摘要
针对发电机定转子潜在缺陷严重影响机组运行安全稳定性的问题,文中提出了基于VGG网络的发电机定转子智能诊断算法。相比于Alex网络,VGG网络采用多个堆叠的小尺寸卷积滤波器代替大尺寸滤波器,减少了算法参数规模,加深了网络结构深度。文中所提算法包括离线训练和在线监测两部分,前者通过本地服务器发电机定转子历史图像进行学习训练,获得满足精度要求的VGG网络模型;后者利用训练好的VGG网络模型,实现发电机定转子的在线实时监测。通过仿真实验表明,相比于Alex网络,文中所提算法训练过程收敛速度更快,计算误差更小,对发电机定转子缺陷识别准确率更高。
In view of the problem that the potential defects of generator stator and rotor seriously affect the safety and stability of unit operation,an intelligent diagnosis algorithm of generator stator and rotor based on VGG network is proposed.Compared with Alex network,VGG network uses multiple stacked small size convolution filters instead of large size convolution filters,which reduces the size of algorithm parameters and deepens the depth of network structure.The proposed algorithm includes two parts:offline training and online monitoring.The former part uses the local server generator stator and rotor historical images for learning and training to obtain the VGG network model which meets the accuracy requirements.The latter part uses the trained VGG network model to realize the online real-time monitoring of generator stator and rotor.The simulation results show that compared with the Alex network,the proposed algorithm has faster convergence speed,smaller calculation error and higher recognition accuracy for generator stator and rotor defects.
作者
李诚
刘昊
蒋希峰
吴军法
韩文刚
高建国
LI Cheng;LIU Hao;JIANG Xifeng;WU Junfa;HAN Wengang;GAO Jianguo(Zhejiang Heika Electric Co.,Ltd.,Hangzhou 311100,China)
出处
《电子科技》
2021年第11期62-66,共5页
Electronic Science and Technology
基金
国网新源控股有限公司科技项目(525736200005)。
关键词
定转子
卷积神经网络
诊断
缺陷
Alex网络
发电机
缺陷识别
在线实时监测
stator and rotor
convolutional neural network
diagnosis
defect
Alex network
generator
defect identification
online real-time monitoring
作者简介
李诚(1987-),男,工程师。研究方向:信息通信与智能系统;吴军法(1991-),男,工程师。研究方向:光电信息工程。
