CN114118251A - Fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network - Google Patents

Fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network Download PDF

Info

Publication number
CN114118251A
CN114118251A CN202111390122.2A CN202111390122A CN114118251A CN 114118251 A CN114118251 A CN 114118251A CN 202111390122 A CN202111390122 A CN 202111390122A CN 114118251 A CN114118251 A CN 114118251A
Authority
CN
China
Prior art keywords
data
information
convolution
source
neural network
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
CN202111390122.2A
Other languages
Chinese (zh)
Inventor
王振浩
许超
李小兰
杨波
李泽曦
张琦
刘东延
卢毅
杨旭
郑舒文
谭澈
赵宁
孙守道
谢杰
赵贝加
张志鹏
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.)
Shenyang Aibeike Technology Co ltd
State Grid Corp of China SGCC
Shenyang Power Supply Co of State Grid Liaoning Electric Power Co Ltd
Original Assignee
Shenyang Aibeike Technology Co ltd
State Grid Corp of China SGCC
Shenyang Power Supply Co of State Grid Liaoning Electric Power Co Ltd
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 Shenyang Aibeike Technology Co ltd, State Grid Corp of China SGCC, Shenyang Power Supply Co of State Grid Liaoning Electric Power Co Ltd filed Critical Shenyang Aibeike Technology Co ltd
Priority to CN202111390122.2A priority Critical patent/CN114118251A/en
Publication of CN114118251A publication Critical patent/CN114118251A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/20—Administration of product repair or maintenance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Biophysics (AREA)
  • General Business, Economics & Management (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • Strategic Management (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention discloses a fault diagnosis and early warning method based on multi-source data fusion and a convolution twin neural network, which is characterized in that collected vibration information of a power transformer, concentration information of dissolved gas in oil, infrared information of the power transformer, voltage information and various types of data in current information are subjected to standardized integration by constructing a data dictionary, data information is extracted through double-space characteristics, fault recognition and classification of the power transformer are realized by utilizing the convolution twin neural network, and the fault diagnosis efficiency can be improved.

Description

Fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network
Technical Field
The invention relates to the technical field of power transformer fault diagnosis, in particular to a fault diagnosis and early warning method based on multi-source data fusion and a convolution twin neural network.
Background
For a long time, the maintenance strategy of the transformer in China mainly adopts regular maintenance taking time as a standard, has the defects of excessive maintenance and insufficient maintenance, has great blindness and compendity, causes great waste of manpower and material resources, and increases the probability of generating new hidden dangers. In consideration of the big data characteristic of the operation information of the transformer and the diversity and complexity of fault types, a new technological revolution and an industrial revolution require the intelligent and high-end transformation of electric equipment. The artificial intelligence algorithm can establish a simulated nervous system for information processing, and can form a characteristic deep learning sample by using a large amount of transformer fault instance data, so that the method has great advantages in the aspect of nonlinear complex fault identification.
With the increase of the capacity of a power grid and the proposal of a concept of a smart power grid, an intelligent substation is produced, and the continuous development of sensing monitoring equipment and the large-scale deployment and application on the surface and inside of a power transformer greatly improve the requirements on the intelligent level of monitoring, diagnosis and positioning of the power transformer equipment, and the data of the intelligent substation gradually presents typical big data characteristics of multiple sources, isomerism and the like.
In order to improve the safe operation level of a large-scale high-voltage power grid, a new method for multi-source information fusion fault diagnosis and residual life prediction of a power transformer is explored, the defects of 'maintenance surplus' and 'maintenance deficiency' existing at present are overcome, and the requirements of a new technological revolution and industrial revolution of power equipment towards intellectualization and high-end transformation are met. Concepts such as artificial intelligence, deep learning and the like are introduced to inject new vitality into the online fault monitoring, diagnosis and positioning of the power transformer under the big data background, and a brand-new thought and research direction are provided.
Disclosure of Invention
In view of the above, the invention provides a fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network, so as to realize on-line fault monitoring of a power transformer through artificial intelligence and deep learning.
