A novel methodology based on clustering techniques for automatic processing of MV feeder daily load patterns

The paper illustrates a methodology for off-line processing historical loading data relevant to MV feeders originating from primary (HV/MV) substations of an electric grid equipped with a telemetering system (STU). The processed data are current values recorded for each feeder every 15 minutes and organized on a daily time basis as 96 component load patterns. The methodology has the main aim of identifying “off-standard” load patterns that can be present in historical data in consequence of contingencies having required MV network re-configurations. The identification task is solved with a pattern-recognition approach. To this aim, a free-clustering problem has been solved by means of an iteratively structured procedure implementing fuzzy-neural constructive and merging hierarchical algorithms. The procedure has been applied to several yearly data sets for various primary substations. The accuracy obtained for each investigated feeder has been never less than 90%. The procedure has been implemented in a user-friendly programme (DETECTOR) that automatically applies to data as directly available from STU