Customer-Phase Identification in Low-Voltage Distribution Networks Prone to Data Losses Using Machine Learning-based Algorithms

Abstract

Customer-Phase Identification is the determination of the exact phase to which each customer is connected in a three-phase distribution network. Knowing such information is critical for utility operators to maintain their customer records, which can later be used to enhance maintenance and repair services, outage and restoration management, load balancing, and the management of behind-the-meter resources. However, managing the customer-phase relationship has emerged as one of the most challenging issues in low-voltage distribution networks due to the continually escalating number of customer connections, the penetration of behind-the-meter resources, and the limited deployment of data measurement devices. To address these concerns, this paper examines three data-driven algorithms: k-medoids, k-means, and hierarchical clustering, and tests their robustness against data losses, measurement errors, and limited smart meter coverage. Results show that k-medoids and k-means clustering algorithms accurately identify the phase labels of customers as compared to hierarchical clustering.