Data center load modeling through optimal energy consumption characteristics: A path to simultaneously enhance energy efficiency and demand response quality

In an era defined by the rapid advancement of artificial intelligence and the global pursuit of “carbon neutrality,” data centers face the dual challenge of enhancing energy efficiency while ensuring high-quality participation in power system demand response. However, conventional linear load models used in demand response programming often force data centers into a trade-off: sacrificing energy efficiency to ensure response quality, or vice versa. This paper presents a hierarchical load modeling framework that captures the optimal energy consumption characteristics of data centers to mitigate this conflict. At the foundational layer, a fine-grained, cross-system energy consumption model is developed to capture the intricate electrical-thermal-performance interactions among the computing, cooling, and power conditioning systems within the data center. Solving the energy optimization problem at this layer yields the optimal energy consumption characteristics of the data center. At the upper layer, these characteristics are analyzed and abstracted into a weakly nonlinear demand response-oriented load model, composed of four patterns that together form a piecewise function—two linear and two nonlinear regions—each corresponding to distinct workload conditions. The nonlinear relations are simplified from cubic to quadratic forms without significant loss of accuracy. Experimental results show that the linear regions achieve $R^2\ge 0.9999$ with mean relative errors below 0.1404 %, while the quadratic regions reach $R^2\ge 0.9982$ with mean relative errors under 0.6259 %. Applied to a typical demand response program, the proposed model reduces electricity costs by 13.40 % to 30.21 %, energy consumption by 24.19 % to 38.31 %, and cumulative curtailment deficit by 98.09 %, compared to conventional linear models.