Energy consumption profiling using Gaussian processes

Abstract

Listen

Translate article

We present a novel clustering approach for time series based on Gaussian process regression in order to discover insights in the spending habits of households. The advantage of the proposed method is that it avoids the pairwise comparison of time series, employed by many existing methods. To this end, it learns a generalized model on several time series at once, based on their likelihood. We have validated our method using a real-world energy consumption dataset of 71 households and compared it with K-medoids and agglomerative clustering, using dynamic time warping. We not only show that our method is superior in terms of scalability but also that the produced results are useful in the decision making process of a company.