基于深度学习理论与相速度的电缆故障在线诊断方法研究

发布时间：2018-01-06 19:30

本文关键词：基于深度学习理论与相速度的电缆故障在线诊断方法研究　出处：《西安科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着电力电缆的广泛使用,电缆故障的离线检测方法给电力部门带来了巨大的诊断压力,为了保证电网的安全运行同时降低人工维修成本,电缆故障的诊断方法应由离线检测诊断转变为在线检测。现阶段电缆故障的诊断仍以离线的方法为主,在线诊断方法多仍处于探索研究阶段,很多理论尚存在很多问题,导致在线诊断的方法的真实性和准确性欠佳难以运用到实际中,不能满足新形势下电缆故障诊断技术的新要求。针对以上提出的问题,本文建立一个地下电缆分布系统仿真模型用于采集不同情况下不同故障类型的电压和电流信号,引入深度学习的概念来分析电缆故障的类型,把故障等效的阻抗融入到电缆中根据电缆双端信号波形的相位差来确定故障距离。本文取得了一下的研究成果:(1)建立了一个拥有16根地下电力电缆分布的三相供电系统仿真模型,在不同位置电缆上设置了不同类型的故障,从理论上的角度模拟了电缆发生各种故障的情况,弥补了实际系统中故障可调性差和数据不足的缺点。(2)创建基于深度学习理论的深度信念网络(DBN)和卷积神经网络(CNN)用于电缆故障的识别。该深度神经网络利用大量的故障数据能够自动完成故障信号特征的分类并提取将故障准确地定位到具体电缆上并识别出故障类型。(3)针对行波测距方法的缺点,提出基于相速度的方法来获取故障距离,根据实际的电缆双端电压电流波形的相位差,推导出相位差与故障距离的数学表达式,所需要的数据易采集计算过程简单。(4)基于MATLAB-GUI设计了一个可视化的检测系统,将仿真模型、电缆故障识别、故障距离计算算法与波形显示功能集成到该系统里,故障设置方便简单、识别与计算结果直观形象。通过实验对比了基于深度学习的DBN和CNN与传统浅层神经网络对电缆故障的识别,DBN和CNN的故障平均识别正确率为89%、93%,传统的BP为50.8%,RBF为67%,SVM为83%。与传统的电缆诊断方法相比,本文所提出的方法利用海量的数据反映了电缆运行的状况和故障发生的规律,在故障类型识别正确率和故障定位精准性都有明显的提高,可作为电缆实际运行中故障诊断技术的有效补充,具有一定的理论意义和使用价值。
[Abstract]:With the wide use of power cables, off-line detection of cable faults has brought huge diagnostic pressure to the power sector, in order to ensure the safe operation of the grid and reduce the cost of manual maintenance. The method of cable fault diagnosis should be changed from off-line detection to on-line detection. At present, the main method of cable fault diagnosis is off-line, and most on-line diagnosis methods are still in the stage of exploration and research. There are still many problems in many theories, which leads to the lack of authenticity and accuracy of online diagnosis methods. Can not meet the new requirements of cable fault diagnosis technology under the new situation. In this paper, a simulation model of underground cable distribution system is established to collect voltage and current signals of different fault types under different conditions, and the concept of depth learning is introduced to analyze the types of cable faults. The equivalent impedance of the fault is integrated into the cable to determine the fault distance according to the phase difference of the signal waveform of the two ends of the cable. A simulation model of three-phase power supply system with 16 underground power cables is established. Different types of faults are set up on the cable in different positions, and the various faults of the cable are simulated from the angle of theory. It makes up for the shortcomings of poor fault tunability and insufficient data in the actual system. (2) Establishment of Deep belief Network (DBN) and Convolutional Neural Network (CNN) based on depth Learning Theory (DLT) and convolutional Neural Network (CNN). For cable fault identification, the depth neural network can automatically classify fault signal features by using a large number of fault data and extract fault accurately to locate the fault on a specific cable and identify the fault type. Aiming at the shortcomings of traveling wave ranging method. A method based on phase velocity is proposed to obtain the fault distance. According to the phase difference of the actual voltage and current waveform, the mathematical expression between the phase difference and the fault distance is derived. A visual detection system is designed based on MATLAB-GUI. The simulation model is used to identify the cable fault. The fault distance calculation algorithm and waveform display function are integrated into the system, and the fault setting is convenient and simple. The results of recognition and calculation are visualized and compared by experiments between DBN and CNN based on depth learning and traditional shallow neural networks for cable fault identification. The average correct rate of fault identification for DBN and CNN is 89 / 93.The traditional BP is 50.8 and 673SVM is 833.Compared with the traditional cable diagnosis method. The method presented in this paper reflects the status of cable operation and the rule of fault occurrence using massive data, and improves the accuracy of fault type identification and fault location obviously. It can be used as an effective supplement of fault diagnosis technology in practical operation of cable and has certain theoretical significance and practical value.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM75

【参考文献】