Applicability of using time series subsequences to study office plug load appliances

•This study is the first of its kind to analyse the applicability of time series subsequence mining for characterizing and classifying individual appliances and their events within aggregate time series data.•A novel idea of “Bag of Rules” (BoR) for appliances’ transient events is proposed. An example of a Multi-Function printing Device’s (MFD) potential subsequence rules across multiple operational states are mined to towards developing BoR.•A Naïve classification approach employing BoR for an MFD is introduced as a proof-of-concept. This classification results suggests the applicability of time series subsequence mining in the context of NILM in offices.

Energy management in offices requires efficient methods (e.g. non-intrusive load monitoring techniques, NILM) to monitor the large number of workstations and office appliances. The key purpose of this study is to ascertain the applicability of using time series subsequence data mining to study and classify the transient operations of typical appliances in an office. The approach involves discovering hidden subsequences (i.e. feature extraction) that are characteristic of individual appliance transient states, using an extension of Symbolic Aggregate approXimation (SAX). Such characteristic features are used to create a repository of rules to help supervised classification of aggregate time series measurements. It is one of the first studies to demonstrate the potential of classifying subsequence features into individual appliances and their states within large aggregate time series data using a “Bag of Rules” approach. The results indicate that distinct, characteristic patterns represent office appliances and their states, in the form of SAX grammar rules. These patterns can then be used for NILM with promising results. This ongoing study demonstrates SAX based time series subsequence mining as a proof-of-concept; not only to discover similarities presented by appliance events but also to demonstrate their applicability to disambiguate aggregate signatures in the context of office NILM.