DC-Side Grid-Forming-Based Control Coordination in a Hybrid DC/AC Microgrid for AI Data Centers

Artificial intelligence (AI) data centers present significant challenges for microgrid power systems due to rapid and high-magnitude load variations generated by GPU clusters executing machine learning workloads. These load transients, which can exceed 30% of rated capacity within sub-second timescales, impose severe stress on conventional microgrid control strategies designed for slower industrial loads. This paper proposes a DC-side grid-forming-based hierarchical control coordination strategy for hybrid DC/AC microgrids supplying AI data centers. The proposed approach integrates virtual-inertia-based battery control with a three-phase adaptive power allocation strategy to coordinate energy sources with substantially different dynamic response characteristics, including battery energy storage (<100ms), solar photovoltaic generation (2-3s), and biofuel-based generation (5-15s). The coordination framework progressively transfers load responsibility from fast battery support to slower renewable and AC-side sources while maintaining the DC bus voltage within ±5% during rapid load transients. Simulation and hardware-in-The-loop validation results demonstrate improved DC voltage regulation compared with conventional droop-based control, particularly under the large and fast load variations characteristic of AI computing workloads. In addition, the proposed strategy reduces battery cycling stress through adaptive load redistribution, thereby improving the balance between transient response capability, battery utilization, grid interaction, and renewable energy integration for sustainable AI data center operation.