Abstract: integrated energy systems (IES) improve energy efficiency through coordinated optimization of electricity, heat, cooling, and gas, yet their scheduling optimization involves multienergy coupling, multidevice coordination, and complex constraints. Conventional approaches rely on experts to manually construct optimization models, suffering from high technical barriers, time-consuming development cycles, and limited transferability. In recent years, large language models (LLMs) have shown promise in natural language understanding and code generation; however, their direct application to IES scheduling optimization still faces challenges, including requirement ambiguity, attention drift in long contexts, and high debugging costs. To address these issues, this study proposes MASEO, a multiagent collaborative framework for IES scheduling optimization. MASEO offers three main contributions. First, the optimization workflow is decomposed into four sequential stages-information acquisition, mathematical modeling, code implementation, and execution validation-each handled by a dedicated LLM agent operating in series. Second, a structured checklist is introduced to standardize the information collection process, complemented by an error-classification-based traceable repair mechanism that routes failures to the responsible upstream agent for targeted correction. Third, case studies on a Beijing data center IES and a German community IES validate the framework: the annualized cost deviation in the Beijing scenario is approximately 1.4%, and the total cost in the German scenario is consistent with the expert benchmark. Compared with a single-agent baseline, modeling accuracy improves by approximately 35% and the code execution pass rate increases from 63% to 94%; ablation studies further validate the independent contribution of each core mechanism; multibackbone and sensitivity experiments further provide configuration guidance for deployment. The results suggest that MASEO can help reduce the dependence of IES scheduling optimization on specialized modeling expertise, offering a reference pathway for the reliable deployment of large language models in energy system scheduling optimization.

Similar articles