Abstract

This paper focuses on the detection of utilization patterns in electricity residential consumption, which are closely related to the occupant characteristics (e.g. number, age, occupancy, and social class). Our goal is to identify groups of appliances that are often used together via their statistically relatedness. This relation might be obvious (as in TV and Home Theater), or not. The results can be used, for example, to guide a recommendations letter from the energy supplier to the final user, suggesting specific change of habits in order to improve the residence’s energy efficiency. We propose here a methodology for identifying patterns from a large sets of system status, which is a computationally hard task defined in ℝ ⁿ with n being the number of appliances involved. The approach consist in the following steps: (i) the Principal Component Analysis method is employed to reduce the set dimensionality to ℝ ³ with explained variance from 68% to 90% to guarantee minimum information loses, (ii) the k-means method to clustering appliances into different groups and (iii) the elbow method was used to define the best number of clusters for each house with explained variance of at least 93% and reaching more than 99% for the best k. Numerical tests using the UK-DALE dataset are presented to show the effectiveness of the proposed solution. The main contribution of this work is a method with low computational cost that requires no other information than a large set of reliable system status (binary vectors) to reveal utilization patterns in the residence.