WO2020015277A1 - 一种基于全景信息的弧光故障识别装置及方法 - Google Patents

一种基于全景信息的弧光故障识别装置及方法 Download PDF

Info

Publication number
WO2020015277A1
WO2020015277A1 PCT/CN2018/119211 CN2018119211W WO2020015277A1 WO 2020015277 A1 WO2020015277 A1 WO 2020015277A1 CN 2018119211 W CN2018119211 W CN 2018119211W WO 2020015277 A1 WO2020015277 A1 WO 2020015277A1
Authority
WO
WIPO (PCT)
Prior art keywords
global
information
kalman filter
local
sensor
Prior art date
Application number
PCT/CN2018/119211
Other languages
English (en)
French (fr)
Inventor
陈力
周斌
沈健
彭奇
周绍亮
张敏
沈宇龙
王猛
Original Assignee
国电南瑞科技股份有限公司
国电南瑞南京控制系统有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 国电南瑞科技股份有限公司, 国电南瑞南京控制系统有限公司 filed Critical 国电南瑞科技股份有限公司
Publication of WO2020015277A1 publication Critical patent/WO2020015277A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
    • G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate

Definitions

  • the invention belongs to the technical field of electric power automation, and particularly relates to an arc light fault identification device and method based on panoramic information.
  • the arc short-circuit fault in the switch cabinet is a serious power distribution system failure.
  • the arc effect caused by the internal arc combustion release can burn expensive switching equipment in severe cases, and the short-circuit current impact generated at the same time can damage the main transformer and cause long-term Power outages can cause more serious casualties to nearby personnel.
  • the fault arc detection technology mostly uses sensors to detect arc light generated when a fault arc is burned, and compares it with a preset threshold value to determine whether a fault arc has occurred.
  • the working environment of the switchgear is complicated, and the external noise interference is large.
  • the physical characteristics generated when the fault arc burns are very uncertain. Therefore, when a single sensor is used to detect the fault arc, the conflict between detection sensitivity and accuracy It is difficult to solve, resulting in a high false detection rate, which seriously affects the safe and stable operation of the power system.
  • the embodiment of the present invention is expected to provide an arc light fault recognition algorithm based on panoramic information, which uses a plurality of different types of sensors to detect a variety of physical characteristics during arc burning, and simultaneously proposes an error that captures arc current characteristics Back Propagation (BP, Error, Propagation, Training) neural network algorithm, which combines the sensor information and arc current information to determine the arc protection action, and performs Kalman filter filtering, thereby reducing the probability of misoperation and refusal and effectively improving the protection reliability.
  • BP Back Propagation
  • Propagation Training
  • An embodiment of the present invention provides an arc light fault identification device based on panoramic information, including a sensor unit, a local Kalman filter, an information fusion unit, a global Kalman filter, and an arc light recognition unit;
  • the sensor unit, the local Kalman filter, the information fusion unit, the global Kalman filter, and the arc recognition unit are sequentially connected in sequence;
  • the sensor unit includes several sensors.
  • the sensor unit is configured to detect the electrical quantity when an arc occurs.
  • Each sensor is connected to a local Kalman filter to filter the sensor output data.
  • the global filtering of the panorama information is performed based on the global Kalman filter to perform arc light judgment.
  • the sensor unit includes an arc light sensor, a temperature sensor, and a pressure sensor.
  • the information fusion unit includes a local information fusion module and a global information fusion module;
  • the local information fusion module fuses information output by similar sensors
  • the global information fusion module fuses the panoramic information output by all local information fusion modules.
  • the panoramic information is characteristic information of arc light detection for the spectral characteristics of solitary light, including current characteristics and acousto-optic characteristics.
  • An arc light fault identification method based on panoramic information includes the following steps:
  • the sensor unit collects electrical quantity information when an arc occurs.
  • Step S2 includes the following steps:
  • the sensor unit is connected with a local Kalman filter of multi-parameter sensor data, analyzes the characteristic changes of the current in the time domain and the frequency domain when the arc fault occurs, extracts the characteristic values as the input of the BP neural network, and uses the BP neural network to detect the arc Fault, use genetic algorithm to get the initial weight of BP neural network to realize circuit breaker arc recognition;
  • neurons in the input layer including the maximum and minimum amplitude difference I diff of the current in the sampling time window, the energy of the current in the low and middle frequency bands, and the first node of the wavelet packet decomposition tree in the low and middle frequency bands Node (4, 1), Root Mean Square (RMS) of the second node Node (4, 2), the third node Node (4, 3), and the fourth node Node (4, 4);
  • the number of neurons in the output layer is one;
  • H the number of neurons in the hidden layer of the neural network
  • I represents the number of neurons in the input layer of the neural network
  • O represents the number of neurons in the output layer of the neural network, ⁇ is the elastic momentum, and the value ranges from 1 to 10;
  • Extract 6 feature quantities from the original current sampling data to form N sample data as training samples.
  • the conditions for stopping the neural network training are Epoch ⁇ 6000, output error MSE ⁇ 0.001, or gradient descent Gradi-ent ⁇ 1.00e -10 ;
  • X 0 is the initial value of the global state
  • P 0 is the initial value of the global covariance matrix
  • Q 0 is the initial value of the global system covariance matrix
  • the information distribution factors of the global Kalman filter and the local Kalman filter are according to the sensor detection characteristics. set up;
  • the initial value of the P 0 global covariance matrix and the initial value of the Q 0 global system covariance matrix have no dispersion, and take 0 or a zero drift error;
  • P m is the global covariance matrix and Q m is the global system covariance matrix, where m is the number of sensors, the global Kalman filter, and the information distribution factor for each local Kalman filter. Allocation based on formula (2) and formula (3);
  • Formula (2) is a global Kalman filter information factor allocation formula, where Q i is the global system covariance matrix of discretely sampled data at the i-th point of the sensor, and Q is the global Kalman filter allocation factor;
  • Formula (3) is the formula for the local Kalman filter information factor allocation, where P i is the global covariance matrix of discrete sampled data values at the i-th point of the sensor, P is the local Kalman filter allocation factor; ⁇ i is the weight factor;
  • Equation (4) is applicable to the same type of sensors, where P i is the global covariance matrix of discrete sampling data at the i-th point of the sensor, Point i is the discrete time samples state estimate vector; D k is the k th sampling data sensor, the Z i value for the i th observation point, Is the transpose of the i-th data, R i is the residual of the i-th data;
  • Q all i is the global system covariance matrix of the ith group of data after local fusion of various types of sensors. Is the state estimation vector of discrete time sampling of the i-th group of data after local fusion of various types of sensors; D k ′ represents the data of the k-th sensor after local fusion, and Z ′ i is the observation value of the i-th point after local fusion of the data, H ′ T i is the transpose of the i-th point data after local data fusion, and R ′ i is the residual of the i-th point data after local data fusion;
  • the value of ⁇ satisfies the output error (MSE) of the neural network ⁇ 0.001 or the gradient descent ⁇ 1.00e -10 ;
  • This application discloses a method for identifying arc faults based on panoramic information, a BP neural network algorithm that captures the characteristics of arc currents, overcomes the problems of inflexibility and instability of analog filters, and is based on BP neural network models, Kalman filtering, and The fusion of panoramic information shows strong robustness due to the influence of temperature changes and circuit noise interference changes, which greatly reduces the probability of arc protection misoperation.
  • the present application discloses an arc light fault identification device based on panoramic information, and a panoramic information filter constructed by multiple types of sensors, which performs real-time filtering on a plurality of different fault arc fault characteristics, which can well eliminate noise interference and improve reliability.
  • the noise ratio improves the accuracy and reliability of fault arc detection, and improves the fault tolerance and resilience of the system. Effectively improve arc detection capabilities.
  • the probability of correct operation of the arc fault is above 99.9%.
  • FIG. 1 is a schematic diagram of an arc light fault identification device based on panoramic information according to an embodiment of the present invention
