Machine Learning-Enabled Distribution Network Phase Identification

The distribution network has typically been the least observable and most dynamic and locally controlled element in the power grid. Complete information about the network topology is continuously changing and is not always readily available when needed. This makes the phase identification and network rebalancing a hard, costly, and time-consuming task for electric utilities, however, it is of great importance to future grid planning and advanced distribution management system (ADMS) type operation. Phase identification traditionally is executed manually, although there are existing voltage measurement based methods that are not always reliable. This paper develops a machine learning based data mining method for an accurate and efficient phase identification of residential customers in a distribution network by leveraging power consumption data collected through the advanced metering infrastructure (AMI). The proposed method uses a high-pass filter to remove the redundant and irrelevant parts of the power consumption time series, then identifies the residential customers' phase connectivity by proposing a modified clustering algorithm. Simulation results show the effectiveness of the proposed method in phase identification in both small and large networks and under complete and incomplete data scenarios.

Automated summary

By utilizing machine learning data mining methods and power consumption data collected through advanced metering infrastructure, this paper proposes an effective approach for accurate phase identification of residential customers in distribution networks. The method combines a high-pass filter to remove redundant information with a modified clustering algorithm to identify phase connectivity, showing effectiveness in both small and large network simulations with complete or incomplete data scenarios.