面向电网两票的风险评估与异常检测方法

Risk assessment and anomaly detection Method for Power Grid Work and Operation Tickets

摘要:

电网工作票与操作票（两票）审核结果的风险处置是电力安全管理的关键环节，现有方式在风险等级细粒度判定、上下文敏感性利用和结论可追溯性方面存在不足。为此，提出一种融合上下文敏感性建模与规则约束的两票风险评估方法：先建立审核问题到风险事件的映射机制，定义涵盖信息缺失、术语规范、对象冲突等七类风险的统一类型体系，再从问题特征、规则违反强度和上下文敏感性构建风险特征向量，经加权评分与阈值分级完成局部评估与票据级聚合，最后设计异常票识别与证据链生成模块，使风险结论可分级、可预警、可追溯。在 10 000 条样本上的实验中，该方法的风险准确率、Macro-F1、二次加权 Kappa、异常票 F1 和证据匹配准确率分别为 0.834 7、0.873 1、0.891 5、0.964 0 和 0.962 0，均优于规则与纯语义等基线，表明两者融合可有效提升风险分级准确性与异常票识别能力。

Abstract:

Risk disposition of audit results for power grid “two-ticket” system (i.e., highly constrained oper-ation and work dispatch instructions, including work tickets and operation tickets) is a critical step in power safety management. However, existing approaches are limited in fine-grained risk grading, context sensitivity utilization, and traceability of risk conclusions. A risk assessment method integrating context sensitivity modeling and rule constraints was proposed. First, a map-ping mechanism from audit problems to risk events was established, with a unified taxonomy of seven risk types. Then, risk feature vectors were constructed from problem features, rule vio-la-tion intensity, and context sensitivity, and a weighted scoring and threshold grading scheme was designed for local risk evaluation and ticket-level risk aggregation. Finally, an anomaly ticket identification mechanism and a risk evidence chain generation module were developed to make risk conclusions gradable, alertable, and traceable. Experiments on 10 000 two-ticket samples showed that the proposed method achieved a risk accuracy of 0.834 7, a Macro-F1 of 0.873 1, a quadratic weighted Kappa of 0.891 5, an anomaly ticket F1 of 0.964 0, and an evidence matching accuracy of 0.962 0,outperforming all baselines and validating the effectiveness of context sensi-tivity modeling and rule constraint fusion in improving risk grading accuracy and anomaly ticket identification.