  • FIG. 2 is a schematic structural diagram of a BP neural network according to an embodiment of the present invention.
  • FIG. 3 is a filter architecture diagram provided by an embodiment of the present invention.
  • an arc light fault identification device based on panoramic information includes a sensor unit, a local Kalman filter, an information fusion unit, a global Kalman filter, and an arc light recognition unit;
  • the sensor unit, the local Kalman filter, the information fusion unit, the global Kalman filter, and the arc recognition unit are sequentially connected in sequence;
  • the sensor unit includes several sensors.
  • the sensor unit is configured to detect the electrical quantity when arcing occurs.
  • Each sensor is connected to a local Kalman filter.
  • the local Kalman filter is based on filtering algorithms of different sensors to filter the sensor output data.
  • the global filtering of the panorama information is performed based on the global Kalman filter to perform arc light judgment.
  • the sensor unit includes an arc light sensor, a temperature sensor, and a pressure sensor.
  • an arc light sensor When an arc light occurs, light is the most direct phenomenon, but light will be blocked and will be interfered by visible light pollution. Therefore, a variety of sensors are required to cooperate with each other.
  • the energy density of the arc light is large, and the temperature near the arc will rise rapidly, and the temperature rise will increase the gas pressure. Therefore, these sensors are used to detect the electrical quantity when arcing occurs.
  • Each sensor has unique waveform data, such as current and pressure. Different information needs to be captured, so the filtering algorithms for different sensors are different, and each sensor is connected to a local Kalman filter. Because the weight of each type of information is different, for example, the most important arc sensor should occupy the main position, so a variety of sensors will be multiplied by weight coefficients for data information fusion according to different installation positions and types. To further filter out abnormal data, a global filter for panoramic information is needed.
  • the Kalman filter is a filter for identifying jitter data, that is, filtering fluctuation data. These fluctuation data will cause no movement.
  • a reasonable covariance value (dispersion) can be obtained. This is reasonable.
  • the degree of dispersion defines whether arcing really occurs or is disturbing.
  • a set of data [10,10.1,10,9.9 ...]
  • This set of data has a very low dispersion. Within the range of the dispersion threshold, it is very stable, no interference, and no arcing occurs, but if it is [10,10.1,90,99,99.1,100,100 . «] Such a dispersion is higher than the dispersion threshold, but since the subsequent data is stable, it will not be too high, and this is where true arcing occurs.
  • this application uses different types of sensors, but these sensors have different data characteristics due to the different objects collected, and different local Kalman filters need to be designed for different sensors, that is, for different sensors.
  • Sensor select a dispersion corresponding to a threshold value.
  • Each type of sensor has an information distribution principle, that is, it has different weights. If the value is 100, the arc is considered to be 80, but the weight of pressure such as sound is relatively small. It plays the role of auxiliary reference, that is, even if the sound sensor passes its Kalman filtering, it outputs a full-code value. From the perspective of the global filter, it is 10 points, while the data of the arc sensor accounts for the proportion It will be big. The values of the data fusion of these different sensors are also fluctuating, and there is also the risk of false triggering.
  • the global Kalman filter is used to filter out the fluctuations of data fusion of different sensors.
  • the local information fusion module fuses information output by similar sensors
  • the global information fusion module fuses the panoramic information output by all local information fusion modules.
  • the panoramic information is characteristic information of arc light detection for the spectral characteristics of solitary light, including current characteristics and acousto-optic characteristics.
  • FIG. 2 is a schematic structural diagram of a BP neural network model according to an embodiment of the present invention.
  • the parameters include the number of input neurons, the number of network layers, the number of hidden layer neurons, and the number of output neurons. Set according to the actual situation.
  • the sensor unit collects electrical quantity information when an arc occurs.
  • Step S2 includes the following steps:
  • the sensor unit is connected with a local Kalman filter of multi-parameter sensor data, analyzes the characteristic changes of the current in the time domain and the frequency domain when the arc fault occurs, extracts the characteristic values as the input of the BP neural network, and uses the BP neural network to detect the arc Fault, aiming at the shortcomings of slow convergence of BP neural network, using genetic algorithm to obtain the initial weight of BP neural network, speeding up the training convergence speed of neural network, and realizing circuit breaker arc recognition;
  • neurons in the input layer including the maximum and minimum amplitude difference I diff of the current in the sampling time window, the energy of the current in the low and middle frequency band (less than 100KHz), and the first node of the wavelet packet decomposition tree in the low and middle frequency band Node (4 , 1), the average root mean square value (RMS) of the second node Node (4, 2), the third node Node (4, 3), and the fourth node Node (4, 4);
  • the number of neurons in the output layer is one, and the output detection is based on arc recognition, so only one output neuron is required to represent "0" or "1".
  • H the number of neurons in the hidden layer of the neural network
  • I represents the number of neurons in the input layer of the neural network
  • O represents the number of neurons in the output layer of the neural network
  • ⁇ is the elastic momentum
  • the value ranges from 1 to 10.
  • the value of ⁇ satisfies the output error (MSE) of the neural network ⁇ 0.001 or the gradient descent ⁇ 1.00e -10 .
  • Extract 6 feature quantities from the original current sampling data to form N sample data as training samples (including N / 2 arc fault samples and N / 2 normal case samples).
  • the stop condition of the neural network training is the number of training iterations Epoch ⁇ 6000 Output error MSE ⁇ 0.001 or gradient descent Gradi-ent ⁇ 1.00e -10 ;
  • X 0 is the initial value of the global state
  • P 0 is the initial value of the global covariance matrix
  • Q 0 is the initial value of the global system covariance matrix
  • the information distribution factors of the global Kalman filter and the local Kalman filter are according to the sensor detection characteristics. Setting;
  • local Kalman filters such as the initial value of the temperature sensor filter, it can be set to 50 degrees Celsius according to the internal temperature conditions of the switchgear, the light intensity can be set according to natural light, and the initial value of the pressure sensor can be set to 0. ;
  • the initial value of the global covariance matrix of P 0 and the initial value of the global covariance matrix of Q 0 have no dispersion. Take 0 or a zero drift error;
  • a local Kalman filter such as the initial value of a temperature sensor filter
  • the light intensity is set according to natural light
  • the initial value of the pressure sensor can be set to 0.
  • P m is the global covariance matrix and Q m is the global system covariance matrix, where m is the number of sensors, the global Kalman filter, and the information distribution factor for each local Kalman filter. Allocation based on formula (2) and formula (3);
  • Formula (2) is a global Kalman filter information factor allocation formula, where Q i is the global system covariance matrix of discretely sampled data at the i-th point of the sensor, and Q is the global Kalman filter allocation factor;
  • Formula (3) is the formula for the local Kalman filter information factor allocation, where P i is the global covariance matrix of discrete sampled data values at the i-th point of the sensor, P is the local Kalman filter allocation factor; ⁇ i is the weight factor;
  • Equation (4) is applicable to the same type of sensor, such as an arc sensor or a temperature sensor.
  • P i is the global covariance matrix of the discrete sampling data at the i-th point of the sensor
  • Point i is the discrete time samples state estimate vector
  • D k is the k th sampling data sensor
  • the Z i value for the i th observation point Is the transpose of the i-th data
  • R i is the residual of the i-th data
  • Q all i is the global system covariance matrix of the ith group of data after local fusion of various types of sensors. Is the state estimation vector of discrete time sampling of the i-th group of data after local fusion of various types of sensors; D k ′ represents the data of the k-th sensor after local fusion, and Z ′ i is the observation value of the i-th point after local fusion of the data, H ′ T i is the transpose of the i-th point data after local data fusion, and R ′ i is the residual of the i-th point data after local data fusion.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Testing Or Calibration Of Command Recording Devices (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

