Profiling household appliance electricity usage with N-gram language modeling

Household appliance classification, based on electricity usage patterns, is gaining a momentum in an era where energy saving has become a priority and connected objects are leveraged to influence consumers' behaviors. In this respect, electricity usage profiling of household appliances is an important step for identifying malfunctioning devices and generating automatic alerts about unusual consumptions. To support the implementation of such profiling tasks in a practical and scalable way, a language modeling approach for classifying time series representing appliances' energy consumption readings is developed in this paper. This approach captures both local discriminative features of consumption profiles and overall curve shapes to yield high classification accuracy, even in the presence of unavoidable noise. Our approach is evaluated against state-of-the-art approaches, and validated based on a large public dataset from 27 different households with energy readings during a one year span. Our experiment findings also show that how the approach remains efficient even in the presence of noise.