grid topology inference

This study addresses the critical limitation imposed by the high confidentiality of urban power grid topology data, which hinders distribution network research and innovation. The authors propose a novel methodology that leverages entirely open data—comprising publicly available electrical infrastructure records and OpenStreetMap—to reconstruct full-scale grid topologies from high-voltage transmission networks down to individual buildings. By integrating graph-theoretic algorithms to construct the medium- and high-voltage backbone and applying geospatial machine learning to cluster building-level electricity demand and infer low-voltage connections, the framework enables comprehensive network reconstruction without proprietary data. Validated in the Alna district of Oslo, Norway, the approach successfully replicates a complete grid encompassing 7,330 buildings and all key electrical assets, thereby facilitating power flow optimization, cascading failure simulations, and resilience analysis under high penetration of distributed renewable energy resources.

Accurately reconstructing distribution network topology from multi-source, heterogeneous, and low-quality utility data—such as GIS metadata and voltage time-series signals—remains challenging due to data uncertainty and physical inconsistency. Method: This paper proposes a confidence-aware, end-to-end learnable inference framework that jointly models spatial layout and dynamic electrical behavior. It incorporates transformer capacity constraints and radial topology priors, soft-handles data uncertainty while hard-enforcing physical feasibility, and preserves structurally critical yet low-quality information. Crucially, it quantifies connection reliability for each branch. Results: Evaluated on three real-world Oncor feeders (>8,000 meters), the method achieves >95% topology reconstruction accuracy, significantly improves confidence calibration, and enhances computational efficiency. It establishes a robust topological foundation for digital twins in high-uncertainty distribution grid scenarios.

This study addresses the lack of standardized cross-national comparative analysis of topological diversity and robustness in high-voltage transmission grids (≥110 kV) across 15 European countries. Method: A unified complex network model was constructed for each national grid, incorporating substations where applicable. Vulnerability was systematically assessed via Monte Carlo simulations of both random and targeted node/edge removal. Topological metrics—including the power-law exponent of degree distribution—were quantified and correlated with functional resilience. Contribution/Results: The study establishes the first standardized topological benchmark for multinational HV grids, demonstrating that the degree distribution’s decay rate effectively discriminates between highly resilient and highly vulnerable systems. It further reveals substantial sensitivity of vulnerability assessments to modeling granularity—particularly the inclusion or exclusion of substations. Crucially, it derives quantitative mappings between structural features and empirical robustness, providing a theoretical foundation and methodological framework for grid resilience evaluation and transnational infrastructure planning.

In digital twin applications for distribution networks, conventional structural similarity metrics—such as subgraph isomorphism and graph edit distance—fail due to the absence of one-to-one node/edge correspondence between synthetic and physical networks. To address this, this paper introduces the multiscale flat norm—a geometric measure-theoretic metric—into power network structural validation for the first time. Our method integrates planar triangulation with linear programming optimization to automatically localize structural discrepancy regions (“patches”) and establishes theoretical stability bounds, thereby overcoming topological mismatch limitations in distance quantification. Experiments on a real U.S. county-scale distribution network demonstrate that, compared to the Hausdorff distance, our approach jointly captures topological and geometric discrepancies with higher fidelity, significantly enhancing the interpretability and reliability of digital twin quality assessment.

This study addresses the challenge of fault location in sparsely measured, partially observable distribution networks. The authors propose a “measurement-only” graph construction strategy combined with a spatiotemporal graph neural network (STGNN) to systematically compare modeling performance between full-topology graphs and subgraphs comprising only measurement nodes. The proposed architecture integrates GraphSAGE with an enhanced GATv2 mechanism and is evaluated on the IEEE 123-bus feeder. Experimental results demonstrate that the method achieves up to an 11-percentage-point improvement in F1 score over RNN-based baselines, reduces training time by a factor of six, and exhibits greater stability, thereby significantly enhancing both the efficiency and robustness of fault location in distribution systems.

This study addresses the scarcity of realistic distribution network datasets—a key bottleneck in benchmarking planning and operational tools under high penetration of distributed energy resources. To overcome this, the authors propose a generative adversarial network (GAN)-based framework for synthesizing distribution grid layouts, uniquely integrating rasterized image representations with GANs to enable both unconditional generation and geographically conditioned synthesis incorporating street maps and customer distribution. Through GIS preprocessing, image-based topological encoding, and a multi-resolution training strategy, the method successfully reproduces realistic topologies aligned with geographic structures across low-, medium-, and high-voltage scenarios. The approach offers data-driven layout recommendations for electrifying new areas while also highlighting persistent challenges in training stability and modeling electrical constraints.