Two-Stage Energy Scheduling for Multi-Tenant Data Centers: Day-Ahead Workload Coordination and Real-Time Rolling Dispatch

The demand for data center computing and renewable power generation do not always align, which leads to inefficiencies in energy utilization. Flexible coordination of computing demand can address this issue. However, in multi-tenant data centers, users autonomously schedule some tasks, so operators cannot directly dispatch the corresponding loads. This paper proposes a two-stage energy dispatch method for multi-tenant data centers directly supplied by wind and photovoltaic generation. During the day-ahead stage, the operator sets a stochastic Stackelberg game and establishes the internal contract tariff. Contracted tenants act as followers and create load plans that are feasible for their services. During the real-time phase, a rolling model determines the admission of potential external cloud tasks and energy dispatch. The service price follows from the aggregate demand at the optimized admission quantity. Numerical results demonstrate that all contracted tasks are completed. Compared with the baseline without workload coordination, renewable energy utilization rises from 88.46% to 98.76%. The operator’s profit increases by 8.28%, and the cost per compute unit falls by 9.05%. Computing service volume increases, while grid dependence and operational emissions decline. Workload coordination improves profitability and operating flexibility under the tested grid restrictions and renewable-supply disturbances while maintaining contractual service.