Dynamic time warping based non-intrusive load transient identification. (1st June 2017)

• Record Type: Journal Article
• Authors: Liu, Bo; Luan, Wenpeng; Yu, Yixin
• Is Part Of: Applied energy. Volume 195(2017)
• Page Start: 634
• Page End: 645
• Publication Date: 2017-06-01
• Subjects: Non-intrusive load monitoring -- Transient power waveform -- Dynamic time warping -- Nearest neighbor classification -- Load transient identification; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2017.03.010
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: DTW algorithm is used to measure the similarity of variable-length TPW time-series. A transient identification method using DTW-based integrated distance is proposed. The integrated distance is designed combining multiple types of TPW signatures. Comparison and field tests verify the proposed method's accuracy. The proposed method is easy to implement in an embedded system at a reasonable cost. Abstract: Non-intrusive load monitoring (NILM) is a novel and cost-effective technology for monitoring load electricity energy consumption details. In the event-based NILM, transient power waveform (TPW) time-series can be used as signatures to identify the transients of the electrical appliances in the aggregated load, and then to determine their operating states, estimate their power demand and cumulative energy consumption. In this paper, for load transient identification, the dynamic time warping (DTW) algorithm is adopted for the first time to measure the similarity between the variable-length raw TPW sample and template time-series. Accordingly, a nearest neighbor transient identification method is proposed to identify the appliance creating the TPW sample time-series, in which the DTW-based integrated distance is used to measure the similarity of TPW signatures. Three schemes to calculate the integrated distance are designed, combining multiple types of TPW signatures. Comparison tests with existing methods are conducted using public datasets. The comparison test results indicate that the proposed load transient identification method cannot only improve the accuracy of load transient identification, but also is easy to implement at a reasonable cost. Ultimately, the proposed method is implemented in an embedded system. The field test results show that it can identify the operating states of electrical appliances with high accuracy.