Smart Datacenter Energy Management Powered by Convexhull and Physics Informed Machine Learning

The rising energy demand in data centers presents a critical operational challenge for telecommunication operators. This challenge arises from the exponential growth of heterogeneous data traffic, which intensifies both energy consumption and operational costs. To mitigate this issue, an adaptive temperature regulation framework is proposed, wherein a convex hull mechanism is employed to learn hotspot regions and provide real-time thermal insights to a physics-informed DeepONet that models the heat generation dynamics within the data center. The DeepONet subsequently extracts hotspot features and informs the HVAC system for intelligent, energy-efficient temperature control. The HVAC control problem is formulated as a Markov Decision Process (MDP) and optimized using the Asynchronous Advantage Actor-Critic (A3C) algorithm to jointly enhance energy efficiency and thermal stability. Simulation results demonstrate that the proposed ConvexHull-DeepONet-A3C (CDA3C) framework achieves a 0.10% higher F1-score, 0.10% lower convergence loss, and 0.30% lower energy consumption compared with baseline DeepONet and A3C models.