Advanced Metering Infrastructure Data Driven Phase Identification in Smart Grid

Many important distribution network applications, such as load balancing, state-estimation, and network reconfiguration, depend on accurate phase connectivity information. The existing data-driven phase identification algorithms have a few drawbacks. First, the existing algorithms require the number of phase connections as an input. Second, they can not provide accurate results when there is a mix of phase-toneutral and phase-to-phase connected smart meters, or when the distribution circuit is less unbalanced. This paper develops an advanced metering infrastructure (AMI) data driven phase identification algorithm that addresses the drawbacks of the existing solutions in two ways. First, it leverages a nonlinear dimensionality reduction technique to extract key features from the voltage time series. Second, a constraint-driven hybrid clustering (CHC) algorithm is developed to dynamically create smart meter clusters with arbitrary shapes. The field validation results show that the proposed algorithm outperforms the existing ones. The improvement in the phase identification accuracy is more pronounced for distribution feeders that are less unbalanced. In addition, this paper discovers that more granular voltage time series leads to higher phase identification accuracy. Keywords—AMI; density-based clustering; phase identification; smart grid; t-SNE.