Abstract
For more than a decade researchers have successfully analyzed smart-meter data to identify consumption patterns. Numerous projects have applied K-Means and other clustering algorithms from machine learning to identify various consumption patterns hidden in the smart-meter data. What motivates both researchers and private stakeholders is the possibility of producing consumption-clustering solutions applicable outside academia to facilitate value propositions for both utilities and consumers. However, for clusters to be truly applicable beyond academia, they need to be defined in such a way that they are meaningful and stable. Therefore, it is important to study the stability of the clusters across time periods to ensure that cluster solutions remain the same and that the transition between clusters is understood and quantified.