Demand Flexibility Quantification for Data Centers: A Chance Constrained Bi-Level Model

Data centers have become demand-side resources in power systems through flexible load adjustment enabled by computing job scheduling. However, the quantification of this remains unexplored due to challenges including complex constraints and real-time large-scale optimization requirements in job scheduling operations. This paper proposes a chance-constrained bi-level model to quantify the demand flexibility of job scheduling in data centers, which integrates power capping and job scheduling processes. The upper-level model determines the power cap of the data center in each time interval according to the given Service Level Agreement (SLA) constraints, while the lower-level makes specific job scheduling decisions. The proposed model enables optimization of job scheduling strategies under varying power capping constraints, thereby evaluating the maximum permissible data center demand flexibility space while maintaining qualityof-service requirements. We propose a regression-based solution method to handle the high computing overhead of solving bilevel problems in real-time. Case studies using real-world job data validate the effectiveness of the proposed method.