Tackling the Power Challenge of AI Data Centers: Dynamics Modeling, Forecasting, and Optimization

Yuzhuo Li – University of Alberta

The rapid growth of AI data centers introduces unprecedented challenges to power systems. Unlike conventional loads, AI workloads driven by large language models can exhibit algorithm-influenced, user-bahaviour-induced, fast, disruptive transient dynamics with sharp power surges and significant peak-to-idle ratios. This talk presents our recent efforts in addressing these challenges, spanning dynamic modelling of AI-induced power transients, data-driven load forecasting, and optimization strategies for power management and grid integration. The goal is to bridge the gap between AI computing infrastructure and power system operation. Together, these efforts aim to enable grid-friendly AI data center integration.