Data-driven consumer-phase identification in low-voltage distribution networks considering prosumers

Low-voltage distribution networks (LVDNs) face growing challenges in managing consumer-phase identification (CPI), a critical task for load balancing, maintenance, and integrating behind-the-meter resources like rooftop PV systems. Traditional CPI methods—field inspections, additional sensors, or signal injection—are labor-intensive, costly, and error-prone. Existing data-driven approaches rely heavily on voltage correlation, which suffers from measurement noise and inaccuracies in prosumer-dominated grids, limiting their reliability.

A new study introduces a data-driven CPI framework addressing these limitations. Published in Frontiers in Energy, the research innovates by integrating two key advancements: (1) singular value decomposition (SVD) for denoising smart meter voltage data, improving signal quality; and (2) a KCL-based optimization model with a softargmax activation function, converting integer programming into a probabilistic problem to reduce computational complexity. Unlike prior voltage-correlation methods, this model accounts for prosumers’ impact by balancing feeder and consumer currents.

The study processes synthetic smart meter data (voltage/current measurements at 15-minute intervals) from an IEEE-906 test system with 55 consumers. First, SVD denoising removes Gaussian noise, enhancing voltage signal integrity. Then, a two-objective optimization model minimizes current mismatches between feeder and aggregated consumer currents, using softargmax to relax binary connectivity variables into probabilities. This approach ensures each consumer is uniquely assigned to one phase while tolerating measurement errors.

The framework improves CPI accuracy in noisy environments and prosumer-integrated grids, reducing reliance on expensive sensors or manual inspections. By leveraging existing AMI data, it offers a cost-effective tool for distribution system operators to optimize asset utilization, reduce phase imbalance, and enhance hosting capacity for renewable energy. The model’s robustness to real-world uncertainties marks a step toward smarter, more resilient low-voltage network management.

Original source: https://link.springer.com/article/10.1007/s11708-024-0946-4

https://journal.hep.com.cn/fie/EN/10.1007/s11708-024-0946-4

Sharable link: https://rdcu.be/eQ38x