Stable energy management for highway electric vehicle charging based on reinforcement learning

Abstract

With the growing global awareness of carbon neutrality and environmental protection, the rapid increase in electric vehicles poses an urgent challenge for highway energy management: how to achieve stable and rational scheduling of the power supply system. Previous research has utilized reinforcement learning to achieve significant success in the scheduling decisions of power supply systems, demonstrating its immense potential. However, achieving long-term stable and environmentally friendly power supply scheduling strategies in large-scale and complex highway energy management systems remains a significant challenge in current research. To fill this gap, we propose HEM-GPT, a large-scale highway energy management framework based on the Generative Pre-trained Transformer architecture. This framework includes an efficient representation module for predicting long-term power supply decision actions and a stable decision-making learning paradigm to enhance the robustness and generalization ability. By applying a linear Q-value decomposition method to the action space, HEM-GPT can effectively reduce the computational burden and complexity of the decision space in large-scale systems. Furthermore, we implement an online adaptive fine-tuning mechanism to bolster the model’s stability and its adaptability to new scenarios. The results show that HEM-GPT reduces the cost by 45.5% compared to the best baseline in terms of long-term scheduling capability for the future.