Privacy-preserving energy scheduling for data center clusters based on evolutionary game theory

Abstract

The substantial energy consumption of data centers challenges power grid stability and sustainable power development. While coordinating data centers for demand response is a viable solution, existing centralized approaches face scalability bottlenecks and privacy risks, and traditional game-theoretic models often rely on ideal assumptions of perfect rationality. To address these gaps, this paper proposes a privacy-preserving decentralized scheduling framework based on evolutionary game theory. Unlike centralized methods that require full state disclosure, the proposed framework models the customer directrix load (CDL) tracking problem as a population evolution process, enabling data centers to optimize strategies using only local information and aggregated signals. To address the limitations of standard game-theoretic approaches which often neglect physical constraints, this work incorporates a computation-power coupling model and employs a bounded rationality perspective. Furthermore, to solve the optimization in high-dimensional discrete spaces where traditional replicator dynamics fail, a hybrid algorithm (EGT-QCDE) integrating Q-learning with compound differential evolution is developed. Case studies based on real-world data validate that the proposed method guides the data center cluster to an evolutionarily stable strategy. The solution achieves reductions in overall operational costs while maintaining accurate tracking of the power grid’s dispatch targets.