[go: up one dir, main page]

CN112307918A - Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network - Google Patents

Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network Download PDF

Info

Publication number
CN112307918A
CN112307918A CN202011132498.9A CN202011132498A CN112307918A CN 112307918 A CN112307918 A CN 112307918A CN 202011132498 A CN202011132498 A CN 202011132498A CN 112307918 A CN112307918 A CN 112307918A
Authority
CN
China
Prior art keywords
transformer
layer
membership function
neural network
parameters
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
CN202011132498.9A
Other languages
Chinese (zh)
Inventor
李涛
张琛亮
郭春林
朱柯佳
马慧远
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
North China Electric Power University
State Grid Beijing Electric Power Co Ltd
State Grid Corp of China SGCC
Original Assignee
North China Electric Power University
State Grid Beijing Electric Power Co Ltd
State Grid Corp of China SGCC
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 North China Electric Power University, State Grid Beijing Electric Power Co Ltd, State Grid Corp of China SGCC filed Critical North China Electric Power University
Priority to CN202011132498.9A priority Critical patent/CN112307918A/en
Publication of CN112307918A publication Critical patent/CN112307918A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M7/00—Vibration-testing of structures; Shock-testing of structures
    • G01M7/02—Vibration-testing by means of a shake table
    • G01M7/025—Measuring arrangements
    • 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
    • G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
    • G01R31/62—Testing of transformers
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Signal Processing (AREA)
  • Power Engineering (AREA)
  • Protection Of Transformers (AREA)

Abstract

本发明公开了一种基于模糊神经网络的变压器振动故障诊断方法，将采集到的变压器振动信号输入网络进行训练，并基于训练好的神经网络进行参数拟合，得到基于振动的变压器直流偏磁故障概率曲线，可以实现变压器直流偏磁的在线诊断，根据振动信号，实时判断变压器是否发生直流偏磁，同时在诊断方法中引入先验知识，减小了样本需求量，且提高了故障诊断的准确率。

Figure 202011132498

The invention discloses a transformer vibration fault diagnosis method based on a fuzzy neural network. The collected transformer vibration signals are input into the network for training, and parameter fitting is performed based on the trained neural network to obtain a vibration-based transformer DC bias fault. The probability curve can realize the online diagnosis of the DC bias of the transformer. According to the vibration signal, it can judge whether the transformer has DC bias in real time. At the same time, the prior knowledge is introduced into the diagnosis method, which reduces the sample demand and improves the accuracy of fault diagnosis. Rate.

