A Collaborative Optimization Method for Integrated Energy Systems Based on an LLM-Assisted Carbon Quota Constraint Mechanism

The carbon-factor accounting method is widely used for carbon emissions (CEs) evaluation and carbon quotas (CQs) allocation in integrated energy systems (IESs). However, its linear mapping model cannot capture the real-time influence of external and environmental factors, which weakens the constraint effect of CQs on CEs and limits the overall energy–carbon optimization capability. To address this issue, this article proposes a large language model (LLM)-assisted deep reinforcement learning (DRL) optimization method to enhance the constraint effect of CQs on CEs in IESs. First, a nonlinear CQ modeling method based on LLM semantic reasoning is proposed, breaking the dependence of the linear carbon-factor method on expert experience. Second, considering information including energy structure, market changes, policy orientation, and environmental constraints, an interpretable nonlinear CQ accounting method is designed based on LLM to enhance the constraint effect of CQs on CEs. Finally, a trigger mechanism is designed to achieve collaborative optimization through automatic interaction between LLM and DRL. Simulation results indicate that the optimized CQ mechanism enforces a more effective constraint on CE behaviors, enabling timelier response and enhanced energy–carbon optimization performance.