作者：祁宏昌¹, 张斌², 黄嘉盛¹, 刘远², 吴倩¹, 洪晓斌²

作者单位：1. 广州供电局输电管理所, 广东 广州 510310;
2. 华南理工大学机械与汽车工程学院, 广东 广州 510641

关键词：Lamb波;高压电缆;短时傅里叶变换;液位检测;深度置信网络

摘要：

瓷套式高压电缆终端内部液态介质关乎高压电缆的运行安全，其内液位定时检测可以有效排除因介质泄露引起的安全隐患。该文提出一种基于STFT和DBN的瓷套式高压电缆终端液位智能检测方法。首先，对超声兰姆波信号进行信号分割，并通过短时傅里叶变换获取分段信号的时频表示；然后，提取分段时频信号的有效值、峰峰值、峭度和波形因子等统计特征；最后将提取的特征作为深度置信网络模型的输入，并将液位分为15个区间作为网络输出。实验训练迭代5 000次时，所提方法的液位识别准确率为92%；当训练迭代10 000次时，测试正确率达到100%。实验结果表明，该方法可以准确地识别液位的高度范围，并提供一定的维护指导。

Liquid level intelligent detection of high voltage cable porcelain termination using short-time Fourier transform and deep brief networks
QI Hongchang¹, ZHANG Bin², HUANG Jiasheng¹, LIU Yuan², WU Qian¹, HONG Xiaobin²
1. Transmission Management Institute of Guangzhou Power Supply Bureau, Guangzhou 510310, China;
2. School of Mechanical and Automotive Engineering, South China University of Technology, Guangzhou 510641, China
Abstract: Internal silicone oil of porcelain bushing type (PBT) high voltage cable terminal is about the safety operation, and its internal liquid level detection can effectively prevent the security risks caused by oil leaks. To solve above problems, an intelligent detection method based on STFT and DBN is proposed to detect the internal liquid level of PBT terminals. First, the ultrasonic lamb wave signals are divided into many segments, and these segments are processed by short-time Fourier transform (STFT) to obtain time-frequency features, including root mean square, peak-peak value, kurtosis and shape factor. Then, extracted features are input into deep brief network (DBN) model for training, and the liquid level was divided into 15 categories as network output. The liquid level recognition accuracy of proposed method is 92% when training 5 000 iterations, only a handful of states appear adjacent miscalculation. The test accuracy reached 100% when training 10 000 iterations. Experimental results show that proposed method can accurately identify the height range of the liquid level, and provides a maintenance guidance.
Keywords: Lamb wave;high voltage cable;short-time Fourier transform;liquid level detection;deep belief networks
2019, 45(4):47-52  收稿日期: 2018-11-20;收到修改稿日期: 2018-12-23
基金项目: 中国南方电网项目（GZM2015-1-0011）
作者简介: 祁宏昌(1985-),男,陕西宝鸡市人,硕士,研究方向为电力系统及其自动化
参考文献
[1] 程明, 马崇, 陈韶瑜, 等. 基于超声波的变电站充油瓷套油位检测方法[J]. 河北电力技术, 2014, 33(6):41-42
[2] 莫润阳, 牛海清, 郭然, 等. 瓷套式电缆终端油位的超声检测[J]. 西北大学学报(自然科学版), 2015, 45(5):745-748
[3] 何存富, 怀保玲, 杜婷, 等. 基于兰姆波的大型罐体液位定点检测方法[J]. 机械工程学报, 2007, 43(6):99-104
[4] 徐鸿, 郭鹏, 田振华, 等. 非浸入式超声导波液位测量方法研究[J]. 仪器仪表学报, 2017, 38(5):1150-1158
[5] 艾春安, 蔡笑风, 刘继方, 等. 短时傅里叶变换声-超声检测信号缺陷识别[J]. 中国测试, 2015(4):29-31
[6] 刘瑾, 雷毅. 小波分析在超声检测信号处理中的应用[J]. 电焊机, 2010, 40(7):77-80
[7] 辜清. 基于PCA和BP神经网络的内检测技术在油气集输管道中的应用[J]. 油气田地面工程, 2018, 37(4):48-52
[8] 任欢. 基于GA-BP神经网络的埋地管道土壤腐蚀性评价[J]. 管道技术与设备, 2018(3):41-44
[9] 吴坚, 郑照红, 薛家祥. 深度置信网络光伏发电短时功率预测研究[J]. 中国测试, 2018, 44(5):6-11
[10] 姚志强. 基于深度置信网络的管网泄漏故障诊断方法研究[J]. 中国安全生产科学技术, 2018(4):101-106
[11] HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786):504-507
[12] FISCHER A, IGEL C. An introduction to restricted Boltzmann machines[C]//Iberoamerican Congress on Pattern Recognition. Berlin:Springer, 2012.
[13] 李巍华, 单外平, 曾雪琼. 基于深度信念网络的轴承故障分类识别[J]. 振动工程学报, 2016, 29(2):340-347