基于改进SVM的发电设备故障特征识别与分析  

Fault feature recognition and analysis of power generation equipment based on improved SVM

作　　者：钟伟津 蒋璆 朱海锋 刘春林 郑旭彬 ZHONG Weijin;JIANG Qiu;ZHU Haifeng;LIU Chunlin;ZHENG Xubin(China Southern Power Grid Energy Storage Co.,Ltd.,Guangzhou 511400,China)

机构地区：[1]南方电网储能股份有限公司,广东广州511400

出　　处：《电子设计工程》2025年第6期122-126,共5页Electronic Design Engineering

摘　　要：针对发电设备的长期运行过程中,在机械应力和不均匀热分布的影响下会产生疲劳裂纹、变形等问题,若未能及时发现并处理,则可能造成严重后果。针对这一问题,文中提出了一种融合卷积自编码器和多分类支持向量机的发电设备状态监测与故障诊断技术。该技术方案以发电设备热力图像、应力分布图像和裂纹分布图为监测指标,利用卷积自编码器对各监测图像的深度特征指标进行提取,再引入支持向量机算法对所提取指标进行故障诊断,从而实现对发电设备的无接触监测与诊断。通过对某抽水蓄能电站发电机进行的实验测试结果表明,所提方法对多种状态下的故障识别准确率均可达到85%以上,验证了该技术方案的有效性和工程价值。During the long-term operation of power generation equipment,fatigue cracks,deformation,and other problems may occur under the influence of mechanical stress and uneven heat distribution.If not detected and dealt with in a timely manner,serious consequences may occur.In response to this issue,a power generation equipment status monitoring and fault diagnosis technology combining Convolutional Auto Encoder and multi classification Support Vector Machine is proposed in the article.This technical solution uses thermal images,stress distribution images,and crack distribution maps of power generation equipment as monitoring indicators.Convolutional Auto Encoder are used to extract the depth feature indicators of each monitoring image,and Support Vector Machine algorithms are introduced to diagnose faults in the extracted indicators,thus achieving contactless monitoring and fault diagnosis of power generation equipment.The experimental test results on a pumped storage power station generator show that the proposed method can achieve fault identification accuracy of over 85%in various states,verifying the effectiveness and engineering value of this technical solution.

关 键 词：发电设备 深度学习 故障监测 支持向量机 卷积自编码器 

分 类 号：TP391.44[自动化与计算机技术—计算机应用技术] TN929.5[自动化与计算机技术—计算机科学与技术]