load identification considering stable states of devices; the transient-state analysis focuses on the transitional states in energy consumption profile; while the last technique concentrates on determining atypical features of the electrical instruments to monitor them [2].…”

“…In addition to the aforementioned NILM methodologies, robust and fast classification techniques, referred to as shapelets, have been introduced for proposing efficacious NILM solution in [2]. Shapelets are generally defined as interpretable time-series subsequences of a dataset, which can categorize an unlabeled sequence if it comprises an occurrence of the shapelets within a specified learning distance [2].…”

“…In addition to the aforementioned NILM methodologies, robust and fast classification techniques, referred to as shapelets, have been introduced for proposing efficacious NILM solution in [2]. Shapelets are generally defined as interpretable time-series subsequences of a dataset, which can categorize an unlabeled sequence if it comprises an occurrence of the shapelets within a specified learning distance [2]. In this work, analysis starts with determining time windowed current envelopes from normalized data and then, time-series shapelets are derived from these envelopes to extract feature components based on which a multi-class classification model is formed to validate the device disaggregation performance.…”