Tensor-Based Optimal Temperature Control of CRACs in Multi-Datacenters

Abstract

Optimal temperature control of computer room air conditioners (CRACs) is an effective way to lower energy costs in high-density data centers. To break the barrier of dimensionality when jointly optimizing the temperature of CRACs in multiple data centers (multi-datacenters) configured in-row cooler systems, a tensor-based sequential quadratic programming (SQP) method is presented in this study. This method takes the temperature tensors as input and transforms the original high-dimensional optimization problem into several lower-dimensional sub-problems. An effective approach to solving the non-convex optimization of these sub-problems is also proposed. The transformation method not only reduces the dimensionality of optimization but also increases its convergence speed. On the foundation of the tensor-based SQP, an optimized control strategy is constructed to adjust the outlet temperature of CRACs, with the objective of enhancing the cooling efficiency. The convergence behavior, maximum input temperature of racks, and cooling power are analyzed to illustrate the effectiveness of the proposed tensor-based optimization and control strategy.