Phase preserving profile generation from measurement data by clustering and performance analysis: a tool for network planning and operation

The need for improved operational efficiency planning accuracy leads to a growing number of sensors and other monitoring sources in our power system. New methods for properly dealing with this increasing amount of data are required. This paper presents how clustering can help to drastically reduce the processing time of energy data time series. The developed approach categorizes similar load behavior by means of agglomerative hierarchical clustering based on their correlation coefficient. It includes the determination of the best number of clusters to model different load patterns with respect to the total error given as a key performance indicator. The results are a reduced set of representative three phase load profiles based on the data input and clustering configurations. The accuracy of these representative profiles is validated by resembling the original data set. Dependent on available computational resources a network operator can use this to intelligently compress measurement data while keeping the required accuracy. The method is demonstrated on data from the testbed of Aspern Smart City Research in Seestadt Aspern, Austria.