Energy disaggregation via data mining

Abstract

Electrical energy consumption of a residence is usually monitored by a meter installed at its entrance and it is composed by the sum of consumptions of installed devices. Energy disaggregation estimates the consumption of each device at each instant of time. This paper presents two main contributions. First, we address disaggregation using data mining techniques by clustering methods, that is, k-Means and Expectation and Maximization (EM). We demonstrate that we can obtain superior disaggregation accuracy from more complex methods. The second contribution, we elaborate clusters (dictionaries) considering that the states of operation of the devices and the signal of total consumption are dependent instances. We use Reference Energy Disaggregation Data Set (REDD), Waikato Environment for Knowledge Analysis (WEKA) and MATLAB.