一种基于全景信息的弧光故障识别装方法，实现断路器弧光识别，属于电力自动化技术领域，通过分析电弧故障发生时电流在时域、频域的特征变化，提取合适的特征值作为误差反向传播(BP)神经网络的输入，采用BP神经网络来检测电弧故障，针对BP神经网络收敛慢的缺点，采用遗传算法获得BP神经网络的初始权值，加快神经网络的训练收敛速度；该方法的输出信噪比高，从而提高了故障电弧检测的准确性和可靠性。

Description

一种基于全景信息的弧光故障识别装置及方法
相关申请的交叉引用
本申请基于申请号为201810802425.2、申请日为2018年07月20日的中国专利申请提出，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此引入本申请作为参考。
技术领域
本发明属于电力自动化技术领域，尤其涉及一种基于全景信息的弧光故障识别装置及方法。
背景技术
随着配电网容量的逐渐增大,中低压母线故障对电力系统安全运行的影响越来越大。开关柜内弧光短路故障是一种严重的配电系统故障，其内部电弧燃烧释放所产生的电弧效应，严重时可烧毁昂贵的开关设备，同时产生的短路电流冲击可损坏主变压器,造成长时间停电，更严重的可造成附近人员的伤亡事故。
相关技术中，故障电弧检测技术多采用传感器对故障电弧燃烧时产生的弧光进行检测，并与预设的阈值进行比较，进而判断是否有故障电弧产生。而开关柜工作环境复杂、外界噪声干扰较大，再加上故障电弧燃烧时产生的物理特征具有很大的不确定性，因此采用单一传感器检测故障电弧时，检测灵敏度和准确度之间的矛盾难以解决，致使误检率居高不下，严重影响了电力系统的安全稳定运行。
发明内容
有鉴于此，本发明实施例期望提供一种基于全景信息的弧光故障识别 算法，采用多个不同类的传感器对电弧燃烧时的多种物理特征进行检测，同时提出一种捕捉弧光电流特性的误差反向传播(BP，Error Back Propagation Training)神经网络算法，结合各传感器信息与弧光电流信息决策弧光保护动作，进行卡尔曼滤波器滤波，从而降低误动与拒动概率，有效提升保护可靠性。
本发明实施例提供一种基于全景信息的弧光故障识别装置，包括传感器单元、局部卡尔曼滤波器、信息融合单元、全局卡尔曼滤波器和弧光识别单元；
所述传感器单元、局部卡尔曼滤波器、信息融合单元、全局卡尔曼滤波器和弧光识别单元依次顺序连接；
传感器单元包括若干传感器，传感器单元，配置为检测弧光发生时的电气量，每个传感器连接一个局部卡尔曼滤波器，对传感器输出数据进行滤波；
局部卡尔曼滤波器输出的信息经过信息融合单元进行信息融合后，基于全局卡尔曼滤波器进行全景信息的全局滤波，进行弧光判定。
上述方案中，传感器单元包括弧光传感器、温度传感器和压力传感器。
信息融合单元包括局部信息融合模块和全局信息融合模块；
局部信息融合模块对同类传感器输出的信息进行融合；
全局信息融合模块融合所有局部信息融合模块输出的全景信息。
全景信息是指针对孤光光谱特性的弧光检测的特性信息，包括电流特性和声光特性。
一种基于全景信息的弧光故障识别方法，包括以下步骤：
S1，传感器单元采集弧光发生时的电气量信息。
S2，建立BP神经网络，确定BP神经网络参数。
步骤S2包括以下步骤：
(201)，传感器单元连接多参量传感器数据的局部卡尔曼滤波器，分 析电弧故障发生时电流在时域和频域的特征变化，提取特征值作为BP神经网络的输入，采用BP神经网络检测电弧故障，采用遗传算法获得BP神经网络的初始权值，实现断路器弧光识别；
(202)，确定BP神经网络输入层与输出层神经元个数：
输入层神经元为6个，包括采样时间窗口内电流最大最小幅值差I diff、电流在中低频段的能量、电流在中低频段内小波包分解树第一节点Node(4，1)、第2节点Node(4，2)、第3节点Node(4，3)和第4节点Node(4，4)的平均均方根值(RMS，Root Mean Square)；
输出层神经元个数为1个；
(203)，确定隐藏层神经元个数：
基于公式(1)计算隐藏层神经元个数：
Figure PCTCN2018119211-appb-000001
式中H：表示神经网络隐藏层的神经元个数；
I：表示的是神经网络输入层的神经元个数；
O：表示的是神经网络输出层的神经元个数，α为弹性动量，取值范围为1～10；
S3，大数据神经网络训练。
对原始电流采样数据提取6个特征量形成N个样本数据作为训练样本，神经网络训练停止条件是训练迭代次数Epoch≥6000、输出误差MSE≤0.001或者梯度下降Gradi-ent≤1.00e －10；
S4，局部卡尔曼滤波器和全局卡尔曼滤波器初始值设定。
X 0是全局状态初始值，P 0是全局协方差阵初始值，Q 0是全局系统协方差阵初始值，全局卡尔曼滤波器和局部卡尔曼滤波器的信息分配因子按传感器侦测量特性设定；
P 0全局协方差阵初始值和Q 0全局系统协方差阵初始值是没有离散度的， 取0或者一个零漂误差；
S5，基于测量估计量和状态估计量完成采样点信息融合。
设定初始值在系统开始时刻，P m是全局协方差阵，Q m是全局系统协方差阵，其中m为传感器个数，全局卡尔曼滤波器和每个局部卡尔曼滤波器的信息分配因子基于公式(2)和公式(3)进行分配；
Figure PCTCN2018119211-appb-000002
公式(2)为全局卡尔曼滤波器信息因子分配公式，其中
Figure PCTCN2018119211-appb-000003
Q i是传感器第i点离散采样数据的全局系统协方差阵，Q为全局卡尔曼滤波器分配因子；
Figure PCTCN2018119211-appb-000004
公式(3)为局部卡尔曼滤波器信息因子分配公式，其中
Figure PCTCN2018119211-appb-000005
P i是传感器第i点离散采样数据值的全局协方差阵，P为局部卡尔曼滤波器分配因子；β i为权重因子；
公式(2)，公式(3)中权重因子β i遵守信息守恒定理，
且β 1+β 2+…β m-1+β m＝1(0≤β i≤1)，i＝1,2,3…m；
