引用本文: | 周海成,石恒初,曾令森,王 飞,欧阳勇.基于关系超图增强Transformer的智能站二次设备故障诊断研究[J].电力系统保护与控制,2024,52(12):123-132.[点击复制] |
ZHOU Haicheng,SHI Hengchu,ZENG Lingsen,WANG Fei,OUYANG Yong.Fault diagnosis of an intelligent substation secondary device based on a relationalhypergraph-enhanced Transformer[J].Power System Protection and Control,2024,52(12):123-132[点击复制] |
|
摘要: |
随着智能变电站二次设备的状态感知与自描述能力不断提升,在提高电网调控细粒度的同时,其海量、驳杂、离散的状态信息也使故障诊断难度倍增。为提高二次设备故障诊断精度与效率,提出基于关系超图增强Transformer的二次设备故障诊断算法。首先利用Apriori算法挖掘故障信号间的关联规则,构建关系超图。然后利用超图卷积神经网络(hypergraph convolutional neural network, HGCN)和微调标准Transformer网络学习故障特征间的高阶关系和上下文表达,再经过误差反向传播、非线性传递函数预测故障类型。最后,以某地区一年的二次设备运行数据作为算例进行分析。结果表明,所提方法能够去除冗余信息干扰,准确定位故障元件和诊断故障类型,为智能运维提供支持。 |
关键词: 关系超图 超图卷积神经网络 Transformer 故障预测 二次设备 设备关联模型 |
DOI:10.19783/j.cnki.pspc.231475 |
投稿时间:2023-11-19修订日期:2024-04-25 |
基金项目:中国南方电网有限责任公司科技项目资助(YNKJXM20220093) |
|
Fault diagnosis of an intelligent substation secondary device based on a relationalhypergraph-enhanced Transformer |
ZHOU Haicheng1,SHI Hengchu2,ZENG Lingsen3,WANG Fei1,OUYANG Yong |
(1. Pu’er Power Supply Bureau of Yunnan Power Grid Co., Ltd., Pu’er 665000, China; 2. Yunnan Electric Power
Dispatching Control Center, Kunming 650011, China; 3. Wuhan Huadian Shuncheng Science Technology Co., Ltd.,
Wuhan 430072, China; 4. School of Computer Science, Hubei University of Technology, Wuhan 430072, China) |
Abstract: |
With the continuous improvement of state perception and self description capabilities of secondary equipment in intelligent substations, it not only improves fine-grained regulation of the power grid, but also doubles the difficulty of fault diagnosis because of its massive, complex, and discrete state information. To improve the accuracy and efficiency of secondary equipment fault diagnosis, a secondary equipment fault diagnosis algorithm based on a relational hypergraph-enhanced Transformer is proposed. First, a Priori algorithm is used to mine the association rules between fault signals and a relationship hypergraph is constructed. Then, a hypergraph convolutional neural network (HGCN) and a fine-tuned standard Transformer network are used to learn high-order relationships and contextual expressions between fault features, and then fault types are predicted through error backpropagation and a nonlinear transfer function. Finally, the annual operational data of secondary equipment in a certain region is taken as an example for analysis. The results show that the proposed method can remove redundant information interference, accurately locate faulty components and diagnose fault types, providing support for intelligent operation and maintenance. |
Key words: relational hypergraph HGCN Transformer fault prediction secondary equipment device association model |