Grid Frequency Stability Support Potential of Data Center: A Quantitative Assessment of Flexibility

TL;DR

Data centers increasingly influence grid dynamics due to large, rapid loads; the paper tackles secure unit commitment with data-center frequency support by embedding nonlinear nadir constraints into MILP using a decision-tree–based constraint-learning (DT-CL) module. The Safe UC framework couples traditional UC with data-driven safety constraints, translating learned nadir boundaries into MILP-ready inequalities to allow tractable optimization of $R^{\mathrm{DC}}_t$ and other decisions. A key contribution is the Marginal Flexibility Value (MFV), defined as $\text{MFV} = \frac{C_i - C_j}{\phi_j - \phi_i}$, which quantifies the system-wide cost savings per 1% increase in DC flexibility and demonstrates diminishing returns as DC participation grows. Case studies on a modified IEEE 118-bus system and a 2030 high-renewable scenario show that increasing fast DC flexibility reduces total costs and improves wind integration, with faster response times yielding greater savings and higher MFV in future, higher-DC contexts; DT-CL also outperforms traditional PLA and KRL linearization methods in accuracy and feasibility, enabling practical deployment of data-center–enabled frequency support.

Abstract

The rapid expansion of data center infrastructure is reshaping power system dynamics by significantly increasing electricity demand while also offering potential for fast and controllable flexibility. To ensure reliable operation under such conditions, the frequency secured unit commitment problem must be solved with enhanced modeling of demand side frequency response. In this work, we propose a data-driven linearization framework based on decision tree based constraint learning to embed nonlinear nadir frequency constraints into mixed-integer linear programming. This approach enables tractable optimization of generation schedules and fast frequency response from data centers. Through case studies on both a benchmark system and a 2030 future scenario with higher DC penetration, we demonstrate that increasing the proportion of flexible DC load consistently improves system cost efficiency and supports renewable integration. However, this benefit exhibits diminishing marginal returns, motivating the introduction of the Marginal Flexibility Value metric to quantify the economic value of additional flexibility. The results highlight that as DCs become a larger share of system load, their active participation in frequency response will be increasingly indispensable for maintaining both economic and secure system operations.

Figures (8)

• Figure 1: Strategies for enabling data center flexibility
• Figure 2: Outline of Safe UC modelling.
• Figure 3: Assumed frequency response process of data center and conventional generators.
• Figure 4: An illustration on a simple decision tree with splitting knots.
• Figure 5: Comparison between conventional piecewise linear and decision tree linear models.