融合LSTM和DBM的电力电缆故障诊断方法研究  

Research on Power Cable Fault Diagnosis Method Integrating LSTM and DBM

作　　者：李楠 温雯 LI Nan;WEN Wen(State Grid Hebei Electric Power Co.,Ltd.,Handan Power Supply Branch,Handan 056004,China;State Grid Hebei Electric Power Co.,Ltd.,Daming Power Supply Branch,Handan 056900,China)

机构地区：[1]国网河北省电力有限公司邯郸供电分公司,河北邯郸056004 [2]国网河北省电力有限公司大名供电分公司,河北邯郸056900

出　　处：《现代建筑电气》2024年第12期6-11,共6页Modern Architecture Electric

摘　　要：在电力系统正常安全运行中,电力电缆易受到自然气象以及人为因素造成的电缆损坏等故障,严重破坏了电力线路的正常运行,降低了电力信号的正常传输,影响电力系统的稳定性和可靠性。基于此,设计了一种基于长短期神经网络和深度玻尔兹曼算法相融合改进长短期神经网络电力电缆故障诊断方法,将电力电缆上的电力信号作为长短期神经网络输入层的数据源进行实时采集和监控,利用神经网络的隐藏层对各类电力信号进行处理与识别,将发生故障时的异常数据进行精确提取,并作用于输出层进行表达,提升了对电力电缆故障识别的准确率。进一步引入深度玻尔兹曼算法,对隐藏层的故障数据识别因子进行调节,提升了长短期神经网络的收敛速度,改善了算法的收敛特性。最后,将该电力电缆故障诊断方法应用于电力信号监控平台中,并与传统的BP神经网络以及其他方法进行对比,改进长短期神经网络下的电力电缆故障信号识别率可达97.3%,故障信号诊断准确率可达96.61%,改善了传统方法的故障识别能力,为电力系统的安全运行提供重要保障。In the normal and safe operation of the power system,power cables are prone to faults caused by natural weather and human factors,which seriously disrupt the normal operation of power lines,reduce the normal transmission of power signals,and affect the stability and reliability of the power system.Based on this,an improved method for power cable fault diagnosis based on the fusion of long short-term neural networks and deep Boltzmann algorithm was designed.The power signal on the power cable was used as the data source for the input layer of the long short-term neural network for real-time collection and monitoring.The hidden layer of the neural network was used to process and identify various types of power signals,accurately extracting abnormal data when a fault occurred,and acting on the output layer for expression,improving the accuracy of power cable fault recognition.Furthermore,the deep Boltzmann algorithm was introduced to adjust the fault data recognition factor of the hidden layer,which improved the convergence speed of the long short-term neural network and the convergence characteristics of the algorithm.Finally,the power cable fault diagnosis method was applied to the power signal monitoring platform and compared with traditional BP neural networks and other methods.The improved long short-term neural network achieved a fault signal recognition rate of 97.3% and a fault signal diagnosis accuracy of 96.61%,which improved the fault recognition ability of traditional methods and provided important guarantees for the safe operation of the power system.

关 键 词：长短期神经网络 电力电缆 深度玻尔兹曼算法 故障诊断 

分 类 号：TU852[建筑科学]