Joint Scheduling of Workload Demand and Energy Supply in Low-carbon Data Centers with Decision-Dependent Uncertainty Set

TL;DR

This paper introduces a novel joint scheduling approach for workloads and energy supply in low-carbon data centers, explicitly modeling decision-dependent uncertainty to optimize costs and improve resilience.

Contribution

It proposes a workload classification model with decision-dependent uncertainty sets and a mixed integer program for optimal scheduling in low-carbon IDC environments.

Findings

Significant reduction in energy operational costs.

Effective workload scheduling under uncertainty.

Enhanced system resilience with the proposed algorithm.

Abstract

This paper addresses the joint scheduling problem of stochastic workloads and a hydrogen-enabled distributed energy system in a low-carbon Internet data centers (IDC). Although such workloads can be shifted over temporal and spatial horizons, it poses challenges when they cannot be accurately predicted, resulting in significant efficiency degradation and high operational cost of the energy system. The problem becomes even more difficult when the workload shifting decisions would influence their randomness, which is natural for the IDC workloads. To tackle these issues, we propose a workload classification model based on the decision-dependent uncertainty set, where the spatiotemporal elasticity of different types of random workloads are clearly identified and the decision dependencies are explicitly described as linear constraints. Thus, a mixed integer program is then established for the optimal scheduling of both workload demand and energy supply. To enhance system resilience against high volatility scenarios, a rolling horizon algorithm is developed to ensure nonanticipativity and full-scenario feasibility. Numerical tests demonstrate that the proposed method exhibits effective workload scheduling decisions with dramatic energy operational cost reductions compared to the benchmarks under most of the uncertain scenarios.