Abstract: In order to achieve digital and intelligent upgrading of traditional industrial parks, and to support high-quality regional development, there is an urgent need for intelligent dispatch models in parks. Therefore, this paper combines the smart park management system and the physical model of the park-level integrated energy system (PIES) to establish a cluster architecture of the PIES. A three-stage collaborative operation method for green electricity trading in the PIES cluster is proposed to solve the problem of green electricity trading in multiple PIES, and to achieve accurate prediction of distributed renewable energy and on-site consumption. Firstly, based on the large language model LLAMA-7B, green electricity prediction is achieved, and further green electricity power is divided into purchasing and selling electricity PIES. Secondly, based on the green electricity price quota curve prediction model and dynamic green electricity pricing strategy, the differential prices for green electricity trading between parks is established. On this basis, a multi-objective low-carbon economic optimization operation model is established to solve the contradiction between economic and environmental factors brought about by the green electricity exchange. The example analysis shows that the proposed model can comprehensively schedule the economic cost, actual carbon emissions, and renewable energy utilization rate of the PIES cluster, which has a positive promoting effect on the intelligent dispatch of multiple PIES.