Figure 202011132498

Description

Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network
Technical Field
The invention relates to a transformer fault method based on vibration, in particular to a transformer direct-current magnetic bias diagnosis method based on a fuzzy neural network, and belongs to the technical field of power transformers and artificial intelligence.
Background
The power transformer is a core device in a power system, bears the core tasks of electric energy conversion and transmission, is one of the most important devices in a power grid, and has great influence on the power grid due to transformer faults, and even can cause serious adverse social influence and economic loss. However, the existing method for diagnosing the health state of the transformer mostly needs shutdown for maintenance, and the fault diagnosis means for the online operation state of the transformer is limited. In recent years, with the rapid development of artificial intelligence, methods such as deep learning and big data are also introduced to the conventional problem of fault diagnosis of transformers. However, the artificial intelligence algorithm usually requires more data to achieve a good training effect, and the fault data of the transformer is just rare compared with the data in normal operation, so that the accuracy of fault diagnosis is affected due to insufficient training data.
Object of the Invention
The invention aims to overcome the defects in the prior art, adopts the idea of combining fuzzy mathematics and neural network to design a method suitable for detecting the state of a transformer, gives the probability of the transformer generating direct current magnetic biasing by processing and analyzing the vibration signal of the transformer, and mainly solves the following problems:
1. the current situation that shutdown maintenance is needed when the health state of the transformer is diagnosed at present is solved, online state detection of the transformer is achieved, and the state of the transformer is diagnosed in real time.
2. The problem of rare data when an artificial intelligence method is adopted to carry out fault diagnosis on the transformer is solved, and the sample demand during training is reduced by adopting a fuzzy neural network framework and introducing priori knowledge.
Disclosure of Invention
The invention provides a diagnosis method of transformer direct current magnetic biasing based on a fuzzy neural network, which comprises the following steps:
step 1: selecting fundamental frequency amplitude p of transformer vibrationf50Frequency complexity FC, ratio of odd-even sub-harmonic amplitudes λoeAs the characteristic quantity, a vibration sensor is adopted to collect vibration signal data of the transformer during working, and the data is analyzed and processed to obtain the characteristic quantity parameter of the transformer at the moment;
step 2: constructing a membership function and a neural network, and initializing related parameters;
and step 3: dividing the sample into a training set and a verification set, and training the neural network containing the membership function by using the training set until the error meets the requirement;
and 4, step 4: verifying the effectiveness of the trained model on a verification set;
and 5: searching key values in the membership function of the three characteristic quantities by using a trained model through a traversal method, thereby determining membership function parameters and obtaining a fault probability curve for fault diagnosis; the output of the trained model can only represent whether the transformer has faults or not, namely the output set is [0,1 ];
step 6: and obtaining the fault probabilities corresponding to the three characteristic quantities according to the fault probability curve, and taking the weighted average as the final fault probability of the transformer, namely the probability of DC magnetic biasing.
Further, the frequency complexity FC, the ratio λ of the odd-even sub-harmonic amplitudesoeThe calculation method (2) is shown in the following formulas (1) and (2):
Figure BDA0002735594610000021
wherein the fundamental frequency amplitude pf5050Hz frequency amplitude of 100-2000 Hz;
Figure BDA0002735594610000022
further, the neural network architecture in step 2 is composed of six layers, which are an input layer, a quantized input layer, 3 hidden layers and an output layer, wherein each hidden layer is provided with 6 neurons; the membership function is an S-shaped function, and is shown as a formula (3).
Figure BDA0002735594610000031
The first layer of the neural network is an input layer, x1,x2,x3The number of the nodes is 3, and the input layer transmits the collected characteristic quantity data to the second layer; the second layer is a quantitative input layer, the input variable is fuzzified through a membership function, the number of nodes is three, and each node represents a fuzzy set; the third to the fifth layers are hidden layers of the network; and the sixth layer is an output layer, the output results are 0 and 1, 1 represents that the transformer generates direct current magnetic biasing, and 0 represents that the transformer has no direct current magnetic biasing.
