Building Energy Disaggregation using Spatiotemporal Pattern Network

Abstract

Numerous studies on non-intrusive load monitoring (NILM) of electrical demand have been performed for the purpose of identifying load components only using univariate data, such as the identification of a certain type of end-use (e.g., lighting load) via whole building electricity consumption time series. However, additional time series data may become useful in providing distinguishable features for energy disaggregation which can be rendered as a multivariate time series data analysis problem. This paper presents a novel probabilistic graphical modeling approach called the spatiotemporal pattern network (STPN) for addressing such problem of pattern extraction from multivariate time-series data with application to building energy disaggregation. The proposed scheme shows promise in dealing with multivariate time-series with widely different characteristics for the improvement in energy disaggregation performance. We use multiple real data sets to validate the STPN framework along with performance comparison with the state-of-the-art techniques such as factorial hidden Markov models (FHMM) and combinatorial optimization (CO).