融合注意力机制的输电线路故障概率预测模型

输电线路运行环境恶劣,发生故障的概率和气象条件直接相关.由于目前极端气象条件在全世界范围内表现活跃,研究不同气象条件下输电线路发生故障的概率对提高电网运行稳定水平有重要意义.该文从分析气象相关输电线路典型故障的作用机理和统计特征出发,提出使用融合注意力机制的深度神经网络进行输电线路故障概率预测,使用停电数据对模型进行检...     

基于径向基网络的同杆双回路故障类型判定

由于同杆双回线发生故障时两回线之间存在非线性耦合 ,故用传统方法无法对同杆双回线进行故障判定 ,而径向基神经网络可以通过故障线路相关电压和电流对线路故障进行准确分类 ,网络输入样本在不同故障发生地点、发生时间、过渡电阻情况下得到。仿真结果表明该网络能够进行同杆双回路故障判定。     

改进BP神经网络的SVM变压器故障诊断

油中溶解气体是变压器故障诊断的重要依据,为了融合以及扩充变压器油中溶解气体含量的特征信息,提高变压器故障诊断准确率,本文提出了改进BP神经网络的SVM（Support Vector Machine）变压器故障诊断方法。首先,通过改进的BP神经网络将5维的气体特征信息进行融合并扩充到128维;然后,在改进的BP神经网络中使用每层提取的特征向量作为SVM的输入对变压器故障进行诊断,增加改进的BP神经网络中诊断准确率较高的特征向量的权重;最后,选择累积权重最大的特征向量作为输入,使用SVM进行变压器的故障诊断。该方法经过多层神经网络的映射使提取的气体特征信息融合及扩充后具有更加明显的特征区别,从而可以有效的提高SVM的诊断准确率。实验结果表明,本文所提出的算法与BP神经网络和SVM的变压器故障诊断方法相比诊断准确率有较大的提升。同时,随着训练数据样本的增加,模型的诊断准确率具有一定的提升。     

基于脑启发视觉神经元网络输电线路部件识别的研究

输电线路关键部件的检测对电力系统的设备安全及系统稳定运行起着关键作用。基于机器视觉和深度学习的输电塔线巡检技术检测耗能大,花费时间长,难以满足高效低耗与实时检测的要求。为了完成对输电线路部的快速准确识别,本文将生物视觉理论中特征提取计算模型和更为高效的脉冲神经网络相结合,提出了一种基于脑启发的多层神经元网络模型以模拟视觉感知的流程。在频率编码、突触电导和侧抑制连接等多种生物可信神经机制下,通过无监督脉冲放电时间依赖可塑性规则进行特征学习,并基于有监督规则进行识别和分类,验证了该模型在输电线路部件识别上的良好性能,并基于现场可编程门阵列芯片实现了所提出的神经元网络模型,为之后应用于输电线路故障的在线智能化检测提供了一种可能途径。     

基于多通道卷积双向长短时记忆网络的输电线故障分类

针对单通道故障分类器不能全面表达三相故障特征信息引起分类精度不高的问题,提出了一种基于多通道卷积双向长短时记忆神经网络(MCCNN-BiLSTM)的输电线故障分类方法.该方法可同时输入故障三相信号,并能有效提取故障信号的空间和时间特征,实现了三相故障信号特征的全面提取,有效地提高了神经网络的分类的精度.基于735 kV...     

Signal analysis‐based fault classification and estimation of fault location of an unbalanced network using S‐transform and neural network

This paper demonstrates a technique for the diagnosis of the type of fault and the faulty phase on an overhead transmission line, followed by locating the particular fault on the affected phase. The power system network considered in this study is a three‐phase transmission line with unbalanced loading simulated in the PowerSim Toolbox of MATLAB. S‐transform is used to compute the energy components of the voltage signals of the three phases of the transmission line. These features are used as input vectors of a probabilistic neural network (PNN) for fault detection and classification. Detection of the faulty phase(s) is followed by estimation of fault location. The voltage signal of the affected phase is processed to generate the S‐matrix. The frequency components of the S‐matrices for different fault locations are used as input vectors for training a backpropagation neural network (BPNN). The results are obtained with satisfactory accuracy and speed. All the simulations have been done in MATLAB environment for different values of fault locations, fault resistances, and fault inception angles. The effect of noise on the simulated voltage signals has been investigated. The analysis has been further extended by implementing the proposed method in a modified version of IEEJ West 10 machine system model. © 2016 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

