Constrained Discrete Bayesian Optimization for Base Station Sleep Mode in Cellular Networks
Abstract
Sleep mode strategies, which deactivate underutilized base stations (BSs), are a promising solution for reducing the energy consumption of cellular networks. However, selecting which BSs to deactivate is a challenging combinatorial optimization problem under Quality of Service (QoS) constraints, such as cell-edge throughput and coverage. Moreover, evaluating the impact of a sleep mode decision on the QoS is costly, since it requires either digital twin-like system-level simulation or testing on the live network. In this paper, we propose a discrete constrained Bayesian Optimization (BO) framework for BS deactivation. We use Random Forest (RF) surrogates for throughput and coverage to build a low-complexity acquisition function (AF) that combines active-BS reduction with joint constraint satisfaction. Numerical results show that the proposed method outperforms greedy heuristics and random search, and achieves competitive energy-QoS trade-offs with respect to a constrained discrete BO baseline, while reducing the computational complexity of the BO decision step.