基于LSTM的高压电网换流变故障诊断方法

A Fault Diagnosis Method for Converter Transformers in High-Voltage Power Grids Based on LSTM

中文关键词:换流变  故障诊断  长短期记忆  粒子群

keywords:converter transformer  fault diagnosis  long short-term memory（LSTM）  particle swarm optimization（PSO）

中文摘要:随着风光等清洁能源发电在电网中占比的逐渐增加，直流输电工程中的换流变压器一旦出现故障，将影响换流站整流或逆变工作的正常进行。基于换流变油色谱数据分析，提出一种行之有效的故障诊断方法。首先，建立长短期记忆网络（long-short-term-memory，LSTM）的换流变故障诊断模型，并选取换流变油色谱数据中的特征气体及其比值关系，发掘特征参量；其次，引入改进后的粒子群算法（improved particle swarm optimization，IPSO），优化LSTM 的 5 个超参数；最后，将1 213组换流变故障数据分为测试集、训练集后导入基于IPSO-LSTM的故障诊断模型。算例分析表明：所提方法能够诊断换流变故障，并有效区分故障类型，其诊断准确率达到93.44%，能准确反映换流变的运行状况。

Abstract:With the increasing proportion of clean energy power generation such as wind and solar in power grids，any fault of converter transformers in DC transmission projects will disrupt the normal rectification or inversion operation of converter stations. Based on the analysis of oil chromatographic data of converter transformers，an effective fault diagnosis method is proposed. First，a fault diagnosis model for converter transformers based on the long short-term memory （LSTM） network is established，and the characteristic gases and their ratio relationships in the oil chromatographic data of converter transformers are selected to extract characteristic parameters. Second，an improved particle swarm optimization （IPSO） algorithm is introduced to optimize five hyperparameters of the LSTM network. Finally，1213 groups of converter transformer fault data are divided into test and training sets and input into the fault diagnosis model based on IPSO-LSTM. Case analysis shows that the proposed method can diagnose converter transformer faults and effectively distinguish fault types，with a diagnosis accuracy of 93.44%，which can accurately reflect the operating status of converter transformers.