则局部卡尔曼滤波器算法为公式(4)、全局卡尔曼滤波器算法为公式(5)；
Figure PCTCN2018119211-appb-000006
公式(4)适用于同种类型的传感器，其中P i为传感器第i点离散采样数据的全局协方差阵，
Figure PCTCN2018119211-appb-000007
是第i点离散时间采样的状态估计矢量；D k是第k个传感器采样数据，Z i为第i点的观测值，
Figure PCTCN2018119211-appb-000008
为第i点数据的转置，R i为第i点数据的残差；
Figure PCTCN2018119211-appb-000009
公式(5)中，Q 全i是各类型传感器局部融合后第i组数据的全局系统协方差阵，
Figure PCTCN2018119211-appb-000010
为是各类型传感器局部融合后，第i组数据的离散时间采样的状态估计矢量；D k′表示第k个传感器局部融合后数据，Z′ i为数据局部融合后第i点的观测值，H′ T i为数据局部融合后第i点数据的转置，R′ i为数据局部融合后第i点数据的残差；
通过公式(4)，对故障电弧检测系统中的每个传感器探测到信号实施局部滤波处理；把每个局部卡尔曼滤波器的处理结果按传感器类别局部融合后，进行全局权重分配，实现信息融合，全局卡尔曼滤波器继续根据公式(5)对接收到融合信息实施滤波，分析弧光是否发生；
S6，基于全局卡尔曼滤波器输出的全景信息得出弧光发生信号，用于逻辑跳闸出口。
上述方案中，α的取值，满足神经网络的输出误差(MSE)≤0.001或者梯度下降≤1.00e －10；
α＝6时，神经网络输出误差最小。
应用本发明实施例具备以下有益技术效果：
(1)，本申请公开一种基于全景信息的弧光故障识别方法，捕捉弧光电流特性的BP神经网络算法，克服模拟滤波器不灵活和不稳定的问题，基于BP神经网络模型、卡尔曼滤波和全景信息的融合，对温度的变化、电路噪声的干扰改变带来的影响体现出坚强的鲁棒性，大幅降低了弧光保护误动的概率。
(2)，本申请公开一种基于全景信息的弧光故障识别装置，多种类传感器构建的全景信息滤波器，对多个不同故障电弧故障特征进行实时滤波，能够很好地消除噪声干扰和提高信噪比，提高了故障电弧检测的准确性和可靠性，完善了系统的容错能力和复原能力。有效提高电弧侦测能力。
(3)，基于本申请上述弧光故障识别方法的弧光保护装置，电弧故障正确动作概率在99.9％以上。
附图说明
图1为本发明实施例提供的基于全景信息的弧光故障识别装置示意图；
图2为本发明实施例提供的BP神经网络结构示意图；
图3为本发明实施例提供的滤波器架构图。
具体实施方式
以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
如图1所示，一种基于全景信息的弧光故障识别装置，包括传感器单元、局部卡尔曼滤波器、信息融合单元、全局卡尔曼滤波器和弧光识别单元；
传感器单元、局部卡尔曼滤波器、信息融合单元、全局卡尔曼滤波器和弧光识别单元依次顺序连接；
传感器单元包括若干传感器，传感器单元，配置为检测弧光发生时的电气量，每个传感器连接一个局部卡尔曼滤波器，局部卡尔曼滤波器基于不同传感器的滤波算法，对传感器输出数据进行滤波；
局部卡尔曼滤波器输出的信息经过信息融合单元进行信息融合后，基于全局卡尔曼滤波器进行全景信息的全局滤波，进行弧光判定。
传感器单元包括弧光传感器、温度传感器和压力传感器；在弧光发生时，光是最直接的现象，但是光会被遮挡，也会受到可见光污染干扰，所以需要多种传感器互相配合。电弧光能量密度大，燃弧附近温度会急速上升，温度上升会增加气体压强。所以这些传感器都是检测弧光发生时的电气量的。
每种传感器都有独特的波形数据，比如电流的和压力的就差异很大。需要捕捉不同的信息，所以针对不同传感器的滤波算法不同，每个传感器连接一个局部卡尔曼滤波器。由于每种信息的权重不同，比如最主要的弧光传感器，应占主要地位，所以多种传感器根据安装位置，种类不同，会乘以权重系数，进行数据信息融合，在所有传感器信息融合后，需要进一步滤除异常数据，需要一个针对全景信息的全局滤波器。
卡尔曼滤波器，为识别抖动数据的滤波器，也就是滤除波动数据，这些波动数据会引起勿动，通过神经网络的训练，可以获得一个合理的协方差值(离散度)，这个合理的离散度界定是否是真的有弧光发生还是干扰。
本实施例一组数据[10,10.1,10,9.9……]这组数据离散度很低，在离散度门限值范围内，也就是很平稳，没有干扰，也没有弧光发生，但如果是[10,10.1,90,99,99.1,100,100…….]这样的离散度高于离散度门限值，但是由于后续数据是平稳的，所以不会太高，这就是真弧光发生了。
为了提高弧光识别的灵敏度，本申请采用了不同类别的传感器，但是这些传感器由于采集的对象不同，数据的特征是不同的，需要对不同的传感器设计不同的局部卡尔曼滤波器，也就是针对不同的传感器，取选择好一个对应门限值的离散度。
每个类型的传感器，都有一个信息分配原则，即占有不同的权重，如果数值满值是100，当数值达到80就认为有弧光，但是这个数值中，诸如声音，压力的权重就比较小，起到辅助参考的作用，也就是说，即使声音传感器，经过它的卡尔曼滤波后，输出了一个满码值，在全局滤波器看来，它就是10分，而弧光传感器的数据，占比重就会很大。这些不同传感器的数据融合后的值，也是波动的，也会有误触发的风险，通过全局卡尔曼进行滤除不同传感器的数据融合后的波动。
局部信息融合模块对同类传感器输出的信息进行融合；
全局信息融合模块融合所有局部信息融合模块输出的全景信息。
全景信息是指针对孤光光谱特性的弧光检测的特性信息，包括电流特性和声光特性。采用多种传感器布置与监测信息融合；将混沌理论、小波分析、模糊理论和人工智能多学科综合技术应用于故障电弧弧声、弧光的辨识，降低弧光误识别，提高动作可靠性。
BP神经网络的设计与训练以及全景信息的融合。图2是本发明实施例提供的BP神经网络模型的结构示意图，当中的参数包括输入神经元个数、网络层数、隐藏层神经元个数以及输出神经元个数，需要根据电弧故障检测的实际情况进行设定。
一种基于全景信息的弧光故障识别方法，