Further, the step 3 specifically includes:
step S31: inputting a training sample and expected output, and setting the learning error and the maximum training frequency;
step S32: initializing parameters of a membership function and each connection weight of nodes in a neural network;
step S33: inputting samples, fuzzifying the samples by using a membership function of the 2 nd layer, calculating the fuzzified samples by using the 3 rd to 5 th layers, and outputting the fuzzified samples by using the 6 th layer;
step S34: calculating the square error E (i) between the obtained target value and the actual value, and judging whether the error requirement is met;
step S35: if the requirement is not met, back propagation is carried out, parameter adjustment quantity of each layer is calculated, the parameters are updated, and if the requirement is met, the trained network and the trained parameters are stored.
Further, the determining parameters of the membership function in step 5 refers to determining parameters α and β of a membership function s (x), and specifically includes:
when the characteristic quantity X is larger than a certain specific value, no matter how the other two characteristic quantities take values, the neural network can judge that the neural network is in a fault state, and at the moment, the characteristic quantity X is recorded as XHNamely, as shown in formula (4):
Figure BDA0002735594610000032
taking epsilon as a minimum value, the formula (5) is shown as follows:
Figure BDA0002735594610000041
the finishing can be obtained as shown in formula (6):
Figure BDA0002735594610000042
finding a point X on the membership function curveMSo that the output of the trained model in step 5 changes from 0 to 1, assuming a point XMThe corresponding failure probability in the membership function curve s (x) is 1/2, i.e. as shown in equation (7),
Figure BDA0002735594610000043
the finishing is shown as a formula (8):
β＝-XM＝f(X1，X2) (8)，
wherein, X1，X2Are two characteristic quantities other than X; f (X)1，X2) Fitting X by testing the existing networkMCurve (c) of (d).
Drawings
FIG. 1 is a diagram of a neural network architecture to which embodiments of the present invention are applied.
FIG. 2 is a flow chart of neural network training in an embodiment of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and the specific examples.
In order to solve the problem of data labeling with insufficient sample quantity required by the existing diagnosis method, the invention constructs the labeling method of the transformer vibration data by using the thought of fuzzy mathematics, taking the vibration signal of the transformer as input, taking an expert knowledge base as a core, and taking the data labeled with the transformer state information as output, constructs the membership function by using a fitting method, reduces the requirement on expert knowledge, and verifies the effectiveness of the method by using a labeling experiment on sample data.
In the embodiment, a diagnosis method of the transformer direct current magnetic biasing based on the fuzzy neural network is disclosed, which comprises the following steps:
step S1: and obtaining the characteristic quantity parameters of the transformer at a certain moment.
The method comprises the steps of installing vibration sensors on the top surface and four side surfaces of a transformer, collecting vibration signals of the transformer during working, analyzing and processing the signals to obtain characteristic quantities, and labeling the obtained characteristic quantities to be divided into non-direct-current magnetic biasing and direct-current magnetic biasing. In this embodiment, the fundamental frequency amplitude p of the transformer vibration is selectedf50Frequency complexity FC, ratio of odd-even sub-harmonic amplitudes λoeAs the characteristic amount.
The frequency complexity and the ratio of odd-even sub-harmonic amplitudes are calculated as shown in the following formulas (1) and (2), respectively:
Figure BDA0002735594610000052
where FC is the frequency complexity, pf50In 100-2000Hz50Hz frequency amplitude.
Figure BDA0002735594610000051
Wherein λoeIs the ratio of the amplitudes of odd and even sub-harmonics
Step 2: constructing a membership function and a neural network, and initializing related parameters;
the neural network architecture applied in this embodiment is shown in fig. 1, and has six layers, namely an input layer, a quantized input layer, a 3-layer hidden layer, and an output layer, where each hidden layer has 6 neurons.
The first layer is an input layer, x1,x2,x3The number of the nodes is 3, and the input layer transmits the collected characteristic quantity data to the second layer.
