Abstract: During operation, power transformers continuously emit vibration signals, which contain a large number of pulses and fluctuations caused by mechanical faults. These signals serve as the primary data source for assessing the operating condition of power transformers. To address the issues of low diagnostic accuracy, poor real-time performance, and inability to effectively handle multi-fault scenarios in traditional power transformer fault diagnosis techniques. The acoustic fingerprint characteristics of internal vibration signals in power transformers were analyzed and a fault diagnosis model integrating dual-threshold preliminary screening and edge-level dynamic graph convolution based on acoustic properties was constructed. Firstly, a normal/fault operational state classification module was proposed to evaluate real-time transformer status using acoustic short-time energy and zero-crossing rate. Second, an Edge-Level Dynamic Graph Convolutional Network (ELDGCN) wass designed to adaptively learn spatial correlations among acoustic fingerprint signals by optimizing edge weights. Finally, the algorithm was deployed to real-world power transformer scenarios for testing. Experimental results show that the method proposed significantly improves the fault identification efficiency, with the recognition rate for datasets containing mixed faults reaching 98.17%.

Key words: power transformer, acoustic short-time energy, zero-crossing rate, fault detection, acoustic fingerprint signal, Edge-Level Dynamic Graph Convolutional Network (ELDGCN)

摘要： 电力变压器在运行过程中会持续发出振动信号，其中包含大量由机械故障引起的脉冲和波动，这些信号是评估电力变压器运行状况的主要数据来源。为解决传统电力变压器故障诊断技术诊断精度低、实时性差且无法有效应对多故障场景的问题，分析了电力变压器内部振动声信号的声纹特征，并基于声学特征构建了一个融合双阈值初筛与边级动态图卷积的故障诊断方法。首先，提出一种利用声学短时能量和过零率双阈值来评估电力变压器实时运行状况的正常/故障运行状态分类模块；其次，设计了一种边级动态图卷积网络（ELDGCN），通过优化边的权重自适应地学习声纹信号之间的空间相关性；最后将算法部署到真实的电力变压器场景下进行测试。实验结果表明，采用本文方法显著提升了故障的识别效率，数据集含混合故障的识别率达98.17%。

关键词: 电力变压器, 声学短时能量, 过零率, 故障检测, 声纹信号, 边级动态图卷积网络