基于LSTM的汽轮发电机线圈的早期异常检测  

Early Anomaly Detection of Turbo-generator’s Coil based on LSTM

作　　者：陈聪 徐俊元 王尊 CHEN Cong;XU Junyuan;WANG Zun(China Power Hua Chuang Electricity Technology Research Co.,Ltd..,Shanghai 200086,China;China Power Hua Chuang(Suzhou)Electricity Technology Research Co.,Ltd.,Suzhou 215123,China)

机构地区：[1]中电华创电力技术研究有限公司,上海200086 [2]中电华创(苏州)电力技术研究有限公司,江苏苏州215123

出　　处：《大电机技术》2022年第4期6-11,共6页Large Electric Machine and Hydraulic Turbine

基　　金：中国电力国际发展有限公司科技项目《发电机定子绕组热故障诊断技术研究》(2020-005-ZGDL-KJ-X)。

摘　　要：汽轮发电机线圈的异常检测通常以定子线棒出水温度作为重要指标。目前广泛使用的温度阈值检测法为了降低温度波动的影响而采取较高的阈值,检出异常相对滞后。为了更早发现异常,本文提出一种基于长短时记忆神经网络(LSTM)的预测性检测模型,利用定子线棒出水温度和定子电流时间序列对短期出水温度最大温差进行预测,以预测值与真实值的均方根误差(RMSE)作为异常检测的判断依据。最后,在型号为QFSN-660-2-22的汽轮发电机上进行了实验,结果表明该方法能够在异常程度不足以达到现有检测阈值时,提前检出异常,并且优于传统预测方法。因此,本文所提方法能够应用于汽轮发电机线圈的早期异常检测。The outlet-water temperature of stator bars is an important indicator in anomaly detection of turbo-generator’s coil.At present,the highly applied monitoring method based on temperature threshold adopts a relatively high threshold to reduce the impact of temperature fluctuation,so that the anomaly detection lags behind.In order to detect anomaly earlier,a predictive monitoring method based on long short-term memory(LSTM)neural network is proposed in this paper.It can predict the maximum difference sequence of the outlet-water temperature of stator bars in a short time based on the sequences of outlet-water temperature and stator current.And the root mean square error(RMSE)between the predicted sequence and the real sequence is used for anomaly detection.Finally,experiments are carried out on the QFSN-660-2-22 turbo-generator.The results show that this method can detect slight anomaly before the indicator reaches the threshold and performs better than traditional predictive method.Therefore,the method proposed in this paper is suitable for early anomaly detection of turbo-generator’s coil.

关 键 词：早期异常检测 定子线棒 出水温度 LSTM 

分 类 号：TM311[电气工程—电机]