A Data-Driven Approach for Topology Correction in Low Voltage Distribution Networks with PVs

TL;DR

This work addresses the challenge of outdated and incomplete LVDN topologies by introducing a data-driven topology correction framework that relies solely on voltage magnitude data. The approach combines (i) switch-state identification via Random Forest with labeled switch-bar encoding, (ii) correlation-based user-feeder identification using a modified Fisher z-transformed PCC within a hierarchical clustering scheme, and (iii) phase-label identification using normalized voltage features, augmented by a time-based data selection to mitigate PV-induced disturbances. The framework demonstrates robustness to incomplete smart meter data and varying network complexity, achieving high accuracy in switch-state and phase labeling while maintaining feasible performance for partial recordings. Practically, this enhances network observability and supports load balancing and PV integration with privacy-preserving data usage.

Abstract

To correct the outdated and incomplete topology of low voltage distribution networks (LVDNs) solely based on voltage magnitudes, a data driven approach is developed based on machine learning algorithms and correlation analysis. Meanwhile, to address the similarity among smart meter (SM) data induced by distributed photovoltaic (PV) systems, a time based SM data selection strategy is combined with the proposed correlation analysis. Unlike offline approaches, the proposed approach uses up to date voltage magnitudes to help distribution system operators determine switch states via supervised learning and refine user feeder connections and customer phase labels through a modified hierarchical clustering algorithm. The feasibility and robustness of the proposed approach are validated using modified real world LVDNs and multiple incomplete SM datasets collected from customers in the Netherlands. The results demonstrate that the time-based SM data selection strategy effectively mitigates its impact on phase identification, and the corrected topology not only improves network observability but also supports network operators in load balancing and PV consumption.

Figures (10)

• Figure 1: Deployment patterns of LVNDs: (a) feeder deployment under one street, (b) distribution of MV/LV transformers and (c) underground feeders connected to the same transformer.
• Figure 2: Proposed topology correction approach composed of three steps: (1) switch state identification, (2) user-feeder connection identification and (3) phase labels identification.
• Figure 3: Illustrative example of switch deployment and states in LVDNs: 1 represents the activated switch and vice versa abb2025.
• Figure 4: A practical LVDN with multiple parallel feeders along the same street demonstrates the common challenge of ambiguous use-feeder connections and phase labelling, illustrating the physical complexity that makes accurate network topology identification (i.e., step (2) and (3)) difficult.
• Figure 5: Illustrative example showing the sequential process and intermediate outputs of the proposed topology correction framework, demonstrating (1) switch states, (2) user-feeder connections and (3) phase labels.