The second layer is a quantitative input layer, the input variable is fuzzified through a membership function, the number of nodes is three, each node represents a fuzzy set and is used for calculating the membership value of the input component belonging to each fuzzy set, according to experience, the larger the characteristic quantity value is, the higher the possibility of the transformer generating direct current magnetic biasing is, so that the membership function of the system input variable selects an S-shaped function, as shown in formula (3):
Figure BDA0002735594610000061
the third to fifth layers are hidden layers of the network.
And the sixth layer is an output layer of the network, the output results are 0 and 1, 1 represents that the transformer generates direct current magnetic biasing, and 0 represents that the transformer has no direct current magnetic biasing.
The learning process of the fuzzy neural network is mainly divided into two stages, firstly, membership function parameters of each node of a fuzzy layer, namely alpha and beta in S (x), are solved according to an input transformer vibration characteristic quantity sample; then, after determining the number of neurons and their parameters, the weight between the hidden layer and the output layer is calculated, in the parameter optimization process, each gradient is calculated by adopting an error back propagation algorithm, then the parameters to be learned are adjusted by utilizing an optimization algorithm, and in the embodiment, the parameters α and β are optimized by adopting a first-order gradient optimization algorithm.
Step S3: training a neural network by using a training set sample, and specifically comprising the following steps:
step S31, inputting training sample and expected output, setting learning error and maximum training times
Step S32, initializing parameters of membership function and each connection weight of nodes in neural network
And step S33, inputting the sample, fuzzifying the sample by using the membership function of the 2 nd layer, calculating the fuzzified sample by using the 3 rd to 5 th layers, and outputting the fuzzified sample by using the 6 th layer.
Step S34, the square error e (i) between the calculated target value and the actual value is determined whether the error requirement is satisfied.
And step S35, if the requirement is not met, performing back propagation, calculating parameter adjustment quantity of each layer, updating the parameters, and if the requirement is met, storing the trained network and the trained parameters.
And step S36, verifying the validity of the trained model on the verification set.
Step S37, the output of the trained model can only indicate whether the transformer has a fault, i.e. the output set is [0,1], and in order to indicate the fault by the fault probability, we need to determine the parameters of the original membership function, i.e. α and β, by using the trained model.
To do this, we need to determine at least two points to determine the parameters of the sigmoid function.
When the characteristic quantity X is larger than a certain specific value, no matter how the other two characteristic quantities take values, the neural network can judge that the neural network is in a fault state, and at the moment, the characteristic quantity X is recorded as XH. It can be known that there are, for example,
Figure BDA0002735594610000071
taking epsilon as a minimum value, the above formula can be expressed,
Figure BDA0002735594610000072
after the finishing, the product can be obtained,
Figure BDA0002735594610000073
in a specific calculation, we can take epsilon as 0.001, (alpha, beta) epsilon [ (alpha, beta)1，β1)，(α2，β2)，(α3，β3) Represents pf50、λoeParameters in the membership function of the FC; xHRepresenting the input feature quantity variable.
For the characteristic quantity X, when we determine the other two characteristic quantities, one point X can be found on the membership function curveMSo that the output of the trained model changes from 0 to 1, it can be generally considered that this point corresponds to a failure probability of 1/2 in the membership function curve, which sometimes,
Figure BDA0002735594610000074
after the finishing, the product can be obtained,
β＝-XM
find XMThe precondition for this is to determine the other two characteristic quantities, i.e. to determine
β＝-XM＝f(X1，X2)
Wherein, X1，X2Are two characteristic quantities other than X; f (X)1，X2) X can be fitted by carrying out value test on the existing networkMCurve (c) of (d).
Determine f (X)1，X2) Then, we can construct the membership function of three characteristic quantities, namely a fault probability curve.
And step S38, inputting the three characteristic quantities into the fault probability curves respectively to obtain corresponding fault probabilities, and taking the weighted average of the fault probabilities to obtain the final fault probability of the transformer, namely the probability of the occurrence of the direct current magnetic biasing.