The invention provides a technical scheme, in particular to a fault diagnosis and early warning method based on multi-source data fusion and a convolution twin neural network, which comprises the following steps:
s1: acquiring vibration information of the power transformer, concentration information of dissolved gas in oil, infrared information of the power transformer, voltage information and current information in real time and acquiring data;
s2: preprocessing the acquired power transformer data, eliminating abnormal and redundant data, and constructing a data dictionary to realize standardized integrated classification of multi-source heterogeneous data;
s3: and after extracting principal component data characteristics of abnormal state data in the multi-source heterogeneous data subjected to standardized integrated classification, inputting the extracted principal component data characteristics into a trained multiple convolution twin network model, preliminarily diagnosing, forming a multi-evidence body, fusing by utilizing an improved DS evidence theory, and finally obtaining a fault type and early warning.
Preferably, in step S2, the constructing a data dictionary realizes standardized integrated classification of multi-source heterogeneous data, and specifically includes:
s201: representing the preprocessed power transformer data by adopting a quintuple (epsilon NAME, pi INFORMATION, omega CONTENT, gamma CONT-MATION, tau IDENTIFY), wherein epsilon NAME is a source data NAME, pi INFORMATION is source data marking INFORMATION, omega CONTENT is source data CONTENT, gamma CONT-MATION is source data CONTENT marking INFORMATION, and tau IDENTIFY is identification data;
s202: extracting identification data information in power transformer data expressed by quintuple, and defining identification base as
Figure BDA0003366736240000021
If Cr≥Cr,baseIf the data is successfully identified, adding the data into a data type space corresponding to the data dictionary, otherwise, abandoning the data;
wherein, taunew,kRepresenting newly acquired identification data; tau iskThe identification data of the data type corresponding to the data dictionary is represented; cr,baseThe identification reference value is shown.
Further preferably, in step S3, the training process of the multiple convolution twin network model is as follows:
1) acquiring vibration information of the power transformer, concentration information of dissolved gas in oil, infrared information of the power transformer, voltage information and current information and acquiring data;
2) preprocessing the acquired power transformer data, eliminating abnormal and redundant data, and constructing a data dictionary to realize standardized integrated classification of multi-source heterogeneous data;
3) extracting principal component data characteristics of abnormal state data in the multi-source heterogeneous data classified by the standardized integration, and dividing a training set and a test set;
4) inputting training data in a training set into a multiple convolution twin network for model training;
5) inputting the test data in the test set into the multiple convolution twin network diagnostic model trained in the step 4) for preliminary diagnosis, then forming multiple evidence bodies, and finally obtaining the fault type by utilizing improved DS evidence theory fusion.
Further preferably, in step 2), the data dictionary is constructed to realize standardized integrated classification of multi-source heterogeneous data, and specifically:
2.1) representing the preprocessed power transformer data by quintuple (epsilon NAME, pi INFORMATION, omega CONTENT, gamma CONT-MATION and tau IDENTIFY), wherein epsilon NAME is a source data NAME, pi INFORMATION is source data marking INFORMATION, omega CONTENT is source data CONTENT, gamma CONT-MATION is source data CONTENT marking INFORMATION, and tau IDENTIFY is identification data;
2.2) extracting identification data information in the power transformer data expressed by quintuple and defining an identification base
Figure BDA0003366736240000031
If Cr≥Cr,baseIf the data is successfully identified, adding the data into a data type space corresponding to the data dictionary, otherwise, abandoning the data;
wherein, taunew,kRepresenting newly acquired identification data; tau iskThe identification data of the data type corresponding to the data dictionary is represented; cr,baseThe identification reference value is shown.
Further preferably, in step 3), extracting principal metadata features of abnormal state data in the multi-source heterogeneous data classified by the standardized integration, and dividing a training set and a test set, specifically:
3.1) transforming the acquired data into a matrix X, wherein the size of the matrix X is n multiplied by m, n is the number of samples, and m is the number of variables;
3.2) inputting the matrix X into a PCA space model, reducing high-dimensional data into low-dimensional data, and reserving main characteristic information of the original data, wherein the PCA space model is as follows:
Figure BDA0003366736240000032
wherein, tiAs a principal component vector, piIs a load vector;
3.3) dividing the obtained labeled principal component vectors into a training set and a testing set, wherein the training set accounts for 80 percent, and the testing set accounts for 20 percent.
