Accuracy of clustering as a method to group distribution feeders by PV hosting capacity

Abstract

This paper examines the accuracy of clustering techniques for predicting hosting capacity. Hosting capacity results for 214 study feeders were used to predict a range of hosting capacities for an addition 7929 feeders using clustering techniques. Several methods were explored to try to improve the accuracy for predicting hosting capacity, including increasing the number of clusters, selecting variables that are highly correlated to hosting capacity for clustering, and weighting highly correlated clustering variables. The average normalized interquartile range (ANIQR) is used to compare the accuracy of several clustering methods for predicting hosting capacity.