包括以下步骤：
S1，传感器单元采集弧光发生时的电气量信息。
S2，建立BP神经网络，确定BP神经网络参数。
步骤S2包括以下步骤：
(201)，传感器单元连接多参量传感器数据的局部卡尔曼滤波器，分析电弧故障发生时电流在时域和频域的特征变化，提取特征值作为BP神经网络的输入，采用BP神经网络检测电弧故障，针对BP神经网络收敛慢的缺点，采用遗传算法获得BP神经网络的初始权值，加快神经网络的训练收敛速度，实现断路器弧光识别；
(202)，确定BP神经网络输入层与输出层神经元个数；
输入层神经元为6个，包括采样时间窗口内电流最大最小幅值差I diff、电流在中低频段(小于100KHz)的能量、电流在中低频段内小波包分解树第一节点Node(4，1)、第2节点Node(4，2)、第3节点Node(4，3)、第4节点Node(4，4)的平均均方根值(RMS)；
输出层神经元个数为1个，输出检测以电弧识别为标准，故只需1个输出神经元来表示“0”或者“1”。
(203)确定隐藏层神经元个数；
基于公式(1)计算隐藏层神经元个数：
Figure PCTCN2018119211-appb-000011
式中H：表示神经网络隐藏层的神经元个数；
I：表示的是神经网络输入层的神经元个数；
O：表示的是神经网络输出层的神经元个数，α为弹性动量，取值范围为1～10。
α的取值，满足神经网络的输出误差(MSE)≤0.001或者梯度下降≤1.00e －10。
α的取值具体包括以下步骤考虑到收敛速度与动作精度，神经网络训练停止条件是训练迭代次数Epoch≥6000；实验表明当α＝6时(即隐藏层神经元个数为6，符合前步骤中BP神经网络输入层与输出层神经元个数确定的结果)，神经网络在迭代次数为258停止训练；
若继续增加α(即增加隐藏层神经元个数)，虽然网络在训练的迭代次数有所减少，但网络输出误差却有所增加。故当α＝6即隐藏层取神经元个数为6时，神经网络输出误差最小。
S3，大数据神经网络训练。
对原始电流采样数据提取6个特征量形成N个样本数据作为训练样本(包括N/2个电弧故障样本和N/2个正常情况样本)，神经网络训练停止条件是训练迭代次数Epoch≥6000、输出误差MSE≤0.001或者梯度下降Gradi-ent≤1.00e －10；
S4，局部卡尔曼滤波器和全局卡尔曼滤波器初始值设定。
X 0是全局状态初始值，P 0是全局协方差阵初始值，Q 0是全局系统协方差阵初始值，全局卡尔曼滤波器和局部卡尔曼滤波器的信息分配因子按传感器侦测量特性设定；对于局部卡尔曼滤波器，如温度传感器滤波器初始值，根据开关柜内部温度工况，可设定为50摄氏度，光强度按自然光照度 设定，压力传感器可将初始值设为0；
P 0全局协方差阵初始值和Q 0全局系统协方差阵初始值是没有离散度的，取0或者一个零漂误差；
本实施例，对于局部卡尔曼滤波器，如温度传感器滤波器初始值，根据开关柜内部温度工况，设定为50摄氏度，光强度按自然光照度设定，压力传感器可将初始值设为0；
S5，基于测量估计量和状态估计量完成采样点信息融合。
设定初始值在系统开始时刻，P m是全局协方差阵，Q m是全局系统协方差阵，其中m为传感器个数，全局卡尔曼滤波器和每个局部卡尔曼滤波器的信息分配因子基于公式(2)和公式(3)进行分配；
Figure PCTCN2018119211-appb-000012
公式(2)为全局卡尔曼滤波器信息因子分配公式，其中
Figure PCTCN2018119211-appb-000013
Q i是传感器第i点离散采样数据的全局系统协方差阵，Q为全局卡尔曼滤波器分配因子；
Figure PCTCN2018119211-appb-000014
公式(3)为局部卡尔曼滤波器信息因子分配公式，其中
Figure PCTCN2018119211-appb-000015
P i是传感器第i点离散采样数据值的全局协方差阵,P为局部卡尔曼滤波器分配因子；β i为权重因子；
公式(2)，公式(3)中权重因子β i遵守信息守恒定理，
且β 1+β 2+…β m-1+β m＝1(0≤β i≤1)，i＝1,2,3…m；
则局部卡尔曼滤波器算法为公式(4)、全局卡尔曼滤波器算法为公式(5)；
Figure PCTCN2018119211-appb-000016
公式(4)适用于同种类型的传感器，如同是弧光传感器或同是温度传 感器。其中P i为传感器第i点离散采样数据的全局协方差阵，
Figure PCTCN2018119211-appb-000017
是第i点离散时间采样的状态估计矢量；D k是第k个传感器采样数据，Z i为第i点的观测值，
Figure PCTCN2018119211-appb-000018
为第i点数据的转置，R i为第i点数据的残差；
Figure PCTCN2018119211-appb-000019
公式(5)中，Q 全i是各类型传感器局部融合后第i组数据的全局系统协方差阵，
Figure PCTCN2018119211-appb-000020
为是各类型传感器局部融合后，第i组数据的离散时间采样的状态估计矢量；D k′表示第k个传感器局部融合后数据，Z′ i为数据局部融合后第i点的观测值，H′ T i为数据局部融合后第i点数据的转置，R′ i为数据局部融合后第i点数据的残差。
如图3所示，通过公式(4)，对故障电弧检测系统中的每个传感器探测到信号实施局部滤波处理；把每个局部卡尔曼滤波器的处理结果(
Figure PCTCN2018119211-appb-000021
P l；
Figure PCTCN2018119211-appb-000022
P 2，…)按传感器类别局部融合后，进行全局权重分配，达到信息融合。如图1架构所示。全局卡尔曼滤波器继续根据公式(5)对接收到融合信息实施滤波，分析弧光是否发生。
S6，基于全局卡尔曼滤波器输出的全景信息得出弧光发生信号，用于逻辑跳闸出口。
以上仅是本发明的优选实施方式，应当指出：对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本发明的保护范围。