Claims (5)

1.一种基于模糊神经网络的变压器直流偏磁的诊断方法，其特征在于，包括以下步骤：1. a diagnostic method based on the transformer DC bias of fuzzy neural network, is characterized in that, comprises the following steps: 步骤1：选择变压器振动的基频幅值pf50、频率复杂度FC、奇偶次谐波幅值之比λoe作为特征量，采用振动传感器采集变压器工作时的振动信号数据，将数据分析处理得到该时刻变压器的特征量参数；Step 1: Select the fundamental frequency amplitude p f50 of the transformer vibration, the frequency complexity FC, and the ratio λ oe of the odd and even harmonic amplitudes as the characteristic quantities, use the vibration sensor to collect the vibration signal data of the transformer when it is working, and analyze and process the data to obtain The characteristic parameters of the transformer at this moment; 步骤2：构造隶属度函数及神经网络，并初始化相关参数；Step 2: Construct membership function and neural network, and initialize relevant parameters; 步骤3：将样本分为训练集和验证集，使用训练集对包含隶属度函数的神经网络进行训练，直至误差满足要求；Step 3: Divide the sample into a training set and a validation set, and use the training set to train the neural network including the membership function until the error meets the requirements; 步骤4：在验证集上验证训练好的模型的有效性；Step 4: Verify the validity of the trained model on the validation set; 步骤5：利用训练好的模型，通过遍历法寻找三特征量隶属度函数中的关键值，从而确定隶属度函数参数，获得可用于故障诊断的故障概率曲线；训练好的模型的输出仅可表示变压器的故障与否，即输出集为[0，1]；Step 5: Use the trained model to find the key value in the membership function of the three feature quantities by traversal method, so as to determine the parameters of the membership function, and obtain the fault probability curve that can be used for fault diagnosis; the output of the trained model can only represent Whether the transformer is faulty or not, that is, the output set is [0, 1]; 步骤6：根据故障概率曲线，得到三种特征量对应的故障概率，取其加权平均数为变压器最终的发生故障概率，即发生直流偏磁的概率。Step 6: According to the fault probability curve, the fault probability corresponding to the three characteristic quantities is obtained, and the weighted average is taken as the final fault probability of the transformer, that is, the probability of occurrence of DC bias. 2.根据权利要求1所述的诊断方法，其特征在于，步骤1中所述频率复杂度FC、奇偶次谐波幅值之比λoe的计算方法如下式(1)、(2)所示：2. The diagnostic method according to claim 1, wherein the calculation method of the frequency complexity FC, the ratio λ oe of the odd and even harmonic amplitudes described in the step 1 is shown in the following formulas (1), (2) :
Figure FDA0002735594600000011
Figure FDA0002735594600000011
Figure FDA0002735594600000012
Figure FDA0002735594600000012
 其中，基频幅值pf50为100-2000Hz中50Hz倍频频率幅值。Among them, the fundamental frequency amplitude p f50 is the 50Hz frequency multiplication frequency amplitude in 100-2000Hz. 3.根据权利要求2所述的诊断方法，其特征在于，步骤2中所述神经网络架构共有六层组成，依次分别为输入层、量化输入层、3层隐含层、输出层，其中每个隐含层有6个神经元；所述隶属度函数选择S型函数，如式(3)所示：3. The diagnostic method according to claim 2, wherein the neural network architecture described in step 2 consists of six layers, which are respectively an input layer, a quantized input layer, three hidden layers, and an output layer, wherein each Each hidden layer has 6 neurons; the membership function selects a sigmoid function, as shown in formula (3):
Figure FDA0002735594600000021
Figure FDA0002735594600000021
