A Digital Twin-Driven Framework for Sustainable Computing Infrastructure Using Predictive Analytics and Intelligent Energy Scheduling

Abstract

Sustainable computing infrastructure requires coordinated decisions across server utilization, cooling, workload placement, and electricity carbon intensity. This paper develops a digital twin driven framework that integrates telemetry synchronization, probabilistic demand forecasting, thermal and power estimation, and constrained intelligent scheduling. The proposed twin combines a physical resource model with learned corrections, exposes uncertainty to a receding horizon optimizer, and tests candidate schedules before execution. A reproducible illustrative scenario is specified for a 24 hour, 100 server facility with time varying demand, renewable availability, and grid carbon intensity. All numerical results are explicitly synthetic demonstration values, not outcomes of a deployed system or measured experiment. In the scenario, the integrated policy reduces daily electricity from 2,400 to 1,944 kWh and location based emissions from 1,104 to 777.6 kg CO₂e while meeting an assumed 99.5% service level target. Sensitivity and ablation analyses show the potential importance of forecast quality, thermal constraints, and carbon aware scheduling. The framework provides a testable research protocol, rather than a claim of validated operational performance.