Cluster analysis for primary feeder identification using metering data

This paper presents a methodology to identify the connected feeder of high-usage customers in a primary distribution network. The proposed methodology considers voltage characteristics of radial distribution and actual measurements. Based on 15-minute intervals metering, cluster analysis is applied to categorize customer patterns on the basis of voltage correlation. Afterwards, support vector classification is also introduced for outlier assigning and cluster separation. The feasibility of this method is demonstrated on a practical distribution network of industrial estate area. The result indicates that all of the customers is correctly identified, and its correctness percentage is also better than the existent network representation. Additionally, wavelet reduction offers the same performance as the original time-domain feature but more efficient use of time.