 所述神经网络的第一层为输入层，x1，x2，x3分别为三个输入特征量，节点数为3个，输入层将采集到的特征量数据传递到第二层；第二层为量化输入层，通过隶属度函数将输入变量模糊化，节点数为三个，每个节点代表一个模糊集合；第三到五层为网络的隐含层；第六层为输出层，输出结果为0和1，1代表变压器发生直流偏磁，0代表变压器无直流偏磁。The first layer of the neural network is the input layer, x 1 , x 2 , and x 3 are three input feature quantities respectively, the number of nodes is 3, and the input layer transmits the collected feature quantity data to the second layer; The second layer is the quantized input layer, which fuzzifies the input variables through the membership function. The number of nodes is three, and each node represents a fuzzy set; the third to fifth layers are the hidden layers of the network; the sixth layer is the output layer, The output results are 0 and 1, 1 means that the transformer has DC bias, and 0 means that the transformer has no DC bias. 4.根据权利要求3所述的诊断方法，其特征在于，所述步骤3具体包括：4. The diagnostic method according to claim 3, wherein the step 3 specifically comprises: 步骤S31：输入训练样本及期望输出，设定学习误差那个及最大训练次数；Step S31: input training samples and expected output, set the learning error and the maximum training times; 步骤S32：初始化隶属度函数的参数以及神经网络中节点的各连接权值；Step S32: Initialize the parameters of the membership function and the connection weights of the nodes in the neural network; 步骤S33：输入样本，第2层隶属度函数将其模糊化，并经过第3-5层进行计算，通过第6层输出；Step S33: input the sample, fuzzify it by the membership function of the second layer, calculate it through the third to fifth layers, and output it through the sixth layer; 步骤S34：计算得到的目标值与实际值之间的平方误差E(i)，判断是否满足误差要求；Step S34: Calculate the squared error E(i) between the target value and the actual value, and determine whether the error requirement is met; 步骤S35：若不满足要求，进行反向传播，计算各层参数调整量，并更新参数，若满足要求，保存已训练好的网络及参数。Step S35: If the requirements are not met, perform backpropagation, calculate the parameter adjustment amount of each layer, and update the parameters. If the requirements are met, save the trained network and parameters. 5.根据权利要求4所述的诊断方法，其特征在于，所述步骤5中所述确定隶属度函数参数指确定隶属度函数S(x)的参数α和β，具体为：5. The diagnostic method according to claim 4, wherein the determining the membership function parameters in the step 5 refers to determining the parameters α and β of the membership function S(x), specifically: 当特征量X大于某个特定值时，不论其余两个特征量如何取值，神经网络都会判断其为故障状态，此时记特征量X为XH，即如式(4)所示：When the feature quantity X is greater than a certain value, regardless of the values of the other two feature quantities, the neural network will judge it as a fault state. At this time, the feature quantity X is denoted as X H , as shown in formula (4):
Figure FDA0002735594600000031
Figure FDA0002735594600000031
 取ε作为极小值，则如式(5)所示：Taking ε as the minimum value, it is shown in formula (5):
Figure FDA0002735594600000032
Figure FDA0002735594600000032
 整理可得，如式(6)所示：It can be obtained by sorting, as shown in formula (6):
Figure FDA0002735594600000033
Figure FDA0002735594600000033
 在隶属度函数曲线上找到一个点XM，使得步骤5中训练好的模型的输出从0变到1，假定点XM在隶属度函数曲线S(x)中对应的故障概率为1/2，即如式(7)所示，Find a point X M on the membership function curve, so that the output of the model trained in step 5 changes from 0 to 1, assuming that the failure probability corresponding to the point X M in the membership function curve S(x) is 1/2 , that is, as shown in formula (7),
Figure FDA0002735594600000034
Figure FDA0002735594600000034
 整理得到，如式(8)所示：After finishing, as shown in formula (8): β＝-XM＝f(X1，X2) (8)，β=-X M =f(X 1 , X 2 ) (8), 其中，X1，X2是除了X以外的两个特征量；f(X1，X2)通过对已有网络进行取值测试，拟合出XM的曲线。Among them, X 1 , X 2 are two characteristic quantities other than X; f(X 1 , X 2 ) fits the curve of X M by testing the value of the existing network.
CN202011132498.9A 2020-10-21 2020-10-21 Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network Pending CN112307918A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011132498.9A CN112307918A (en) 2020-10-21 2020-10-21 Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011132498.9A CN112307918A (en) 2020-10-21 2020-10-21 Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network