Further preferably, step 4) inputs training data in the training set into the multiple convolution twin network for model training, specifically:
4.1) initializing convolution twin network parameters and structures, converting input data into a 3 x 3 matrix, then inputting convolution layers with the number of convolution kernels of 16 and the size of 2 x 2 into the network, wherein the expression is as follows:
yi＝f(xi*ωi+bi)
wherein, yiFor the output of convolutional layer i, f () is the activation function ReLU, xiFor input data, ωiAs weights of convolution kernels, biIs a bias vector.
4.2) inputting the convolution layer output to a pooling layer with the size of 2 multiplied by 2, adopting a maximum pooling mode, and having the expression:
Figure BDA0003366736240000033
wherein, ymaxOutputting data for the pooling layer, wherein R is a pooling area;
4.3) inputting output data of the pooling layer into a convolution layer with the size of 2 multiplied by 2 and the number of convolution kernels of 32, inputting the output data of the convolution layer into a maximum pooling layer, and then inputting the output data of the maximum pooling layer into a full-connection layer, wherein the expression is as follows:
y′i＝g(y′iω′i+b′i)
wherein, y'iOutputting data for full connectivity layer, ω'i,b′iRespectively, the weights and bias vectors of the fully connected layers.
4.4) judging the error between the loss function and the expected target, if the error meets the requirement, finishing the model training, wherein the loss function expression is as follows:
Figure BDA0003366736240000041
L(W,(Y,X1,X2)i)＝(1-Y)LG(EW(X1,X2)i)+T
T＝YL1(EW(X1,X2)i)
where P is the total input sample, Y is whether the two input samples are the same fault, LGFor two samples being loss functions of the same type, L1The same type of loss function is used for both samples.
Further preferably, the step 4) of inputting the training data in the training set into the multiple convolution twin network for model training further includes the following steps:
4.5) if the fault type of the input sample is difficult to judge, inputting the original input data into a KICA space, mapping the data to a high-dimensional space for principal component calculation, wherein the calculation method is as follows:
Figure BDA0003366736240000042
Figure BDA0003366736240000043
Figure BDA0003366736240000044
where K is the kernel matrix, ftr() Calculating trace;
4.6) mapping data to a high-dimensional feature space:
Figure BDA0003366736240000045
wherein the feature matrix v ═ v1,v2,...,vh]，Λ＝diag(λ1,λ2,...,λh)；
4.7) calculating an independent principal element s according to the following calculation formula:
s＝W·z；
4.8) inputting the characteristics extracted by the KICA space into a convolution twin network with the same structure and parameters for model training.
Further preferably, step 5) inputs the test data in the test set into a multiple convolution twin network diagnosis model for preliminary diagnosis, then forms multiple evidence bodies, and finally obtains the fault type by utilizing improved DS evidence theory fusion, specifically:
5.1) inputting test data in the test set into the multiple convolution twin network diagnostic model trained in the step 4), and constructing a BPA function according to the obtained classification result, wherein the expression is as follows:
Figure BDA0003366736240000051
wherein m isi(EM) Is EMBPA, EMAs a fault classType AiFor fault diagnosis accuracy of the ith convolution twin neural network, PiMThe classification result belongs to E when the ith convolution twin neural network is used for carrying out diagnosis classification on the sampleMThe probability of (d);
5.2) carrying out evidence correction by using a Langmuir distance function, wherein the correction expression is as follows:
Figure BDA0003366736240000052
wherein d isij(L) is the Langmuir distance of the two evidences, DiAs evidence body miTotal distance from other evidential bodies, Rrel(mi) Evaluating the reliability of the evidence;
5.3) performing evidence fusion according to a DS synthesis rule to finally obtain the fault type of the power transformer, wherein the synthesis rule is as follows:
Figure BDA0003366736240000053
Figure BDA0003366736240000054
wherein K is a conflict factor of the conflict degree of the two evidences.
According to the fault diagnosis and early warning method based on the multi-source data fusion and the convolution twin neural network, the collected vibration information of the power transformer, the concentration information of the dissolved gas in the oil, the infrared information of the power transformer, the voltage information and the current information of each type of data are subjected to standardized integration by constructing a data dictionary, the data information is extracted through double-space characteristics, the fault recognition and classification of the power transformer are realized by utilizing the convolution twin neural network, and the fault diagnosis efficiency can be improved.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.