Claims (6)

  1. 一种基于全景信息的弧光故障识别装置，包括：传感器单元、局部卡尔曼滤波器、信息融合单元、全局卡尔曼滤波器和弧光识别单元；
    所述传感器单元、局部卡尔曼滤波器、信息融合单元、全局卡尔曼滤波器和弧光识别单元依次顺序连接；
    传感器单元包括至少一个传感器，传感器单元，配置为检测弧光发生时的电气量，每个传感器连接一个局部卡尔曼滤波器，对传感器输出数据进行滤波；
    局部卡尔曼滤波器输出的信息经过信息融合单元进行信息融合后，基于全局卡尔曼滤波器进行全景信息的全局滤波，进行弧光判定。
  2. 根据权利要求1所述的基于全景信息的弧光故障识别装置，其中，
    传感器单元包括弧光传感器、温度传感器和压力传感器。
  3. 根据权利要求1所述的基于全景信息的弧光故障识别装置，其中，
    信息融合单元包括局部信息融合模块和全局信息融合模块；
    局部信息融合模块，配置为对同类传感器输出的信息进行融合；
    全局信息融合模块，配置为融合所有局部信息融合模块输出的全景信息。
  4. 根据权利要求1所述的基于全景信息的弧光故障识别装置，其中，
    全景信息是指针对孤光光谱特性的弧光检测的特性信息，包括电流特性和声光特性。
  5. 一种基于全景信息的弧光故障识别方法，包括以下步骤：
    S1，传感器单元采集弧光发生时的电气量信息；
    S2，建立误差反向传播BP神经网络，确定BP神经网络参数；
    步骤S2包括以下步骤：
    201，传感器单元连接多参量传感器数据的局部卡尔曼滤波器，分析电 弧故障发生时电流在时域和频域的特征变化，提取特征值作为BP神经网络的输入，采用BP神经网络检测电弧故障，基于遗传算法获得BP神经网络的初始权值；
    202，确定BP神经网络输入层与输出层神经元个数：
    输入层神经元为6个，包括采样时间窗口内电流最大最小幅值差I diff、电流在中低频段的能量、电流在中低频段内小波包分解树第一节点Node(4，1)、第2节点Node(4，2)、第3节点Node(4，3)和第4节点Node(4，4)的平均均方根值RMS；
    输出层神经元个数为1个；
    203，确定隐藏层神经元个数：
    基于公式(1)计算隐藏层神经元个数：
    Figure PCTCN2018119211-appb-100001
    其中，H表示神经网络隐藏层的神经元个数；
    I表示的神经网络输入层的神经元个数；
    O表示神经网络输出层的神经元个数，α为弹性动量，取值范围为1～10；
    S3，大数据神经网络训练；
    对原始电流采样数据提取6个特征量形成N个样本数据作为训练样本，神经网络训练停止条件是训练迭代次数不小于6000、输出误差不大于0.001或者梯度下降不大于1.00e －10；
    S4，局部卡尔曼滤波器和全局卡尔曼滤波器初始值设定；
    X 0是全局状态初始值，P 0是全局协方差阵初始值，Q 0是全局系统协方差阵初始值，全局卡尔曼滤波器和局部卡尔曼滤波器的信息分配因子按传感器侦测量特性设定；
    P 0全局协方差阵初始值和Q 0全局系统协方差阵初始值是没有离散度的， 取0或者一个零漂误差；
    S5，基于测量估计量和状态估计量完成采样点信息融合；
    设定初始值在系统开始时刻，P m是全局协方差阵，Q m是全局系统协方差阵，其中m为传感器个数，全局卡尔曼滤波器和每个局部卡尔曼滤波器的信息分配因子基于公式(2)和公式(3)进行分配；
    Figure PCTCN2018119211-appb-100002
    公式(2)为全局卡尔曼滤波器信息因子分配公式，其中
    Figure PCTCN2018119211-appb-100003
    Q i是传感器第i点离散采样数据的全局系统协方差阵，Q为全局卡尔曼滤波器分配因子；
    Figure PCTCN2018119211-appb-100004
    公式(3)为局部卡尔曼滤波器信息因子分配公式，其中P i -1＝β iP -1，P i是传感器第i点离散采样数据值的全局协方差阵,P为局部卡尔曼滤波器分配因子；β i为权重因子；
    公式(2)，公式(3)中权重因子β i遵守信息守恒定理，
    且β 1+β 2+…β i…+β m-1+β m＝1(0≤β i≤1)，i＝1,2,3…m；
    则局部卡尔曼滤波器算法为公式(4)、全局卡尔曼滤波器算法为公式(5)；
    Figure PCTCN2018119211-appb-100005
    公式(4)适用于同种类型的传感器，其中P i为传感器第i点离散采样数据的全局协方差阵，
    Figure PCTCN2018119211-appb-100006
    是第i点离散时间采样的状态估计矢量；D k是第k个传感器采样数据，Z i为第i点的观测值，
    Figure PCTCN2018119211-appb-100007
    为第i点数据的转置，R i为第i点数据的残差；
    Figure PCTCN2018119211-appb-100008
    公式(5)中，Q 全i是各类型传感器局部融合后第i组数据的全局系统协方差阵，
    Figure PCTCN2018119211-appb-100009
    为是各类型传感器局部融合后，第i组数据的离散时间采样的状态估计矢量；D k′表示第k个传感器局部融合后数据，Z′ i为数据局部融合后第i点的观测值，
    Figure PCTCN2018119211-appb-100010
    为数据局部融合后第i点数据的转置，R′ i为数据局部融合后第i点数据的残差；
    通过公式(4)，对故障电弧检测系统中的每个传感器探测到信号实施局部滤波处理；把每个局部卡尔曼滤波器的处理结果按传感器类别局部融合后，进行全局权重分配，实现信息融合，全局卡尔曼滤波器继续根据公式(5)对接收到融合信息实施滤波，分析弧光是否发生；
    S6，基于全局卡尔曼滤波器输出的全景信息得出弧光发生信号，所述弧光发生信号用于逻辑跳闸出口。
  6. 根据权利要求1所述的基于全景信息的弧光故障识别方法，其中，
    α的取值，满足神经网络的输出误差不大于0.001或者梯度下降不大于1.00e －10；
    α＝6时，神经网络输出误差最小。
PCT/CN2018/119211 2018-07-20 2018-12-04 一种基于全景信息的弧光故障识别装置及方法 WO2020015277A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810802425.2 2018-07-20
CN201810802425.2A CN109298291A (zh) 2018-07-20 2018-07-20 一种基于全景信息的弧光故障识别装置及方法

