Relieving Structural Transmission Bottlenecks Under Hyperscale AI Loads Through Co-Optimized Data-Center Siting and HVAC-to-HVDC Conversion

초록

Hyperscale artifi cial intelligence (AI) data centers are adding large, spatially concentrated loads while renewable generation reshapes transmission fl ow patterns, yet new corridors remain slow to build because of siting, permitting, and public-acceptance barriers. Conventional expansion planning treats data-center placement as an exogenous load and limits reinforcement to new lines. This study instead develops a centralized social-cost framework that co-optimizes hyperscale data-center siting and HVAC-to-HVDC conversion of screened existing corridors. The resulting mixed-integer quadratic program couples workload routing, latency-constrained server activation, modular data-center expansion, and communication-link investment with DC optimal power fl ow (DC-OPF) dispatch and reinforcement decisions. Case studies on the 193-bus Korean Power Grid test system, with branch ratings deliberately derated to emulate operational security margins, compare data-center siting alone, new AC construction, and conversion. The derated grid already contains a structural transfer bottleneck that siting alone cannot remove; both reinforcement options eliminate the modeled representative-hour load shedding, but conversion does so with substantially less added transfer capability and no new routes. The annualpeak defi cit is structural and persists even when every candidate asset is built, pointing to complementary resources such as energy storage or demand-side fl exibility rather than further line investment. Coordinating data-center siting with existing-corridor conversion thus off ers an effi cient reinforcement alternative when new-corridor development is diffi cult.

키워드