In order to more clearly illustrate the embodiments or technical solutions in the prior art of the present invention, the drawings used in the description of the embodiments or prior art will be briefly described below, and it is obvious for those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.
FIG. 1 is a schematic flow chart of a fault diagnosis and early warning method based on a multi-source data fusion and convolution twin neural network according to an embodiment of the disclosure;
FIG. 2 is a flow chart of the dual spatial feature extraction and convolution twin neural network in an embodiment of the disclosure.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the invention, as detailed in the appended claims.
In order to realize the online fault monitoring of the power transformer through artificial intelligence and deep learning, the embodiment provides a fault diagnosis and early warning method based on multi-source data fusion and a convolution twin neural network, and referring to fig. 1, the diagnosis method comprises the following steps:
s1: acquiring vibration information of the power transformer, concentration information of dissolved gas in oil, infrared information of the power transformer, voltage information and current information in real time and acquiring data;
s2: preprocessing the acquired power transformer data, eliminating abnormal and redundant data, and constructing a data dictionary to realize standardized integrated classification of multi-source heterogeneous data;
s3: and after extracting principal component data characteristics of abnormal state data in the multi-source heterogeneous data subjected to standardized integrated classification, inputting the extracted principal component data characteristics into a trained multiple convolution twin network model, preliminarily diagnosing, forming a multi-evidence body, fusing by utilizing an improved DS evidence theory, and finally obtaining a fault type and early warning.
Step S1 connects vibration sensor, gas sensor, voltage transformer, current transformer and infrared sensor through the FPGA device to acquire vibration data, gas concentration data, voltage current data and infrared data, and through the 5G communication module, the bus communication module is used to transmit the acquired data information to the memory in the edge computing module for storage, and meanwhile, the 5G communication protocol is used to transmit the acquired transformer data to the cloud for storage as a data copy.
The collected data are transmitted to the 5G communication module through an RS485 protocol, the data are transmitted to the cloud server as backup data according to a 5G communication data analysis protocol, meanwhile, the FPGA module transmits the data to a memory of the edge computing module for storage through serial port to USB, and the specific computing process of the edge computing module is as described in steps S2 and S3.
In step S3, a double-space feature extraction method is used to extract principal metadata features of abnormal state data in the multi-source heterogeneous data subjected to standardized, integrated and classified, wherein a specific feature extraction process and a method for improving DS evidence theory fusion will be described in detail in a subsequent model training process, and will not be described in detail here.
In step S2, the constructing of the data dictionary realizes standardized integrated classification of multi-source heterogeneous data, which specifically includes:
s201: representing the preprocessed power transformer data by adopting a quintuple (epsilon NAME, pi INFORMATION, omega CONTENT, gamma CONT-MATION, tau IDENTIFY), wherein epsilon NAME is a source data NAME, pi INFORMATION is source data marking INFORMATION, omega CONTENT is source data CONTENT, gamma CONT-MATION is source data CONTENT marking INFORMATION, and tau IDENTIFY is identification data;
s202: extracting identification data information in power transformer data expressed by quintuple, and defining identification base as
Figure BDA0003366736240000071
If Cr≥Cr,baseIf the data is successfully identified, adding the data into a data type space corresponding to the data dictionary, otherwise, abandoning the data;
wherein, taunew,kRepresenting newly acquired identification data; tau iskThe identification data of the data type corresponding to the data dictionary is represented; cr,baseThe identification reference value is shown.
The training process of the multiple convolution twin network model in the step S3 is as follows:
1) acquiring vibration information of the power transformer, concentration information of dissolved gas in oil, infrared information of the power transformer, voltage information and current information and acquiring data;
2) preprocessing the acquired power transformer data, eliminating abnormal and redundant data, and constructing a data dictionary to realize standardized integrated classification of multi-source heterogeneous data;
3) extracting principal component data characteristics of abnormal state data in the multi-source heterogeneous data classified by the standardized integration, and dividing a training set and a test set;
4) inputting training data in a training set into a multiple convolution twin network for model training;
5) inputting the test data in the test set into the multiple convolution twin network diagnostic model trained in the step 4) for preliminary diagnosis, then forming multiple evidence bodies, and finally obtaining the fault type by utilizing improved DS evidence theory fusion.
