Work

Time Series Patterns Recognition With Dynamic Data Reduction in Nilm Appliance Identification

Year: 2024

Type: article

Abstract: The convergence of advancements of Internet of Things (IoT) capabilities, cou-pled with the accessibility of inexpensive and user-friendly sensors, has propelled the emergence of various new domains, and one such domain is Non-Intrusive Load Monitoring (NILM). A pivotal aspect of these technological advancements is the identification of appliances through the analysis of disaggregated power consumption signatures. The length of these signatures is contingent upon the frequency of data collection, where higher frequencies correspond to lengthier time series. To address this, we introduce a novel dynamic time series data reduction methodology, tailored to efficiently extract regions of interest from extended time series data. Subsequently, the efficacy of appliance classification using these extracted sub-ranges is assessed through the utilization of Matrix Profile techniques. The experimental validation is conducted utilizing the Plug-Load Appliance Identification Dataset, thus offering a concrete empirical basis for our approach's evaluation and verification. The proposed approach has successfully improved the overall accuracy of identifying the appliances in PLAID dataset with a significant margin as compared to the baseline approach.

Source: Journal of Information System and Technology Management

DOI: 10.35631/jistm.937011

Authors: Kwan-Hua Sim, Saad Tariq, Kwan-Yong Sim

Institutions: Swinburne University of Technology, Swinburne University of Technology Sarawak Campus

Language: English

Topic: Smart Grid Energy Management

Subfield: Electrical and Electronic Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: diamond

Funder: Ministry of Higher Education, Malaysia