Improved divergence-based distributionally robust chance-constrained scheduling for geo-distributed Internet data centers

Geo-distributed Internet data centers (IDCs) have computational workloads that are temporally deferrable and spatially transferable, and they also have sufficient or even redundant supporting equipment. Thus, IDCs possess considerable economic potential. The economic potential of IDCs, however, is challenging to fully leverage due to the complex coupling of various pieces of equipment and the non-Gaussian uncertainty of stochastic workloads. To address these challenges, we propose a comprehensive model to unlock IDCs; power flexibility while incorporating an improved distributionally robust chance-constrained method to handle the uncertainty with desirable optimality. First, we explicitly model the power consumption of IDCs considering closely-coupled flexible sources. Then, we introduce chance constraints to develop a risk-averse optimization that can manage the uncertainty in critical constraints. Further, we propose an improved divergence-based distributionally robust chance-constrained model to reformulate each non-Gaussian chance constraint into a convex counterpart. Multiple divergence functions are traversed, and the least conservative one is chosen during the reformulation to ensure energy efficiency. Numerical experiments based on real-world data validate the efficacy of our proposed method, which significantly reduces costs while ensuring the quality of services and the computational efficiency.