| Title |
Locational Assessment of Feasible Connection Capacity Ranges for AI Inference Data Centers in Distribution Networks with Photovoltaics |
| Authors |
강민찬(Min-Chan Kang) ; 조윤진(Dong-Il Cho) ; 조동일(Yun-Jin Cho) ; 남준혁(Jun-Hyuk Nam) ; 문원식(Won-Sik Moon) |
| DOI |
https://doi.org/10.5370/KIEE.2026.75.10.2743 |
| Keywords |
AI inference data center; hosting capacity; distribution network; photovoltaic integration; site selection |
| Abstract |
This study evaluates location-dependent feasible connection capacity intervals for AI inference data centers in a distribution network with photovoltaics. A synthetic 24-hour facility load profile is constructed from public inference request traces, measured processor power characteristics, and power usage effectiveness data. OpenDSS simulations assess 53 candidate buses under three line current limits. In the identified intervals, existing overvoltage determines the lower bounds, while undervoltage or line current constraints determine the upper bounds. The maximum interval width increases from 1.141 MW at 350 A to 1.938 MW at 450 A. Sites selected by the upper-bound and interval-width criteria differ under all three conditions. The results support joint assessment of capacity bounds, interval width, and limiting constraints when planning data center location and size under a fixed load profile. |