Energy Consumption Optimization for 5G Base Stations Based on Deep Reinforcement Learning

With the rapid development of 5G mobile internet, the large-scale deployment of 5G base stations has led to a significant increase in energy consumption. Traditional deep reinforcement learning (DRL) methods face significant challenges due to the immense search space, often resulting in local optima. To address these issues, this paper proposes an innovative Deep Reinforcement Learning-based framework for Online Base Station Switching Decisions(DRLOS), which quantifies base station switching decisions and optimizes action selection, significantly reducing the action space and improving the algorithm's convergence performance and computational efficiency. As an online algorithm, DRLOS dynamically generates base station switching decisions based on real-time traffic variations and optimizes bandwidth and power allocation. The numerical results show that compared with existing distributed multi-agent RL algorithm, the proposed algorithm reduces energy consumption by 58.37%.