The Hierarchical Learning for Energy Saving of Base Stations

With the rapid advancement of 5G technology, there has been a significant increase in the number of 5G base stations (BSs), leading to a substantial rise in power consumption. Traditional artificial intelligence (AI) and deep reinforcement learning (DRL) methods face challenges related to decision cycles and exhibit suboptimal performance. To address these issues, this paper proposes a hierarchical learning framework that integrates an outer Binary Particle Swarm Optimization (BPSO) algorithm with an inner Multi-Agent Proximal Policy Optimization (MAPPO) algorithm, thereby significantly enhancing the convergence speed of the learning process. Moreover, this hierarchical framework effectively partitions and addresses different modules within the base station-specifically, the Remote Radio Unit (RRU) module and Building Baseband Unit (BBU) module. In comparison to existing algorithms, our proposed outer layer achieves a reduction in power consumption by 54.19%, while the inner layer further decreases power consumption by an additional 56.16%.