Unsupervised Learning Methods for Power System Data Analysis

Abstract This chapter focuses on the use of the K-Means clustering algorithm for an enhanced visibility of the electrical distribution system which can be provided by advanced metering infrastructure and supported by big data technologies and parallel cloud computing environments such as Spark and H2O. Based on smart meter data of more than 30,000 loads in the City of Basel, Switzerland, and thanks to an appropriate cluster analysis, it is shown that useful knowledge of the grid state can be gained without any further information concerning the type of consumer and their habits. Once energy data is judiciously prepared, the features extraction is an important step. A graphical user interface is presented which illustrates the potentially great flexibility in the choice of features according to the needs of distribution system operators (DSOs). For example, the distribution of the various types of customers across the power system is of interest to DSOs. This chapter presents thus some pertinent examples of clustering outcomes that are visualized on the map of Basel, which notably enables to easily identify heating and cooling demand or gain insight into the energy consumption throughout the day for different neighborhoods.