Abstract: Addressing the difficulty of existing IEC 61850 Substation Configuration Description (SCD) file verification methods in detecting deep-seated configuration flaws, this paper proposes an intelligent analysis method integrating Large Language Models (LLMs). The method extracts structured information through XML parsing, performs fundamental validation using a rule engine, and emphasizes the use of LLMs for semantic logic analysis, design rationality checks and design intent consistency verification. The results are ultimately integrated to generate a report. Using the example of non-standard configuration in the breaker failure trip logic for the high-voltage side of the main transformer, the method is proved to effectively identify risks overlooked by traditional tools and non-compliant configurations. The research indicates that the introduction of LLMs significantly enhances the depth and breadth of SCD validation, providing reliable support for power grid security verification.