【作者】 杨孝华；

【导师】 廖瑞金；

【作者基本信息】 重庆大学 ， 高电压与绝缘技术， 2002， 硕士

【摘要】 随着电力电缆的广泛应用和故障事故的增加，交联聚乙烯电力电缆的绝缘故障检测技术取得了长足的进展。本文在总结前有的XLPE电力电缆绝缘故障检测方法和借鉴变压器绝缘局部放电的基础上，以理论与实验相结合，研究了基于人工神经网络的以XLPE电力电缆局部放电信号的模式识别方法。本文介绍了交联电力电缆的发展状况，发生局部放电的原因，综合总结了现有的XLPE电力电缆故障检测方法，其他如变压器、互感器等电气设备的故障在线监测方法， 说明了研究XLPE电力电缆局部放电在线监测的理论和实践意义。本文的研究，将为实现XLPE电力电缆局部放电在线监测起到铺垫作用。论文分析了XLPE电力电缆局部放电模式识别原理，分析了局放信号特点；搭建了以脉冲电流传感器感应局放脉冲信号、话路滤波器去除干扰的实验平台；设计并实作了和实际运行中可能出现的局部放电特征相似的放电模型，分别测量了它们的局部放电信号，对实验后的电缆作了解剖，观察分析其放电痕迹。论文介绍了由特征量提取器和模式识别分类器两大模块构成的模式识别系统，阐述了人工神经网络的模式识别原理，本文一方面采用信号的PRPD模式的放电次数和统计算子作为BPNN的输入信号，设计了相应的BPNN模式识别程序；另一方面采用PRPD模式的统计算子作为SART人工神经网络的输入，设计了相应的SART神经网络模式识别程序。论文用设计的人工网络模式识别程序识别测量信号：以PRPD信号模式的放电次数为输入时，BP网络识别率为88％，识别率和发生局放的强弱程度有关，局放信号越强，识别率越高；以PRPD信号模式的统计算子为输入时，BP网络识别率为95％；以PRPD信号模式的统计算子为输入时，SART网络识别率为98％，识别率随着样本量的增加而提高。论文还根据识别结果，和实验现象，分析了出现误判的原因。更多还原

【Abstract】 With the widely application of XLPE cable and the increasing defaults, the detecting technology on XLPE cable insulation has got great improvement. Based on summarizing the present technology on XLPE faults detection and reference the PD detection method on power transformer, this paper introduces a method on recognition PD mode of XLPE. This paper uses discharging quantum and stat. arithmetic operators of chart as inputs to artificial neural network. This paper introduces the development processing of cable, the reason to occur PD. And the detection methods present such as detection on transformer, mutual inductance, are synthesized. Demonstrate the importance of researching on XLPE cable PD measurement.The character of XLPE PD signal is analyzed, and measurement method is designed. Pulse current sensor is used to acquire PD signal by inducting pulse signal, and band-pass filter to exclude interfere. Several discharging models simulation are designed. Their PD signals are measured and measured cables are anatomized to observe their discharging trace.Implement of picking-up character and mode classifying are introduced, and expatiate the principle of artificial neural networks. Soft program is necessary to pick-up the character of PD signals and classify PD mode. Program to pick-up the character of PD signals is compiled. Using the discharging times and stat. operator on PRPD mode of PD signal, mode-classifying implements based on BPNN and SART are compiled.The artificial neural networks designed are applied to recognize measured signals. Conclusions are made: recognition rate of BPNN is 95%, and it is respect to magnitude of discharging signal and increases while signal magnitude increasing; The recognition rate of SART neural network is 98%, higher than that of BPNN, recognition rate increasing with the number of signal group. Then analyze the reason to fault recognition according to the recognition result.更多还原