Looking for Patterns in Residential Electricity Consumption

Abstract

Residential electricity consumption is an important part of general energy use. Its detailed investigation, however, requires rich empirical data, here the data of Swedish households. The individual consumption is a time series of readings at certain time intervals (hourly, every ten minutes, or every minute, say). Series exhibit patterns, in terms of which they may be compared, and it is desirable to model similarity. Classical statistical methods (correlation, factor, and cluster analyses) are presently used for this purpose; they have the advantage of being more explicit than the techniques of adaptive data analysis that may recently have become excessively popular. The present work is methodological, preceding any massive statistical analyses. Factor analysis allowed describing individual styles in terms of time intervals (during a day) of maximal variability. Cluster analysis was used for finding groups of days with similar patterns; the obtained clusters can help interpreting the results of other methods. Comparing two households requires comparing two sets of time series; correlation analysis quantified the similarity between them.