Publications (1)

Publication Number Publication Date
CN112307918A true CN112307918A (en) 2021-02-02

Family

ID=74326843

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011132498.9A Pending CN112307918A (en) 2020-10-21 2020-10-21 Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network

Country Status (1)

Country Link
CN (1) CN112307918A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113670428A (en) * 2021-07-19 2021-11-19 国网四川省电力公司电力科学研究院 A method for online abnormality detection of transformer vibration
CN115453235A (en) * 2022-08-19 2022-12-09 深圳供电局有限公司 Method, device, equipment and storage medium for determining direct current magnetic bias information of transformer
CN120507691A (en) * 2025-07-17 2025-08-19 广东正超电气有限公司 An intelligent monitoring method for internal faults in box-type substations

Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6301572B1 (en) * 1998-12-02 2001-10-09 Lockheed Martin Corporation Neural network based analysis system for vibration analysis and condition monitoring
US6757665B1 (en) * 1999-09-28 2004-06-29 Rockwell Automation Technologies, Inc. Detection of pump cavitation/blockage and seal failure via current signature analysis
CN102570392A (en) * 2012-01-17 2012-07-11 上海电力学院 Method for identifying exciting inrush current of transformer based on improved probability neural network
CN103149470A (en) * 2013-01-29 2013-06-12 北京信息科技大学 Method of distinguishing transformer magnetizing rush current by transformer winding vibration
CN104463187A (en) * 2014-10-22 2015-03-25 宁波力芯科信息科技有限公司 Membership function generator unit, membership function generator array and fuzzy recognizer
CN106569056A (en) * 2016-10-21 2017-04-19 广州供电局有限公司 Power transformer fault diagnosis method and diagnosis device
CN107101714A (en) * 2017-05-09 2017-08-29 华北电力大学（保定） A kind of transformer health Evaluation method based on multi-measuring point vibration signal characteristics
CN109033612A (en) * 2018-07-20 2018-12-18 广西电网有限责任公司电力科学研究院 A kind of Diagnosis Method of Transformer Faults based on vibration noise and BP neural network
CN109669087A (en) * 2019-01-31 2019-04-23 国网河南省电力公司 A kind of method for diagnosing fault of power transformer based on Multi-source Information Fusion
CN109884459A (en) * 2019-01-30 2019-06-14 国网浙江省电力有限公司电力科学研究院 An intelligent online diagnosis and positioning method for winding deformation of power transformers
CN109961239A (en) * 2019-04-03 2019-07-02 杭州安脉盛智能技术有限公司 Transformer fault reasoning by cases method and system
CN110188837A (en) * 2019-06-25 2019-08-30 长春工业大学 A Method of MVB Network Fault Diagnosis Based on Fuzzy Neural
CN110927511A (en) * 2019-10-18 2020-03-27 云南电网有限责任公司 Power grid fault diagnosis system and method using fault waveform
CN111476318A (en) * 2020-04-30 2020-07-31 常州大学 A Transformer Fault Diagnosis Method and System Based on Fuzzy Decision Making
CN111612033A (en) * 2020-04-15 2020-09-01 广东电网有限责任公司 Distribution transformer fault diagnosis method based on gravity search and density peak clustering

Patent Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6301572B1 (en) * 1998-12-02 2001-10-09 Lockheed Martin Corporation Neural network based analysis system for vibration analysis and condition monitoring
US6757665B1 (en) * 1999-09-28 2004-06-29 Rockwell Automation Technologies, Inc. Detection of pump cavitation/blockage and seal failure via current signature analysis
CN102570392A (en) * 2012-01-17 2012-07-11 上海电力学院 Method for identifying exciting inrush current of transformer based on improved probability neural network
CN103149470A (en) * 2013-01-29 2013-06-12 北京信息科技大学 Method of distinguishing transformer magnetizing rush current by transformer winding vibration
CN104463187A (en) * 2014-10-22 2015-03-25 宁波力芯科信息科技有限公司 Membership function generator unit, membership function generator array and fuzzy recognizer
CN106569056A (en) * 2016-10-21 2017-04-19 广州供电局有限公司 Power transformer fault diagnosis method and diagnosis device
CN107101714A (en) * 2017-05-09 2017-08-29 华北电力大学（保定） A kind of transformer health Evaluation method based on multi-measuring point vibration signal characteristics
CN109033612A (en) * 2018-07-20 2018-12-18 广西电网有限责任公司电力科学研究院 A kind of Diagnosis Method of Transformer Faults based on vibration noise and BP neural network
CN109884459A (en) * 2019-01-30 2019-06-14 国网浙江省电力有限公司电力科学研究院 An intelligent online diagnosis and positioning method for winding deformation of power transformers
CN109669087A (en) * 2019-01-31 2019-04-23 国网河南省电力公司 A kind of method for diagnosing fault of power transformer based on Multi-source Information Fusion
CN109961239A (en) * 2019-04-03 2019-07-02 杭州安脉盛智能技术有限公司 Transformer fault reasoning by cases method and system
CN110188837A (en) * 2019-06-25 2019-08-30 长春工业大学 A Method of MVB Network Fault Diagnosis Based on Fuzzy Neural
CN110927511A (en) * 2019-10-18 2020-03-27 云南电网有限责任公司 Power grid fault diagnosis system and method using fault waveform
CN111612033A (en) * 2020-04-15 2020-09-01 广东电网有限责任公司 Distribution transformer fault diagnosis method based on gravity search and density peak clustering
CN111476318A (en) * 2020-04-30 2020-07-31 常州大学 A Transformer Fault Diagnosis Method and System Based on Fuzzy Decision Making