Publications (1)

Publication Number Publication Date
WO2020015277A1 true WO2020015277A1 (zh) 2020-01-23

Family

ID=65167873

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/119211 WO2020015277A1 (zh) 2018-07-20 2018-12-04 一种基于全景信息的弧光故障识别装置及方法

Country Status (2)

Country Link
CN (1) CN109298291A (zh)
WO (1) WO2020015277A1 (zh)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111967309A (zh) * 2020-07-03 2020-11-20 西安电子科技大学 一种电磁信号智能协同识别方法及系统
CN112561280A (zh) * 2020-12-09 2021-03-26 中铁第四勘察设计院集团有限公司 基于自学习收敛故障知识库的设备故障预测方法及其应用
CN113657156A (zh) * 2021-07-13 2021-11-16 西安理工大学 一种基于双轴心轨迹的卷对卷印刷单元故障诊断方法
CN113702739A (zh) * 2021-08-26 2021-11-26 广东电网有限责任公司 一种基于小波分解和神经网络的电弧检测方法及装置
CN113970697A (zh) * 2021-09-09 2022-01-25 北京无线电计量测试研究所 一种模拟电路状态评估方法和装置
CN114019296A (zh) * 2021-09-26 2022-02-08 广西电网有限责任公司电力科学研究院 一种基于bp神经网络的配电线路接地故障辨识方法
CN114077846A (zh) * 2021-10-14 2022-02-22 国网安徽省电力有限公司电力科学研究院 一种基于rf-lstm的故障电流多域识别方法及存储介质
CN114136436A (zh) * 2021-11-25 2022-03-04 国网山东省电力公司青岛供电公司 一种基于esfk和bp算法的输电杆塔大数据巡检系统及方法
CN114167729A (zh) * 2021-12-08 2022-03-11 北京工业大学 基于分布式二段自适应加权堆叠自动编码器的微小故障检测方法及系统
CN114237208A (zh) * 2021-08-12 2022-03-25 国网福建省电力有限公司 一种无人值守智能仓内自动路径导引方法及系统

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113076691A (zh) * 2019-09-23 2021-07-06 华为技术有限公司 直流电弧检测方法、装置、设备、系统及存储介质
CN110987267B (zh) * 2019-12-23 2022-04-22 佳讯飞鸿（北京）智能科技研究院有限公司 转辙机应力信号检测装置和方法及电子设备
CN111255436B (zh) * 2020-01-17 2023-07-28 成都理工大学 一种改善hdil阵列感应测井仪测量曲线质量的方法
CN113049922B (zh) 2020-04-22 2022-11-15 青岛鼎信通讯股份有限公司 一种采用卷积神经网络的故障电弧信号检测方法
CN112904228B (zh) * 2021-01-25 2023-05-26 国网江苏省电力有限公司检修分公司 基于电光信息复合判据的二次回路短路故障电弧识别方法
CN114139577A (zh) * 2021-11-15 2022-03-04 国网天津市电力公司 基于一维卷积自编码网络的故障电弧识别方法及装置
CN116026403A (zh) * 2022-11-09 2023-04-28 国能四川西部能源股份有限公司 开关柜故障预警方法、装置、介质和电子设备
CN117872122A (zh) * 2024-01-24 2024-04-12 广东电网有限责任公司江门供电局 一种基于高压断路器的机械故障诊断方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101702515A (zh) * 2009-11-24 2010-05-05 杭州之江开关股份有限公司 一种断路器相间弧光短路保护装置
US8098465B1 (en) * 2008-03-28 2012-01-17 Reliance Controls Corporation AFCI breaker providing protection for multiple branch circuits in an electrical panel
CN203014364U (zh) * 2013-01-06 2013-06-19 安徽中节节能环保科技有限公司 基于电弧光检测的电气保护装置
CN203504160U (zh) * 2013-10-22 2014-03-26 上海交通大学 弧光保护系统
CN205753370U (zh) * 2016-04-15 2016-11-30 北京劳动保障职业学院 一种弧光保护系统

