Dual-Timescale Carbon Efficiency Optimization for AC Constrained Cellular Networks

Abstract

To address surging carbon footprint in cellular networks, this letter proposes an integrated dual-timescale framework for coupled communication-grid systems, specifically incorporating non-linear Alternating Current (AC) power flow constraints to ensure grid stability. We formulate a carbon efficiency maximization problem and employ Dinkelbach’s reformulation with Lagrangian decomposition to decouple the joint problem into large-timescale grid dispatch and small-timescale power allocation subproblems. Specifically, a novel Physics-Informed Graph Neural Network is developed to solve non-linear AC power flow by embedding physical laws into the learning architecture, while a distributed subgradient method optimizes power allocation under dynamic traffic demands. Simulation results validate that the proposed scheme enhances carbon efficiency by 10.4% and 10.0% gains over DC-Only without and with interference