【作者】 陆宇航；

【导师】 杜伯学；

【作者基本信息】 天津大学 ， 高电压与绝缘技术， 2008， 博士

【摘要】 局部放电在线检测可以及时反映电力设备绝缘老化情况,是状态监测和故障诊断的重要手段。但是,由于现场强烈的电磁干扰严重影响传感器的灵敏度和检测的可靠性,高性能的传感器研制始终是局放在线检测研究中的一个重要课题。光学检测方法抗干扰能力强,绝缘性能好,灵敏度高,对于局放在线检测的研究意义重大。本文提出将根据Faraday磁光效应原理研制的光纤电流传感器应用于局放检测,以光信号为检测量,为局部放电光学检测方法奠定了基础。分析了不同类型光纤电流传感器的特点,利用琼斯矩阵法对光纤电流传感器的偏振光系统进行研究,给出了不同状态偏振光的归一化琼斯矢量,对本文涉及的光学器件推导了相应的Jones矩阵,最终建立了光纤电流传感器的数学模型。研制出有效带宽为300 MHz的全光纤结构光纤电流传感器,设计了相关的信号处理电路;采用双螺线管结构提高传感器灵敏度,螺线管形状的通电导体和光路之间的相对位置关系发生变化不会影响检测结果;分析了它的频率响应特性、抗干扰特性和光路损耗特性,与传统光纤电流传感器相比,该检测系统频率响应范围宽,灵敏度高。当传感器采用全光纤闭合结构光路设计并满足安培环路定律时,其抗电磁干扰性能最优。介绍了线性双折射特性和产生的各种原因;分析了线性双折射对于系统的影响,认为线性双折射是光纤电流传感器检测中影响系统测量误差的主要因素,随着线性双折射的增加,系统灵敏度下降,如果系统没有线性双折射,则其它误差源的影响将容易预测和处理,并探讨了消除线性双折射的可行措施。建立了局部放电实验系统及尖板放电、板板放电、内部放电、沿面放电和悬浮放电五种典型局放模型,通过本文研制的光纤电流传感器采集了各种局部放电信号,实验结果说明在强干扰环境下与其他方法相比较,基于光纤电流传感器的局部放电检测具有抗干扰能力强、绝缘特性好和响应速度快等特点。恰当地提取特征参数是进行局部放电进行模式识别的关键环节,将光纤电流传感器采集到的局部放电时域信号转换为三维时频谱图,全面表征了局放信号的时间分量、频率分量和放电能量的分布,提取的特征参数同时反映了原始信号的时频特征,各分量之间互不干扰,所以更加准确。本文采用基于非线性理论的分形盒子维数和空缺率描述复杂的时频表面,提取出局部放电三维时频谱图的时频特征参数,以描述时频谱图峰值处的陡度和下降沿的变化趋势,简化了特征向量的维数。基于时频分析方法的分形理论和BP神经网络可有效识别不同类型的局部放电,证明光纤电流传感器可用于局部放电检测和故障诊断。更多还原

【Abstract】 Partial discharge (PD) on-line detection can timely reflect the degradation degree of power apparatus, thus becoming a useful technique for condition monitoring and failure diagnosis. Because of high level electromagnetic interference, the development of high quality sensor is the key of PD on-line detection. Optical detection method has many advantages, such as great ability of anti-interference, excellent insulation property, high sensitivity, and is helpful to PD on-line detection.On basis of Faraday Effect, optical-fiber current sensor (OCS) is used for PD detection in the first time. It will possibly set a foundation for optical method applied into PD detection. The characteristics of different types OCS were analyzed and research into polarized light system of OCS using Jones matrix method, then the mathematics model of OCS was set up at last.Full-fiber OCS with effective band of 300 MHz and correlative circuit has been developed. Double constructer was adopted to enhance the sensitivity of system. Then its frequency-response property, anti-interference property and loss property have been analyzed in detail, from which the conclusion can be got that when the closed light path of OCS meet Ampere’s law, the sensor is supposed to have anti-interference feature.The cause of linear birefringence and its property was analyzed. The effects of all the factors on system error are related to the linear birefringence, and the larger the linear birefringence is, the larger the effects of the factors are. Methods to eliminate linear birefringence were discussed.PD test system and five models were set up. After picking classical PD signals with OCS, a comparison has been made between the optical method and other methods used in high level interference environment, and the result shows that PD detection using OCS has many merits such as great ability of anti-interference, high sensitivity and quick response.Suitable extraction of characteristic parameters is vital for PD pattern recognition. Single PD pulse acquired with OCS has been transformed into 3-D time-frequency image. The characteristics were represented with parameters of frequency, time and energy. It has been proposed that direct extraction of 3-D images may be used for PD pattern recognition. Compared with the methods for extraction of characteristic only in time domain or frequency domain, 3-D images can instantaneously reflect the magnitudes of different frequency components and their features in time domain, but there is no interference among different frequency components, thus it is more accurate than other methods.In order to extract the features of 3-D images, the fractal box-counting dimension and vacancy rate are adopted to characterize the complex time-frequency surface, which reflect the gradient at the magnitude and the variation trend at the trailing edge of time-frequency spectrum and can compress the dimensions of vector. The identification results using back propagation neural network (BPNN) and fractal theory based on time-frequency method demonstrates the effectiveness of OCS in PD detection.更多还原