基于预训练语言模型的电网检修计划异常检测

A Pre-trained Language Model-based Anomaly Detection Method for Grid Maintenance Plans

吴海洋 ¹吴子辰 ²吴博科 ³高华 ²王莘然 ²胡伟⁴

扫码查看

作者信息

1. 国网江苏省电力有限公司南京 210000
2. 国网江苏省电力有限公司信息通信分公司南京 210000
3. 国网常州供电公司常州 213003
4. 南京大学计算机软件新技术全国重点实验室南京 210023
折叠

摘要

[目的/意义]电网检修计划异常检测技术可以发现电网检修计划内存在的异常,对于电网通信系统检修活动的正常开展和电力通信系统的稳定运行至关重要.目前,电网检修计划异常检测主要依靠人工判别的方式寻找检修计划的异常,存在准确率低与效率低下的问题.[方法/过程]提供了一种基于预训练语言模型的电网检修计划异常检测方法,使用电网的检修计划数据对预训练语言模型进行微调,使得预训练语言模型能够获取相关领域的专业知识,同时充分利用预训练语言模型的上下文感知能力和领域通用性,使预训练语言模型能够深入理解检修计划的复杂语境,从而发现检修计划中存在的异常.[结果/结论]实验证明,相较于传统方法,该异常检测方法能够有效优化电网检修计划异常检测的性能.

Abstract

[Objective/Significance]Anomaly detection methods are designed to detect anomalies in grid maintenance plans,which are crucial for the normal implementation of grid maintenance and the stable operation of the grid system.Existing methods primarily rely on manual discernment,leading to issues of low accuracy and inefficiency.[Methods/Processes]This paper presents an anomaly detection method for grid maintenance plans based on pre-trained language models(PLMs).The method uses grid maintenance plan data to fine-tune the PLMs,so that the PLMs can obtain professional knowledge in related fields,while fully leveraging the contextual awareness and domain adaptability of PLMs.The combination of the two aspects enables the PLMs can deeply understand the complex context of the maintenance plans and detect anomalies in them.[Results/Conclusions]Experimental results demonstrate that the proposed method effectively enhances the performance of anomaly detection in grid maintenance plans.

关键词

预训练语言模型/异常检测/电网管理/文本分类

Key words

Pre-trained Language Model/Anomaly Detection/Grid Management/Text Classification

引用本文复制引用

基金项目

江苏省电力有限公司项目(J2023108)

出版年

2024

情报工程

CSTPCDCHSSCD

ISSN：

段落导航