A new data-driven method based on Niching Genetic Algorithms for phase and substation identification

Abstract

Knowledge about the customers’ phase connections is strategic and critical for utility companies. It allows them to optimize maintenance and repair operations, implement load balancing, and detect losses, among other benefits. However, this information may be incomplete or outdated due to the undocumented changes in the Low Voltage network. Several methods have been proposed to estimate it. Methods based on data analysis stand out because they do not require costly specialized equipment. This work presents a new method for Phase Identification and Transformer Substation Detection for single-phase customers. Unlike previous approaches, we address the problem through a heuristic optimization, using an Evolutionary Algorithm based on Deterministic Crowding and correlation analysis on load measurements. The algorithm was designed to work with low penetration of smart meters and missing data, obtaining better results in shorter periods. The method was tested using both a public dataset and a dataset from Tucumán province, Argentina. We obtained an average accuracy above 95% on 21 days if almost 30% of the smart meters are available (200 customers in total). In contrast, only 5 days are required to reach the same accuracy if more than 80% of smart meters are available.