Authors:
Sivaprakash Sunkara

Addresses:
Department of Energy, Utilities and Application Development, Tata Consultancy Services, Naperville, Illinois, United States of America.

Abstract:

This paper addresses the key question of planning for the exponential growth in electricity demand driven by the hyper-growth of AI/DL data centres. Problems addressed using traditional probabilistic planning approaches are difficult to model because they involve complex, nonlinear, and high spatial-temporal correlations, as in packed AI computing workloads. Researchers introduce a new approach to generating high-fidelity stress-inducing load profiles that present worst-case scenarios for distribution grid operation using Generative AI, specifically Conditional Diffusion Models. A custom version of the IEEE distribution test feeder dataset is adapted for this study, which includes artificial intelligence workload patterns that can be generated. By conditioning the diffusion process on peak computing events and thermal thresholds, Researchers obtain realistic grid stress scenarios that are not captured by standard historical data interpolation. Researchers implemented it using Python and PyTorch for model training, and OpenDSS for power flow validation. Numerical simulations indicate that the developed generative-based attack procedure can more accurately locate hidden grid instability (e.g., voltage collapse and transformer thermal overload) than conventional Gaussian-based Monte Carlo methods. This work offers utility planners a powerful AI toolkit for proactively addressing grid risks associated with the exploding telecommunications infrastructure.

Keywords: Generative AI; Distribution Grid; Diffusion Models; Load Planning; Grid Resilience; Data Interpolation; Telecommunication Infrastructure; Monte Carlo Approaches.

Received on: 26/12/2024, Revised on: 05/03/2025, Accepted on: 30/04/2025, Published on: 07/12/2025

DOI: 10.69888/FTSIN.2025.000550

FMDB Transactions on Sustainable Intelligent Networks, 2025 Vol. 2 No. 4, Pages: 190-197

• Views : 86
• Downloads : 14

Download PDF