Network Energy Saving Techniques Aided by AI/ML in 5G Networks

Inhalt:
This paper addresses 5G mobile networks, their energy consumption and potential optimisation using energy efficiency techniques with the aid of machine learning. A real-world traffic dataset was provided by Vodafone, containing data from sites in two countries. This dataset was carefully analysed to identify potential energy-saving opportunities that could be stimulated. A time-based energy efficiency technique was developed, aimed at turning off base station components during low or no traffic periods. One aims at predicting low-traffic periods, enabling pre-emptive energy-saving actions that reduce unnecessary power consumption without compromising network performance. To achieve this, an LSTM model was developed, leveraging traffic predictions to guide energy-saving decisions dynamically. The algorithm is designed to deliver substantial energy efficiency, while also being customisable to meet the specific requirements of operators and remaining easy to implement and maintain. The algorithm achieves a high predictive accuracy, showing good results, with potential energy savings up to 2.452 GWh annually, translating to a total reduction of 405.7 tons in CO2 emissions.