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101676744B (zh) * 2007-10-31 2012-07-11 北京航空航天大学 一种复杂背景低信噪比下弱小目标高精度跟踪方法
US8756001B2 (en) * 2011-02-28 2014-06-17 Trusted Positioning Inc. Method and apparatus for improved navigation of a moving platform
US8797045B2 (en) * 2011-03-25 2014-08-05 Doble Lemke Gmbh Device for detecting partial discharge in an insulation system of rotary electric machines
CN102156476B (zh) * 2011-04-14 2013-12-18 山东大学 智能空间与护士机器人多传感器系统及其信息融合方法
CN102679977A (zh) * 2012-06-20 2012-09-19 南京航空航天大学 基于惯性网络的分布式导航装置及其信息融合方法
CN102841131B (zh) * 2012-09-20 2015-04-08 西安科技大学 钢丝绳芯输送带缺陷智能识别方法及系统
CN103776654B (zh) * 2014-02-21 2016-08-31 黑龙江省科学院自动化研究所 多传感器信息融合的故障诊断方法
CN104035067A (zh) * 2014-06-13 2014-09-10 重庆大学 一种基于无线传感器网络的移动机器人自动定位算法
CN105004336A (zh) * 2015-07-10 2015-10-28 中国科学院深圳先进技术研究院 一种机器人的定位方法
CN105352535A (zh) * 2015-09-29 2016-02-24 河海大学 一种基于多传感器数据融合的测量方法
CN105572492B (zh) * 2015-10-22 2018-05-04 北京建筑大学 一种城轨列车辅助逆变器故障诊断装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8098465B1 (en) * 2008-03-28 2012-01-17 Reliance Controls Corporation AFCI breaker providing protection for multiple branch circuits in an electrical panel
CN101702515A (zh) * 2009-11-24 2010-05-05 杭州之江开关股份有限公司 一种断路器相间弧光短路保护装置
CN203014364U (zh) * 2013-01-06 2013-06-19 安徽中节节能环保科技有限公司 基于电弧光检测的电气保护装置
CN203504160U (zh) * 2013-10-22 2014-03-26 上海交通大学 弧光保护系统
CN205753370U (zh) * 2016-04-15 2016-11-30 北京劳动保障职业学院 一种弧光保护系统

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111967309B (zh) * 2020-07-03 2024-02-06 西安电子科技大学 一种电磁信号智能协同识别方法及系统
CN111967309A (zh) * 2020-07-03 2020-11-20 西安电子科技大学 一种电磁信号智能协同识别方法及系统
CN112561280A (zh) * 2020-12-09 2021-03-26 中铁第四勘察设计院集团有限公司 基于自学习收敛故障知识库的设备故障预测方法及其应用
CN112561280B (zh) * 2020-12-09 2023-04-18 中铁第四勘察设计院集团有限公司 基于自学习收敛故障知识库的设备故障预测方法及其应用
CN113657156B (zh) * 2021-07-13 2024-03-08 佛山市富丽图数字印刷有限公司 一种基于双轴心轨迹的卷对卷印刷单元故障诊断方法
CN113657156A (zh) * 2021-07-13 2021-11-16 西安理工大学 一种基于双轴心轨迹的卷对卷印刷单元故障诊断方法
CN114237208A (zh) * 2021-08-12 2022-03-25 国网福建省电力有限公司 一种无人值守智能仓内自动路径导引方法及系统
CN113702739A (zh) * 2021-08-26 2021-11-26 广东电网有限责任公司 一种基于小波分解和神经网络的电弧检测方法及装置
CN113970697B (zh) * 2021-09-09 2023-06-13 北京无线电计量测试研究所 一种模拟电路状态评估方法和装置
CN113970697A (zh) * 2021-09-09 2022-01-25 北京无线电计量测试研究所 一种模拟电路状态评估方法和装置
CN114019296A (zh) * 2021-09-26 2022-02-08 广西电网有限责任公司电力科学研究院 一种基于bp神经网络的配电线路接地故障辨识方法
CN114077846B (zh) * 2021-10-14 2024-06-04 国网安徽省电力有限公司电力科学研究院 一种基于rf-lstm的故障电流多域识别方法及存储介质
CN114077846A (zh) * 2021-10-14 2022-02-22 国网安徽省电力有限公司电力科学研究院 一种基于rf-lstm的故障电流多域识别方法及存储介质
CN114136436B (zh) * 2021-11-25 2023-06-23 国网山东省电力公司青岛供电公司 一种基于eskf和bp算法的输电杆塔大数据巡检系统及方法
CN114136436A (zh) * 2021-11-25 2022-03-04 国网山东省电力公司青岛供电公司 一种基于esfk和bp算法的输电杆塔大数据巡检系统及方法
CN114167729A (zh) * 2021-12-08 2022-03-11 北京工业大学 基于分布式二段自适应加权堆叠自动编码器的微小故障检测方法及系统
CN114167729B (zh) * 2021-12-08 2024-05-28 北京工业大学 基于分布式二段自适应加权堆叠自动编码器的微小故障检测方法及系统

Also Published As

Publication number Publication date
CN109298291A (zh) 2019-02-01

Similar Documents

Publication Publication Date Title
WO2020015277A1 (zh) 一种基于全景信息的弧光故障识别装置及方法
US11489490B2 (en) Arc fault detection method for photovoltaic system based on adaptive kernel function and instantaneous frequency estimation
Ahmadi et al. A new method for detecting series arc fault in photovoltaic systems based on the blind-source separation
CN107340459B (zh) 一种直流故障电弧检测方法及系统
US11656263B2 (en) Effective feature set-based high impedance fault detection
CN107395121B (zh) 基于格拉布斯准则及离群点检测光伏阵列故障检测方法
Yao et al. Fault diagnosis model of adaptive miniature circuit breaker based on fractal theory and probabilistic neural network
WO2019127440A1 (zh) 一种直流电弧的处理方法及装置
CN210111945U (zh) 光伏系统直流侧电弧故障检测系统
CN114584069B (zh) 一种光伏直流电弧检测方法及系统
Souhe et al. Fault detection, classification and location in power distribution smart grid using smart meters data
Gao et al. Series arc fault detection of grid-connected PV system via SVD denoising and IEWT-TWSVM
Ashok et al. Fault diagnosis scheme for cross-country faults in dual-circuit line with emphasis on high-impedance fault syndrome
El Chamie et al. Physics-based features for anomaly detection in power grids with micro-pmus
Lin et al. A dual-window transient energy ratio-based adaptive single-phase reclosure criterion for EHV transmission line
Khond et al. A combined Data Mining and Machine Learning approach for High Impedance Fault Detection in Microgrids
Guillen et al. Hermite transform based algorithm for detection and classification of high impedance faults
Chan et al. Convolutional adversarial neural network (CANN) for fault diagnosis within a power system: addressing the challenge of event correlation for diagnosis by power disturbance monitoring equipment in a smart grid
CN113919198A (zh) 一种基于生成对抗网络的电气火灾监测预警方法
Fernandes et al. Power transformer disturbance classification based on the wavelet transform and artificial neural networks
CN113297786A (zh) 一种基于半监督式机器学习的低压故障电弧感知方法
Bera et al. Autoregressive Coefficients based intelligent protection of transmission lines connected to type-3 wind farms
Khond et al. Data mining methods for bad data detection and event data acquisition in microgrids
Chen et al. A data preparation method for machine-learning-based power system cyber-attack detection
CN114156830A (zh) 一种光电磁同步融合检测弧光保护方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18926749

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18926749

Country of ref document: EP

Kind code of ref document: A1