Return-to-Go Predicting Decision Transformer for Energy-Saving in 5G Base Stations with Photovoltaic Integration

To address the challenges of energy conservation, emission reduction, and the dual-carbon strategy, the integration of photovoltaic solar panels has become increasingly important in the energy management of 5G Base Stations (BS). Efficient energy scheduling is essential to accommodate the growing number of users and traffic demands. To optimize the energy efficiency and service quality of BS, this paper proposes an intelligent decision model based on the Decision Transformer (DT), named Return-to-Go Predicting Decision Transformer (PDT). This model incorporates training for the target return-to-go (RTG) during the training phase, enabling it to auto-matically select an appropriate RTG during inference based on the environment. This approach enhances the model's accuracy and performance in complex environments. Simulation results demonstrate the PDT can dynamically adjust the working status of base stations, as well as the use of photovoltaic solar panels and batteries, thereby achieving optimization of base station energy consumption and efficient utilization of photovoltaic energy, and its adaptability and robustness in scenarios with varying numbers of base stations surpass the current state-of-the-art reinforcement learning algorithms.