Clustering methodology for classifying distribution feeders

Abstract

The screening process for DG interconnection procedures needs to be improved in order to increase the penetration of PV systems on the distribution grid. A significant improvement in the current screening process could be achieved by finding a method to classify the feeders in a utility service territory and determine the sensitivity of particular groups of distribution feeders to the impacts of high PV deployment levels. This paper presents a method for separating a utility's distribution feeders into unique clusters using the k-means clustering algorithm. An approach for determining the feeder variables of interest for use in a clustering algorithm is also described. The Cubic Clustering Criterion is used as a quality metric for determining the optimum number of clusters in a large dataset of over 3000 feeders from western utilities. An approach is illustrated for choosing the feeder variables to be utilized in the clustering process and a method is identified for determining the optimal number of representative clusters.