ABSTRACT To address the problems of insufficient accuracy and poor scenario adaptability in the manual preparation of on‐site electrical operation tickets for distribution networks, this paper proposes an intelligent ticket generation method that integrates domain knowledge enhancement and scenario‐based feature encoding, enabling controllable generation through the collaboration of rules and data. The method builds a distribution‐network safety operation rule base and a standardised terminology table as external knowledge modules that automatically complete mandatory safety steps and unify terminology before and after generation, while preserving the mapping between key steps and safety regulation clauses. Meanwhile, elements, such as voltage level, operation type and equipment identifier, are organised into structured scenario instructions and, together with the task description, are fed into a Chinese BART generation model; the model is responsible for generating the main content of the ticket, and the rule module acts as a safety backstop for key safety steps. Experiments conducted on an operation ticket dataset from a provincial power supply bureau, compared with RawBART, AblateRules and AblateFeat, show that the proposed method significantly outperforms the baselines in BLEU, ROUGE‐L, and the key‐step retention rate, with ROUGE‐L improved by 53. 58% over the baseline. The results indicate that the proposed framework improves the safety‐related step coverage, structural integrity and robustness of generated tickets in offline evaluations on historical data, showing promising potential for intelligent assistance in distribution‐network operation and maintenance. Further validation through expert human evaluation and real‐system shadow‐mode trials will be required to assess deployment reliability and on‐site usability. It should be noted that the model is currently validated on tickets collected under a single utility's regulation and naming system; its generalisability across regions, voltage levels and management practices remains to be verified with independent multi‐source datasets.
Hao et al. (Thu,) studied this question.