摘要
针对电网工程项目业务范畴广、审计环节多、事件关联复杂,导致审计管理与数字化作业融合困难的问题,提出一种基于事件抽取与核查预警的电力工程项目审计方法。该方法首先构建基于事件框架的审计业务过程,减轻建模人员的工作量与业务扩展压力;其次采用模板匹配结合Transformer的方式,将固定格式文本识别后的审计事件作为远程学习知识,生成具有明确提示知识的触发词模板,并根据不同项目主题按需补充知识,从而提升非固定格式文本的事件抽取准确率;最后将识别出的各类审计事件匹配至不同审计阶段及事件框架中,以支持跟随项目进度对相关审计事件进行核查与预警。验证实验表明,在低人工标记文本中,该审计方法事件抽取准确率不低于93%,且提取事件可准确嵌入审计业务过程中,支持对关联审计事件的异常核查预警。
A power engineering project audit method based on event extraction and verification warning is proposed to address the difficulties in integrating audit management and digital operations caused by the wide scope of business,multiple audit processes,and complex event correlations in power grid engineering projects.This method first constructs an audit business process based on an event framework to reduce the workload and business expansion pressure of modeling personnel;Then,using template matching combined with Transformer,the audit events recognized from fixed format text are used as remote learning knowledge to generate trigger word templates with clear prompt knowledge,and knowledge is supplemented as needed according to different project topics,thereby improving the accuracy of event extraction from non fixed format text;Finally,the identified audit events will be matched to different audit stages and event frameworks to support the verification and early warning of relevant audit events in accordance with project progress.The verification experiment shows that in low manually annotated text,the accuracy of event extraction using this audit method is not less than 93%,and the extracted events can be accurately embedded into the audit business process,supporting abnormal verification and early warning of related audit events.
作者
丁鹏程
韩琪
DING Pengcheng;HAN Qi(State Grid Dongying Power Supply Company,Dongying 257100,China)
出处
《软件导刊》
2024年第10期40-47,共8页
Software Guide
基金
国家电网科技项目(520616230005)。
