Pattern classification of harmonic monitoring data using data mining

The rapid increase in computer technology and the availability of very large scale in power quality monitoring data have motivated the utility engineers to extract the useful information that is hidden into the data. This information might be critical issues for diagnoses the power quality problems, predicting expecting failure and alarming for dangerous situation. This paper identifies the underlying classes in harmonic data of MV/LV distribution systems using mixture modelling method. Multidimensional Scaling (MDS) and link analysis are then used to merge similar clusters in the discovered classes and super-group abstractions are formed from the clusters when the selected dissimilarity threshold is exceeded. The rules behind the interested super-groups are extracted using the C5.0 algorithm of classification techniques in supervised learning of data mining.