Knowledge-Data Hybrid-Driven Low-Energy Resource Allocation in Cellular Networks

Due to the growing demand for services and the dense deployment of base stations, the substantial energy consumption has emerged as a significant impediment to the development of cellular networks. Although numerous energy-efficient methods have been proposed, the energy consumption of cellular networks continues to exhibit an upward trend. This is because the existing methods may provide superfluous services to users to achieve higher energy efficiency, resulting in extra energy consumption. In this paper, a low-energy resource allocation method is proposed to minimize the total energy consumption, considering the long-term service demand of each user. However, this problem is challenging to solve due to the the interaction of resource allocation across sequential time slots, such as data queuing and service satisfaction. Therefore, the initial problem is decoupled into multiple subproblems to relax complexity by Lya-punov optimization. Moreover, a knowledge-data hybrid-driven deep reinforcement learning framework is proposed to solve each subproblem. Specifically, the knowledge module is embedded into DRL to make decisions for low-energy based on exact service demand, without requiring extensive exploration and learning. Simulation results show that the energy consumption of the proposed method is reduced by 20.7% compared with the baseline, and the convergence speed is increased by two times.