Abstract:In recent years, the automatic generation and intelligent matching of grid dispatch tasks have garnered widespread attention for enhancing operational safety and efficiency; however, challenges remain in accurately extracting information from large-scale text data and in achieving timely risk diagnosis. To address the complexity and efficiency bottlenecks inherent in the grid operation task generation process, this study employs a text processing technique that integrates deep pre-trained language models with conditional random fields, combined with a rule engine based on fault tree analysis, ultimately proposing a method for automatic generation and intelligent matching of operation tasks based on large models and rule engines. Experimental results demonstrate that, across five different text formats, the model achieves a recall rate of over 90%, with the highest coverage reaching 98.6% when the risk threshold is set at 0.6, and under marginal risk conditions, the misoperation rate is as low as 2.7%. These findings indicate that generation techniques based on large models can accurately extract task requirements and, through the rule engine, automatically match and generate operation tasks, thereby significantly enhancing the intelligence and automation of task generation. This research is of considerable significance for advancing the automation of grid dispatch and the development of safety control technologies in smart grids.