基于模糊神经网络的故障类型识别

提出了基于模糊神经网络的双端电源输电线路故障类型识别的方法,用ATP提取输电线路故障后一周后继电保护安装点的三相电压电流以及反映接地故障的零序电流基频分量及其相应的相角,并采用T-S模型与改进BP算法结合的模糊神经网络,实现故障类型识别。该方法不受故障位置、故障电阻及对两端电源初始相角差、系统运行方式等不确定的因素影响,仿真结果表明该类型识别方法可靠、正确。     

Improved fault analysis technique for protection of Thyristor controlled series compensated transmission line

Transmission lines are major component of a power system. Any fault on them results in outage of power not only in the area fed by them but also in the neighboring area as well. Therefore, protection of them is very important. Nowadays, in order to allow maximum power transfer series compensation both uncontrolled and controlled are used. Due to the introduction of compensating devices the protection methodology of transmission lines requires changes. A new transmission line fault analysis method based on half cycle post fault three-phase current data has been presented in this paper for series compensated transmission line equipped with Thyristor Controlled Series Compensator (TCSC). The proposed two-step methodology has been developed with the help of Discrete Wavelet Transform (DWT) and implementation of Chebyshev Neural Network (ChNN). ChNN is derived from regular neural network, but is functionally superior. The performance of the developed algorithm has been tested over a vast fault pattern data set dynamically generated with EMTDC/PSCAD. The results with extensive testing indicate effectiveness of the developed scheme with higher level of accuracy and speed. The algorithm is capable of doing classification with minimal training.     

Induction machine fault detection using stray flux EMF measurement and neural network-based decision

The aim of this paper is to present the performances of voltage unbalance and rotor fault detections using an external stray flux sensor in a working three-phase induction machine. The automatic classification and fault severity degree evaluation are realized by using a neural network approach based on a multi-layer perceptron (MLP) structure. In this paper, it is proved that a simple external stray flux sensor is more efficient than the classical stator current sensor to detect rotor broken bar and voltage unbalance, using data processing at low-frequency resolution.     

基于人工鱼群神经网络的抽油机故障诊断

BP神经网络应用于抽油机的故障诊断时易陷入局部极值,同时收敛速度也无法保证。在此前提下,提出人工鱼群神经网络算法的抽油机故障诊断新方法,充分利用人工鱼群在全局范围的快速寻优特性以克服BP神经网络收敛速度较慢和易陷入局部最优解的缺点,从而提高故障诊断的准确率和速度。以抽油机的管漏、供液不足、杆断脱、泵漏失、气影响五种故障类型为例,利用MATLAB分别搭建了传统BP神经网络和人工鱼群神经网络的模型,并对两种方法的诊断结果进行了比较。仿真结果充分说明了人工鱼群神经网络在抽油机故障诊断中的可行性、准确性和优越性。     

基于蚁群算法的神经网络配电网故障选线方法

为了克服基于神经网络的故障选线方法收敛速度慢、易于陷入局部极小点的缺点,提出了蚁群算法和神经网络相结合的故障选线方法。利用ATP-EMTP做单相接地仿真试验,得到各线路的零序电流信号,通过小波变换和傅里叶变换提取其中的故障特征作为神经网络的输入。利用蚁群算法对神经网络进行训练,完成训练的神经网络模型即可实现故障选线。仿真结果表明,该方法训练速度快、误判率低。     

基于粗集神经网络的配电网故障选线研究

为了克服基于神经网络的故障选线方法训练时间长和网络结构复杂的缺点,提出了基于粗集神经网络的故障选线方法.利用ATP-EMTP做大量的单相接地故障仿真试验,得到大量的各馈线零序电流信号,通过小波变换和傅立叶变换从中提取各种暂态和稳态故障特征.利用粗集理论对故障特征进行预处理,将约简后的故障特征作为神经网络的输入,约简后的样本作为训练样本.完成训练的神经网络模型即可实现故障选线.仿真和现场验证结果表明,该方法训练速度快、误判率低.     

支持向量机及在电力系统中的应用

支持向量机是20世纪60年代开始研究,在90年代得到快速发展的机器学习技术。为了系统地归纳统计学习理论与支持向量机的基本思想和算法,总结目前该领域的最新研究成果。通过对7种多分类支持向量机训练算法进行深入分析,得出其各算法的优、缺点,还归纳了支持向量机在故障预测和识别、电力系统等方面的应用,特别在电力系统暂态稳定评估与分析、电机故障诊断、高压输电线路故障诊断和定位、双凸极永磁发电机非线性模型、火焰监测以及电力系统负荷预测等方面的成功应用。研究表明,支持向量机克服了传统神经网络算法的局部最优、收敛难以控制、结构设计困难等优点。     

