PULSE: A Modular Framework for Predictive Energy Efficiency in Heterogeneous Data Centers

Abstract

As digital transformation accelerates, data centers have become critical infrastructure, supporting cloud services, artificial intelligence, and real-time applications. This increasing demand has led to significant growth in energy consumption, raising concerns about operational costs and environmental impact. Power Usage Effectiveness (PUE) is the prevailing metric used to assess energy efficiency in data centers, yet optimizing PUE remains a complex and highly site-specific challenge. While prior research has explored the use of machine learning and control systems to reduce energy usage, most existing tools are narrowly focused, difficult to adapt across heterogeneous data center environments, or inaccessible to non-expert users. These limitations hinder the practical application of AI in dynamic and diverse operational contexts. This work introduces PULSE (PUE Unified Learning and Simulation Engine), a novel software platform that integrates deep learning-based prediction models with a natural language assistant to support PUE optimization in real-world settings. The platform fills a gap in current research by offering a user-friendly, modular, and adaptive solution that empowers operators to model, predict, and improve data center energy performance with minimal technical expertise. Results highlight PULSE’s ability to facilitate accessible, data-driven decision-making while adapting to unique infrastructure profiles. PULSE contributes to more intelligent, sustainable, and operationally efficient data center management.