Towards Sustainable Mobile Networks: AI for Zero Touch Automated Battery Control

With the adaption of 5G technology, a massive increase in the number of devicesand applications that utilize mobile communication systems is expected.The mobile communications sector experiences ever-increasing electricity andcarbon footprint costs due to new services and increased demand. Radio accessnetworks (RAN:s) are one of the major contributors to the global carbonfootprint of mobile communication systems, and stand for the main part ofthe power consumption. If the current industry trends continue, the sustainabilityagenda of the telecommunications sector may be compromised. Novelsolutions and improvements are required in order to reduce the costs andcarbon footprint in the sector, while providing growth to society by enhancingmobile communication services. Further, these solutions should not belimited to improving energy efficiency and reducing power consumption, butalso focus on the power source itself.In this thesis, we contribute to tackling the above issues. We introducebatteries as a dynamic power supply in RAN base stations and employ LinearProgramming (LP), Mixed Integer Linear Programming (MILP) and ReinforcementLearning (RL) methods, in particular Deep Q-Network (DQN),to find optimal policies for battery control. This enables cost reductions forRAN operators, e.g., by charging batteries when the electricity price is lowand discharging them when the electricity price is high. The DQN approachis found to be applicable for cost reduction both in monetary and carbonfootprint terms.The results in the thesis have contributed to one patent application andtwo publications are under preparation