Dual-Loop Control in DCVerse: Advancing Reliable Deployment of AI in Data Centers via Digital Twins

2026-04-08

AI Summary

This work addresses the challenges of deploying deep reinforcement learning (DRL) for energy-efficient data center control, where high risks of service disruption, data scarcity, and the absence of real-time policy validation hinder reliable adoption. To overcome these limitations, the authors propose a Digital Twin-based Dual-Loop Control Framework (DLCF), which introduces—for the first time—a dual-loop architecture enabling real-time policy pre-evaluation and expert intervention. By tightly integrating the physical system, a digital twin, and a diverse DRL policy library, DLCF establishes a closed-loop pipeline for data assimilation, training, validation, and optimization. This approach substantially enhances DRL’s sample efficiency, generalization, safety, and interpretability. Validated on a real-world data center cooling system, the framework achieves up to 4.09% energy savings while strictly adhering to service-level agreement (SLA) requirements, thereby improving the trustworthiness and deployability of AI-driven control strategies.

Abstract

The growing scale and complexity of modern data centers present major challenges in balancing energy efficiency with outage risk. Although Deep Reinforcement Learning (DRL) shows strong potential for intelligent control, its deployment in mission-critical systems is limited by data scarcity and the lack of real-time pre-evaluation mechanisms. This paper introduces the Dual-Loop Control Framework (DLCF), a digital twin-based architecture designed to overcome these challenges. The framework comprises three core entities: the physical system, a digital twin, and a policy reservoir of diverse DRL agents. These components interact through a dual-loop mechanism involving real-time data acquisition, data assimilation, DRL policy training, pre-evaluation, and expert verification. Theoretical analysis shows how DLCF can improve sample efficiency, generalization, safety, and optimality. Leveraging DLCF, we implemented the DCVerse platform and validated it through case studies on a real-world data center cooling system. The evaluation shows that our approach achieves up to 4.09% energy savings over conventional control strategies without violating SLA requirements. Additionally, the framework improves policy interpretability and supports more trustworthy DRL deployment. This work provides a foundation for reliable AI-based control in data centers and points toward future extensions for holistic, system-wide optimization.

Problem

• data centers
• energy efficiency
• outage risk
• Deep Reinforcement Learning
• reliable AI deployment

Innovation

• Dual-Loop Control
• Digital Twin
• Deep Reinforcement Learning
• Data Center Optimization
• Policy Pre-evaluation