Advanced SOM & K Mean Method for Load Curve Clustering

Abstract

From the load curve classification for one customer, the main features such as the seasonal factors, the weekday factors influencing on the electricity consumption may be extracted. By this way some utilities can make decision on the tariff by seasons or by day in week. The popular clustering techniques are the SOM & K-mean or Fuzzy K-mean. SOM &Kmean is a prominent approach for clustering with a two-level approach: first, the data set will be clustered using the SOM and in the second level, the SOM will be clustered by K-mean. In the first level, two training algorithms were examined: sequential and batch training. For the second level, the K-mean has the results that are strongly depended on the initial values of the centers. To overcome this, this paper used the subtractive clustering approach proposed by Chiu in 1994 to determine the centers. Because the effective radius in Chiu’s method has some influence on the number of centers, the paper applied the PSO technique to find the optimum radius. To valid the proposed approach, the test on well-known data samples is carried out. The applications for daily load curves of one Southern utility are presented.