A case study on the modeling of system state degradation for data center cooling systems

Abstract

The energy efficiency of a data center largely depends on the performance of its cooling system, which consumes 30% to 40% of the total electricity. Very often, the cooling efficiency degrades over time and the degradation depends on various external factors and internal system states. This study is motivated by the need to estimate the degradation of system internal states from measured system operating data. Based on the thermodynamic law that governs the relationship between system internal states, operating conditions and cooling efficiency, a two-stage physics-based statistical approach for modeling the cooling efficiency is proposed. The model also takes into account the statistical dependence among system state variables, and captures the complex dependence structure by the Archimedean family of copulas with its generator function approximated by a cubic B-splines. The case study demonstrates how statistical models can be constructed for complex physical systems.