GIST Releases Open Model of South Korea Power Grid
Yun-Su Kim of the Gwangju Institute of Science and Technology (GIST) has released the GIST 2217-Bus Test System, an open synthetic model of South Korea's national power transmission grid, according to UPI and an arXiv preprint (arXiv:2606.12791) submitted June 11, 2026. Per the paper, the dataset spans 2,217 buses, 512 generation and renewable sources totalling 144 GW, 3,708 AC line circuits, four HVDC converter links, and 3,324 transformers, serialized to a PSS/E-compatible CSV format and released under CC BY 4.0. The model was built strictly from public sources -- OpenStreetMap/OpenInfraMap overhead-line geometry and published national electricity statistics -- to avoid classified infrastructure data (arXiv:2606.12791). Per UPI, the release targets research into renewable integration, outage analysis, and AI-driven grid management. Korea's islanded grid with coastal generation clusters and heavy Seoul metropolitan demand differs markedly from widely used IEEE test cases, which the paper says has made reproducible domestic research difficult.
What happened
Yun-Su Kim of the Gwangju Institute of Science and Technology (GIST) released the GIST 2217-Bus Test System, a synthetic open model of South Korea's national power transmission network, per UPI and arXiv preprint arXiv:2606.12791 (submitted June 11, revised June 15, 2026). Per UPI and Asiae, the model is distributed free of charge under CC BY 4.0 and avoids classified infrastructure data by using only publicly available sources.
Technical specifications
Per the arXiv paper, the dataset spans 2,217 buses, 512 generation and renewable sources (totalling 144 GW), 3,708 AC line circuits, four HVDC converter links, 3,324 transformers, and reactive resources including shunts and 11 FACTS devices. Topology is derived from the OpenStreetMap/OpenInfraMap power layer via multi-source shortest-path reassembly of overhead-line geometry, gap-filling unreachable substations with a geographic minimum-spanning-tree backbone, and calibrating aggregate circuit length against published national statistics (94/100/109% at 765/345/154 kV per the paper). The dataset is serialized to a PSS/E-compatible CSV schema.
Validation
The arXiv paper reports the model is distributed as a frozen operating point: a single deterministic pandapower Newton-Raphson pass (with reactive limit enforcement and HVDC converter settling) reproduces an 85 GW high-demand snapshot at mean transmission voltage 0.995 pu with 2.6% losses. The paper notes structural consistency with the independent public KPG193 model.
Why it matters
Per the arXiv abstract, no model of the Korean transmission system at native resolution was previously publicly available. Korea's grid is an islanded interconnection with extreme separation between coastal generation clusters and the Seoul Metropolitan Area load center, low renewable penetration, and heavy reliance on extra-high-voltage (EHV) transmission -- characteristics differing markedly from widely used IEEE test cases, which limits reproducibility and geographic transferability of research using imported models.
Context
UPI and Asiae report the release supports research into renewable integration, outage analysis, and AI-driven grid management without exposing classified infrastructure. The arXiv paper frames the release as "a citable platform for power flow, planning, and decarbonization studies" (arXiv:2606.12791). Reporting does not include statements from GIST beyond the published paper and release materials.
Scoring Rationale
An open, PSS/E-compatible national-scale power grid test system from GIST directly supports reproducible ML and power-flow research on Korea's distinctive topology (islanded, EHV-heavy, coastal generation). The impact is solid for power systems and applied ML researchers but domain-specific and geographically concentrated, warranting a 'solid niche research' classification rather than a broader notable-tier score.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
