Appliance Power Pattern Mining Via Motif Discovery in Unsupervised NILM

Abstract

Nonintrusive load monitoring (NILM) can obtain the power consumption information of a single appliance from the aggregated load data. Based on only limited knowledge about the appliance operating characteristics, the unsupervised NILM can autonomously extract load characteristic samples of appliances in unseen scenario, such as power waveform samples, and then form a personalized load signature base. Usually, the power consumption patterns and corresponding power waveform samples of the same appliance will appear repeatedly in the load power data according to users' appliance usage habits, therefore, the load characteristic sample can be extracted by searching for recurring subsequences in the load power time series, which is the core of unsupervised NILM. Motif discovery can discover similar subsequences with fixed or variable length in the time series of a large amount of data. In this research, we use motif discovery method to mine appliance power consumption patterns for the first time, and explore the possibility of the application of motif discovery in unsupervised NILM, to deal with the multiple time-scale characteristics of appliance power consumption patterns more effectively. Test results on private and public datasets show that the proposed time series motif discovery can effectively discover various load characteristic patterns of appliances in unseen scenarios, thus be feasible as part of unsupervised NILM.