Profit-sharing Mechanism for Geo-distributed Data Centers Considering Delay Tolerance Diversity in Electric-Computational Coordination

The spatiotemporal flexibility of workloads in geo-distributed internet data centers (IDCs) offers significant potential for electric and computational coordination, yet privacy-preserving economic interaction among multiple opera tors remains underexplored, limiting market-oriented deploy ment at scale. This paper develops a bilevel optimization model that aligns electricity price signals with coordinated scheduling of four workload classes (interactive, bufferable, interruptible, and uninterruptible) to unlock flexibility and minimize operating cost. To ensure fair profit sharing under limited information sharing, this paper proposes a data-efficient Shapley value estimation scheme using a bidirectional gated recurrent unit (BiGRU)–Attention surrogate with hy perparameters tuned via a subtraction-average-based optimization (SABO) algorithm. Each IDC trains the surrogate lo cally to regress Shapley values from locational marginal prices, electricity consumption, and workload migration fea tures, thereby avoiding exhaustive coalition enumeration and reducing data requirements. A case study on a modified IEEE 30-bus system using a 365-day dataset demonstrates that the proposed coordination strategy reduces total operational costs by 14.06% and achieves Shapley-based sharing fairness. Moreover, results indicate that interactive workloads significantly enhance IDC flexibility and improve system-wide efficiency.