A Data-driven, Multi-setpoint Model Predictive Thermal Control System for Data Centers

Energy, exergy and computing efficiency based data center workload and cooling management

“…Investigators have previously considered the energy-based objective function during joint workload management and cooling control [1,27]. Our first goal is to maximize 𝐶𝑂𝑃 𝑐 by minimizing 𝑃 ̇𝑐𝑜𝑜𝑙 for a prescribed IT load 𝑃 ̇𝐼𝑇 .…”

“…As observed from Table 1, thermally-aware workload distribution algorithms require an accurate dynamic thermal model for the DC. Several options are available, such as (a) data-driven black-box models [27], (b) data-driven hybrid gray-box models [28,29], (c) computational fluid dynamics (CFD) simulation-based heat recirculation matrix (HRM) approaches [12], and (d) physics-based zonal models [22,30]. The first three require significant training data and computational resources, and are the basis of prior workload scheduling algorithms.…”