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
CHENLIANG ZHANG等: "An Evaluation Method of DC Magnetic bias Vibration for Transformer based on Prior Knowledge and Neural Network Modeling", 《JOURNAL OF PHYSICS: CONFERENCE SERIES》, pages 1 - 8 *
RONG-JONG WAI等: "Design of Adaptive Fuzzy-Neural-Network Control for a Single-Stage Boost Inverter", 《IEEE TRANSACTIONS ON POWER ELECTRONICS》, vol. 30, no. 12, pages 7282 - 7298, XP011667335, DOI: 10.1109/TPEL.2015.2396891 *
WEI WANG等: ""fuzzy" "neural network" "frequency complexity"", 《JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY》, pages 1389 *
WEI WANG等: "A rub fault recognition method based on generative adversarial nets", 《JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY》, pages 1389 *
余长厅等: "基于振动噪声及BP神经网络的变压器故障诊断方法研究", 《高压电器》, vol. 56, no. 06, pages 256 - 261 *
张琛亮: "基于混合高斯模型的变压器振动状态诊断方法", 《中国优秀硕士学位论文全文数据库 (工程科技Ⅱ辑)》, no. 01, pages 042 - 773 *
温阳东等: "基于模糊神经网络的电力变压器故障诊断", 《基于模糊神经网络的电力变压器故障诊断》, vol. 21, no. 01, pages 39 - 41 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113670428A (en) * 2021-07-19 2021-11-19 国网四川省电力公司电力科学研究院 A method for online abnormality detection of transformer vibration
CN113670428B (en) * 2021-07-19 2023-12-22 国网四川省电力公司电力科学研究院 Transformer vibration online abnormality detection method
CN115453235A (en) * 2022-08-19 2022-12-09 深圳供电局有限公司 Method, device, equipment and storage medium for determining direct current magnetic bias information of transformer
CN120507691A (en) * 2025-07-17 2025-08-19 广东正超电气有限公司 An intelligent monitoring method for internal faults in box-type substations

Similar Documents

Publication Publication Date Title
CN109460618B (en) Rolling bearing residual life online prediction method and system
Li et al. An ensemble learning-based prognostic approach with degradation-dependent weights for remaining useful life prediction
Soualhi et al. Prognosis of bearing failures using hidden Markov models and the adaptive neuro-fuzzy inference system
CN113609955B (en) A three-phase inverter parameter identification method based on deep learning and digital twin
CN109407654B (en) A nonlinear causal analysis method for industrial data based on sparse deep neural network
CN111638028A (en) High-voltage parallel reactor mechanical state evaluation method based on vibration characteristics
CN112033463B (en) Nuclear power equipment state evaluation and prediction integrated method and system
CN102707256B (en) Fault diagnosis method based on BP-Ada Boost nerve network for electric energy meter
CN109343505A (en) Prediction method of gear remaining life based on long short-term memory network
CN112307918A (en) Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network
CN108256556A (en) Wind-driven generator group wheel box method for diagnosing faults based on depth belief network
CN117556347A (en) Power equipment fault prediction and health management method based on industrial big data
Zhao et al. Bearing health condition prediction using deep belief network
Mao et al. Comparative study on prediction of fuel cell performance using machine learning approaches
CN111581880A (en) Rolling bearing residual life prediction method based on APSO-DBN
Afrasiabi et al. Two-stage deep learning-based wind turbine condition monitoring using SCADA data
CN113554229A (en) Three-phase voltage unbalance abnormal detection method and device
CN113887571A (en) Electronic equipment failure prediction method based on improved SVR algorithm
Zhu et al. Remaining useful life prediction approach based on data model fusion: An application in rolling bearings
CN115015753A (en) Rotary equipment prediction diagnosis method of LSTM prediction model based on wolf optimization algorithm
CN117668623B (en) Multi-sensor cross-domain fault diagnosis method for ship pipeline valve leakage
CN119226913A (en) A method and system for online diagnosis of transformer winding deformation based on machine learning
Luo et al. A novel method for remaining useful life prediction of roller bearings involving the discrepancy and similarity of degradation trajectories
CN118194191A (en) An intelligent prediction method for coal mill faults integrating Stacking ensemble learning and raccoon optimization algorithm
CN117421918A (en) Transformer state prediction model construction method and system for digital twin platform

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
AD01 Patent right deemed abandoned
AD01 Patent right deemed abandoned

Effective date of abandoning: 20241115