In view of the characteristics of gas station safety accident case text, such as no fixed template, different accident descriptions and different length, and the shortcomings of common data extraction methods, such as complex process, low accuracy and poor universality, a new method of structured data extraction from gas station safety accident case text is proposed. For 165 unstructured electronic texts of gas station safety accident cases collected, 2475 structured data corresponding to 165 cases were extracted through case text preprocessing, pattern knowledge construction for large model prompt, large model evaluation based on pattern knowledge and training and fine-tuning based on orange model; In view of the fact that manual verification of the extracted data is time-consuming and laborious, an automatic verification algorithm for the extracted structured data is compiled by using Phthon3.9.0, and the test passes. The verification results show that the method is superior to the traditional methods in timeliness, accuracy, simplicity and so on, which is can deal with new tasks in other fields under the condition of zero training samples, and have cross-domain general capabilities.