Intelligent Energy Management of 5G Base Stations under Electricity-Shortage Conditions Based on Network-Load Forecasting

Abstract. This paper addresses energy-efficient operation of fifth-generation (5G) mobile base stations under conditions of electricity shortage and constrained power supply, an increasingly critical concern for telecommunication infrastructure operating as part of national critical infrastructure. An intelligent adaptive energy-saving system is proposed that combines short-horizon network-load forecasting with predictive, constraint-aware control of the radio access network (RAN). Continuous analysis of traffic and key-performance-indicator telemetry feeds machine-learning forecasters (Random Forest, XGBoost, and a Long Short-Term Memory network), whose 1–6 h load predictions drive a model-predictive controller that schedules advanced sleep modes, transmit-power adaptation, and selective deactivation of surplus carriers and sectors, subject to an explicit quality-of-service (QoS) constraint. The distinctive feature of the approach is the coupling of predictive load models with real-time resource control, which anticipates demand rather than reacting to it, thereby preserving service quality while reducing energy use. In a simulation study based on a synthetic diurnal traffic profile and an EARTH-type base-station power model, the forecast-based controller reduced modelled daily site energy by approximately 24 % (with a 20–31 % range across traffic profiles and horizons) relative to an always-on baseline, at a modelled QoS degradation below 1 %, dominating both rule-based and aggressive strategies on the energy–QoS trade-off. The results indicate that forecast-driven energy management can materially improve the energy autonomy and resilience of 5G infrastructure during power-supply crises and large-scale outages. All numerical results reported here are modelled (simulated) and require empirical validation before deployment.

Keywords: 5G; base station; energy efficiency; network-load forecasting; machine learning; LSTM; quality of service; cyber-physical system; critical-infrastructure resilience; demand response.