一种基于神经网络的高压输电线故障分类器

故障分类器是输电线路保护中的基本模块。文中以一个实际500 kV输电网络为模型,提出了基于BP前馈网络 和Kohonen自组织特征映射网络的高压输电线路故障分类方法,并进行了仿真研究。结果表明：这种方法快 速、可靠,对输入信号容错性强,尤其在高阻接地故障时具有较好的分类性能,可以用于支持高速保护装 置。     

Study of a new method for power system transients classification based on wavelet entropy and neural network

The detection and classification of transient signals are widely applied in many fields of power system. The study of transient signal detection and classification is a sustaining focus of researchers as well as a difficult issue. There are still many problems needed to be solved in this area. Based on the wavelet transform (WT), the idea of entropy and weight coefficient is introduced, and the wavelet energy entropy (WEE) and wavelet entropy weight (WEW) are defined in this paper. The distribution picture of WEE and WEW along with scales are presented for the first time. PSCAD/EMTDC models for six types of transients, namely breaker switching, capacitor switching, short circuit fault, primary arc, lightning disturbance and lightning strike fault, are constructed. With WEE and WEW, the eigenvectors for the six transients are established and a model which uses the eigenvectors as the input of the BP (back-propagation) neural network is set up to realize the classification of these transients. The simulation has been executed based on a 500 kV transmission line model in China and the results show that feature extraction based on WEE and WEW can effectively discover the useful local features. With the help of neural network classifier, it has effective classifying result. This method is applicable in the power system.     

基于深度学习的特高压直流闭锁故障智能调度决策

针对特高压直流闭锁故障的处置策略问题,提出一种基于深度学习的故障特征建模方法及故障后电网调度策略生成方法,所提智能调控决策依据电网直流故障特征和运行环境信息,通过大数据驱动模型训练得到故障后的调度策略。首先根据故障环境信息,利用故障影响相关性提取有效故障信息,构建故障特征模型。然后介绍深度学习类神经网络原理和多层感知器模型,提出利用深度网络提取训练故障前后运行特征,自动生成调控策略的思路。之后利用反向传播算法构建深度学习框架,通过不断计算损失函数和准确率修正训练模型,自动生成有效故障处置策略。最后利用锦苏直流特高压线路相关的电力系统验证了所提方法的有效性。     

Wavelet Packet Entropy and RBFNN Based Fault Detection,Classification and Localization on HVAC Transmission Line

The article presents a technique for fast and accurate detection, classification and localization of faults on the high voltage transmission systems considering the alternator's dynamics and the effect of transformers. The systems have been simulated by ATP/EMTP software and three phase fault currents at one end of the transmission line are recorded with a sampling frequency of 50 kHz. The fault signals are decomposed by wavelet packet decomposition (WPD) up to 3^rd level with mother wavelet db6 to calculate wavelet packet entropy (WPE) which has the ability to measure the uncertainty of fault signals during feature extraction. A properly designed radial basis function neural network (RBFNN) trained with these features can recognize, classify and locate faults faster as it utilizes only half cycle data after fault initiation. This technique has been verified for different fault categories, fault impedances and fault inception angles (FIA) at different locations for two different transmission systems. The investigated results demonstrate that the wavelet packet entropy is very powerful for extracting the features from the fault signals and RBFNN is very accurate for classification and localization of faults on the transmission line including locations close to the generator's end.     

遗传算法在前向神经网络参数估计中的应用

人工神经网络在很多领域有着成功的应用,神经网络参数估计有许多训练算法,BP算法是前向多层神经网络的典型算法,但BP算法有时会陷入局部最小解,遗传算法是一种随机优化技术,它可以发现全局优解。本文介绍了遗传算法在前向多层神经网络参数估计中的应用,并对标准遗传算法进行了适当的改进。结合具体例子给出了算法实现的操作步骤和实验结果。实验数据表明采用遗传算法得到的神经网络参数是最优的,神经网络的性能优于基于BP算法的神经网络性能。     

基于果蝇优化算法的GRNN电网故障诊断

提出基于果蝇优化算法的GRNN电网故障诊断模型,实现GRNN分布参数的优化选择。利用广义回归神经网络(GRNN)相比于其他人工神经网络在逼近能力、分类能力和学习速度上面的优势,建立基于GRNN神经网络的电网故障诊断模型。经分析及测试,该方法能够有效的提高运行人员故障处理效率,快速并准确的实现电网的故障诊断。     

基于二维卷积神经网络的BLDCM驱动系统故障检测方法

针对无刷直流电机驱动系统功率开关管故障诊断存在由于特征提取效果差而导致识别准确率低等问题,提出一种基于二维卷积神经网络2D-CNN(two-dimensional convolution neural network)自适应特征提取的故障检测方法,避免人工提取特征的复杂性及不确定性.以相电流作为故障敏感信号进行FFT预...