Optimization of Energy and Carbon Emissions in Integrated Energy System Based on Deep Reinforcement Learning Assisted by Large Language Model
Liang Zhang, Dong Yue, Gerhard P. Hancke, Chunxia Dou, Liang Yu, Zhiqiang Chen
Abstract
Integrated energy system (IES) facilitates efficient energy conversion and utilization. However, the joint optimization of energy use and carbon emissions (CEs) remains a significant and widely recognized challenge in this field. In this article, to solve the problem, a novel decision-making framework is proposed with leveraging a large language model (LLM) to assist deep reinforcement learning (DRL). First, a dynamic priority trading strategy is designed based on real-time supply and demand, which is adjusted dynamically through a trading matrix. Furthermore, a bidirectional equilibrium pricing mechanism is designed to determine reasonable prices that balance the interests of trading parties. Finally, the powerful inference and analysis capabilities of the LLM are leveraged to optimize DRL algorithms through interactive iterations and feedback loops, thereby enhancing decision-making performance. The experimental results demonstrate that the improved algorithm outperforms the baseline algorithm in terms of cost control, CE limitation.