Evolutionary Decision-Making Mechanism for Geo-Distributed Data Center Aggregation and VPP Strategic Pricing

This paper proposes a co-evolutionary decision-making mechanism for coordinating the dynamic aggregation of geo-distributed data center (DC) clusters into virtual power plants (VPPs) under incomplete information. First, cross-regional VPP aggregation is modeled as an evolutionary game based on spatial workload shifting, where a geographical correction factor is introduced to account for the trade-off between economic incentives and network latency constraints. Second, VPPs are modeled as strategic aggregators with heterogeneous incentive strategies, allowing them to adjust incentive parameters to attract the spatiotemporal flexibility of geo-distributed DC clusters. To address the black-box nature of market responses, a Bayesian optimization-based pricing strategy is developed to search for effective incentive parameters by adapting to the evolutionary responses of DC clusters. Finally, a geographically corrected frequency-adjusted Q-learning algorithm is designed to compute the co-evolutionary stable state. The proposed model-free method supports distributed decision-making without requiring private data sharing and helps reduce the sensitivity of standard Q-learning to initial conditions. Case studies based on real-world production datasets from Google indicate that the proposed mechanism can achieve stable aggregation outcomes and improved utility relative to the compared benchmark methods.