基于多传感器信息融合与混合感受野残差卷积神经网络的调相机转子故障诊断  

Fault Diagnosis of Synchronous Condenser Rotor Based on Multi-sensor Information Fusion and Mixed Receptive Field Residual Convolutional Neural Network

作　　者：钱白云 吕朝阳 张维宁 林翔 朱霄珣[2] 董利江 吴玉华 王鲁东 李震涛 QIAN Baiyun;L Zhaoyang;ZHANG Weining;LIN Xiang;ZHU Xiaoxun;DONG Lijiang;WU Yuhua;WANG Ludong;LI Zhentao(Electric Power Research Institute of Xinjiang Power Grid Corporation,Urumchi 830000,China;Department of Power Engineering,North China Electric Power University,Baoding 071003,China;State Grid Xinjiang Electric Power Co.,Ltd.,Urumchi 830000,China;Urumqi Electric Power Construction and Commissioning Institute of Xinjiang Xinneng Group Co.,Ltd.,Urumchi 830000,China)

机构地区：[1]国网新疆电力有限公司电力科学研究院,乌鲁木齐830000 [2]华北电力大学动力工程系,河北保定071003 [3]国网新疆电力有限公司,乌鲁木齐830000 [4]新疆新能集团有限责任公司乌鲁木齐电力建设调试所,乌鲁木齐830000

出　　处：《计算机测量与控制》2023年第9期29-35,共7页Computer Measurement &Control

基　　金：国家电网有限公司科技项目资助(SGXJDK00NYJS2200102)。

摘　　要：大型调相机是电网对无功调节的关键设备之一,预防调相机因转子故障而产生的安全事故极为重要;为了有效利用设备实际监测过程中多源传感器同步采集的海量数据,并考虑传统卷积神经网络(CNN,convolutional neural network)模型的特征学习能力差、感受野尺度单一等缺点,提出了一种基于SDP-MRFRCNN的调相机转子故障诊断方法;首先通过对称点模式(SDP,symmetrized dot pattern)将调相机多个传感器的振动信号进行信息融合,获取融合多源振动信息的图像,然后构建混合感受野残差卷积神经网络(MRFRCNN,mixed receptive field residual CNN)进行学习,实现调相机转子状态识别;实验结果表明,该方法增强了不同状态特征间的辨别度,具有更高识别精度,分类准确率达到了99.33%。A large-scale synchronous condenser is one of the key equipment with reactive power regulation in power grid.It is very important to prevent the safety accident of synchronous condenser caused by rotor failure.In order to effectively use the massive data synchronously collected by multi-source sensors in the actual monitoring process of the equipment,and considering the shortcomings of traditional convolutional neural network(CNN)models such as poor feature learning ability and single scale of receptive field,a fault diagnosis method of synchronous condenser rotor based on symmetrized dot pattern and mixed receptive field residual convolutional neural network(SDP-MRFRCNN)is proposed.Firstly,the vibration signals of multiple sensors of the synchronous condenser are fused by the symmetrized dot pattern(SDP)to obtain the image fused with multi-source vibration information.Then,the MRFRCNN is constructed to learn and realize the rotor status recognition of the synchronous condenser.The experimental results show that this method enhances the discrimination between different state features,and it has higher recognition accuracy,with an average classification accuracy of 99.33%.

关 键 词：调相机 转子振动 多传感器融合 卷积神经网络 多感受野 残差结构 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]