GeoDistNet: An Open-Source Tool for Synthetic Distribution Network Generation
Yunqi Wang , Xinghuo Yu , and Mahdi Jalili · 2026
Read GeoDistNet: An Open-Source Tool for Synthetic Distribution Network Generation by Yunqi Wang , Xinghuo Yu , and Mahdi Jalili (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-29523.txt.
Introduction
Wang and co-authors present GeoDistNet, an open-source tool that builds synthetic power distribution networks. Feeder studies need models that are both usable and true to real areas. Real feeder data are rarely public, and benchmarks miss local geography. Their tool starts from public map data for a service area. It builds a candidate graph and picks a radial feeder layout. It then assigns standard electrical parameters and loads. A Melbourne case study produced a feeder ready for power-flow runs.
Catalog record
| CREATOR | Yunqi Wang , Xinghuo Yu , and Mahdi Jalili |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31555 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.29523
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
GeoDistNet: An Open-Source Tool for Synthetic Distribution Network Generation Thanks: This work is partially supported by Australian Research Council’s Discovery Program Grant DP240100830. Yunqi Wang , Xinghuo Yu , and Mahdi Jalili Abstract Distribution-level studies increasingly require feeder models that are both electrically usable and structurally representative of practical service areas. However, detailed utility feeder data are rarely accessible, while benchmark systems often fail to capture the geographic organization of real urban and suburban networks. This paper presents GeoDistNet, an open-source tool for synthetic distribution network generation from publicly available geographic information. Starting from map-derived spatial data, the proposed workflow constructs a candidate graph, synthesizes feeder-compatible radial topology through a mixed-integer formulation, assigns representative electrical parameters and loads, and exports the resulting network for power-flow analysis. A Melbourne case study shows that the generated feeder remains geographically interpretable, topologically structured, and directly usable in pandapower under multiple loading levels. GeoDistNet therefore provides a reproducible workflow for bridging publicly accessible GIS data and simulation-ready distribution feeder models when detailed utility networks are unavailable. I Introduction Distribution systems are becoming a central focus in modern power system research due to the rapid growth of distributed energy resources (DERs), electric vehicles, flexible demand, and active network operation. As a result, a broad range of studies including hosting-capacity assessment, voltage regulation, service restoration, network reinforcement, and flexibility coordination, now depend on feeder models that are not only electrically analyzable, but also structurally representative of practical networks. In many of these applications, the realism of feeder topology, route layout, and load distribution directly affects the credibility of the resulting conclusions. Despite this need, access to real utility feeder data remains highly restricted. Detailed distribution network models are typically unavailable outside utilities because of confidentiality, security, and commercial constraints. Consequently, many studies continue to rely on a limited number of benchmark feeders, such as the classical 33-bus test system proposed by reference [ 1 ] . These benchmark systems remain indispensable for algorithm development and comparative assessment. More recently, efforts have also been made to improve their realism for active distribution system studies, for example through the enhanced IEEE 33-bus benchmark in [ 4 ] . Nevertheless, benchmark feeders remain small in number and limited in regional diversity. They are useful as reference systems, but they do not generally preserve the geographic structure, corridor layout, or spatial heterogeneity of real urban and suburban service areas. This limitation has been explicitly recognized in the literature. Postigo Marcos et al. reviewed distribution test feeders used in the United States and emphasized the need for synthetic representative networks that extend beyond a few standard benchmark cases [ 8 ] . More broadly, the authors in [ 2 ] established structural metrics as validation criteria for synthetic grids, showing that realistic network studies require synthetic models to be evaluated against topological characteristics rather than treated as arbitrary graph instances . These observations have motivated growing interest in synthetic grid construction as a practical alternative when utility-grade data are unavailable. At the distribution level, synthetic network generation has advanced considerably in recent years. Mateo et al. demonstrated that large-scale synthetic U.S. electric distribution system models can be constructed with realistic structural and engineering characteristics [ 9 ] . This line of work has shown that synthetic distribution models can serve as a practical bridge between idealized test systems and inaccessible utility feeders. At the same time, it also makes clear that feeder synthesis is not simply a graph-generation problem. A useful distribution network model must satisfy multiple requirements simultaneously: it should reflect plausible spatial organization, remain electrically interpretable, and support the simulation tasks for which it is intended. Existing approaches to synthetic feeder generation can be broadly grouped into three categories. The first adapts existing benchmark feeders by modifying topology, scaling load levels, or adjusting parameters to suit a given study. The second follows statistical or engineering-template approaches, in which feeders are generated to match target topological or asset-level distributions. The third increasingly explores data-driven or graph-learning approaches. For example, Liang et al. proposed FeederGAN, which generates synthetic feeders through deep graph adversarial learning and reproduces feeder topology and component patterns from existing feeder datasets [ 6 ] . These studies confirm that synthetic feeder generation is both necessary and technically viable. However, many such approaches still depend on proprietary feeder inventories, pre-existing utility models, or statistical training data that are not themselves openly available. A related line of work considers geographically informed feeder construction. This direction is especially relevant because practical distribution feeders are strongly shaped by road corridors, land development, and spatial demand layout. Saha et al. proposed a framework for generating synthetic distribution feeders from OpenStreetMap, demonstrating the potential of map-derived topology as a basis for feeder construction [ 10 ] . Trpovski et al. similarly explored synthetic distribution grid generation in a planning-oriented setting, highlighting the importance of geographic and engineering consistency in feeder synthesis [ 12 ] . These studies indicate that publicly accessible geographic information can provide a useful basis for synthetic feeder construction, particularly when confidential utility models are unavailable. Even so, a practical gap remains between publicly available map data and simulation-ready electrical feeder models. A road map is not a distribution network. It contains geometric and spatial information, but it does not directly specify feeder radiality, source location, electrical parameters, or demand allocation. Map-derived graphs are often too dense, too detailed, or too geometrically unconstrained to be used directly in feeder simulation. As a result, a formal construction layer is required to translate raw geographic information into a connected, radial, and electrically usable network representation. This translation is nontrivial: it must preserve dominant spatial structure while removing unnecessary geometric detail, and it must do so in a way that is compatible with downstream electrical analysis. Another practical requirement is reproducibility. If synthetic feeder generation is to be broadly useful, the construction workflow should be transparent and implementable using accessible tools. From the electrical-analysis perspective, pandapower has emerged as a widely used open-source framework for modeling, analysis, and optimization of electric power systems [ 11 ] . OpenDSS has likewise remained a common platform for distribution simulation studies [ 5 ] . On the geographic-data side, OSMnx provides an effective route for acquiring and processing street-network structure from OpenStreetMap [ 3 ] . Together, these developments make it increasingly feasible to build synthetic feeder workflows that are both open and practically useful. This paper addresses the above gap through GeoDistNet , an open-source tool for synthetic distribution network generation from publicly accessible geographic information. Rather than learning feeder structure from confidential utility models or relying solely on statistical templates, GeoDistNet starts from map-derived spatial data, extracts a candidate graph, synthesizes feeder-compatible radial topology through a mixed-integer formulation, assigns representative electrical parameters and loads, and exports the resulting model for power-flow analysis. In this sense, the contribution of the present work does not lie in proposing a new power-flow engine. Instead, it lies in bridging real-world geographic structure and simulation-ready feeder construction within an open and reproducible workflow. The main contributions of this paper are summarized as follows: • GeoDistNet is introduced as an open-source workflow for synthetic distribution feeder generation from publicly available