Abstract: Traditional ecological restoration assessment methods for mining areas still face limitations in terms of accuracy, efficiency, and technological autonomy. The rapid development of large language models （LLMs） has opened new opportunities for intelligent ecological restoration assessment. Based on multiple domestic LLMs and Huawei’s Ascend platform, this study proposed an intelligent assessment method for ecological restoration in mining areas. Using remote sensing datasets from typical mining areas in Guizhou Province, prompt engineering and retrieval-augmented generation techniques were employed to guide the models in analyzing ecological restoration conditions from multi-temporal imagery. The optimal model is determined via ensemble learning to generate assessment results. Results demonstrate that domestic LLMs present strong performance in evaluating ecological restoration effects in mining areas. By integrating the strengths of different models, the method effectively reduces the bias risk of single models, and objectively assesses the ecological restoration status of mining areas over time while automatically generating evaluation reports. The method excels in accuracy, automation, and assessment consistency, significantly reducing manual intervention and improving evaluation efficiency. This research highlights the application potential of LLMs in intelligent ecological restoration assessment for mining areas and provides a feasible pathway for building an independent, domestically controlled evaluation system, demonstrating promising practical application prospects.