In step 2), the data dictionary is constructed to realize the standardized integrated classification of the multi-source heterogeneous data, and the method specifically comprises the following steps:
2.1) representing the preprocessed power transformer data by quintuple (epsilon NAME, pi INFORMATION, omega CONTENT, gamma CONT-MATION and tau IDENTIFY), wherein epsilon NAME is a source data NAME, pi INFORMATION is source data marking INFORMATION, omega CONTENT is source data CONTENT, gamma CONT-MATION is source data CONTENT marking INFORMATION, and tau IDENTIFY is identification data;
2.2) judging the relevance of the data every time new data is collected. And if the data object is judged to be related to the main body, dividing the data object into corresponding data groups for storage. Specifically, identification data information in power transformer data expressed by quintuple is extracted, and an identification base is defined
Figure BDA0003366736240000081
If Cr≥Cr,baseIf the data is successfully identified, adding the data into a data type space corresponding to the data dictionary, otherwise, abandoning the data;
wherein, taunew,kRepresenting newly acquired identification data; tau iskThe identification data of the data type corresponding to the data dictionary is represented; cr,baseThe identification reference value is shown.
Further preferably, in step 3), principal metadata features of abnormal state data in the multi-source heterogeneous data classified by the standardized integration are extracted, and a training set and a test set are divided, wherein the principal metadata features are extracted by a double-space feature extraction method, specifically:
3.1) transforming the acquired data into a matrix X, wherein the size of the matrix X is n multiplied by m, n is the number of samples, and m is the number of variables;
3.2) inputting the matrix X into a PCA space model, reducing high-dimensional data into low-dimensional data, and reserving main characteristic information of the original data, wherein the PCA space model is as follows:
Figure BDA0003366736240000082
wherein, tiAs a principal component vector, piIs a load vector;
3.3) dividing the obtained labeled principal component vectors into a training set and a testing set, wherein the training set accounts for 80 percent, and the testing set accounts for 20 percent.
Further preferably, step 4) inputs training data in the training set into the multiple convolution twin network for model training, specifically:
4.1) initializing convolution twin network parameters and structures, converting input data into a 3 x 3 matrix, then inputting convolution layers with the number of convolution kernels of 16 and the size of 2 x 2 into the network, wherein the expression is as follows:
yi＝f(xi*ωi+bi)
wherein, yiFor the output of convolutional layer i, f () is the activation function ReLU, xiFor input data, ωiAs weights of convolution kernels, biIs a bias vector.
4.2) inputting the convolution layer output to a pooling layer with the size of 2 multiplied by 2, adopting a maximum pooling mode, and having the expression:
Figure BDA0003366736240000091
wherein, ymaxOutputting data for the pooling layer, wherein R is a pooling area;
4.3) inputting output data of the pooling layer into a convolution layer with the size of 2 multiplied by 2 and the number of convolution kernels of 32, inputting the output data of the convolution layer into a maximum pooling layer, and then inputting the output data of the maximum pooling layer into a full-connection layer, wherein the expression is as follows:
y′i＝g(y′iω′i+b′i)
wherein, y'iOutputting data for full connectivity layer, ω'i,b′iRespectively, the weights and bias vectors of the fully connected layers.
4.4) judging the error between the loss function and the expected target, if the error meets the requirement, finishing the model training, wherein the loss function expression is as follows:
Figure BDA0003366736240000092
L(W,(Y,X1,X2)i)＝(1-Y)LG(EW(X1,X2)i)+T
T＝YL1(EW(X1,X2)i)
where P is the total input sample, Y is whether the two input samples are the same fault, LGFor two samples being loss functions of the same type, L1The same type of loss function is used for both samples.
Further preferably, the step 4) of inputting the training data in the training set into the multiple convolution twin network for model training further includes the following steps:
4.5) if the fault type of the input sample is difficult to judge, inputting the original input data into a KICA space, mapping the data to a high-dimensional space for principal component calculation, wherein the calculation method is as follows:
Figure BDA0003366736240000093
Figure BDA0003366736240000094
Figure BDA0003366736240000095
where K is the kernel matrix, ftr() Calculating trace;
4.6) mapping data to a high-dimensional feature space:
Figure BDA0003366736240000096
wherein the feature matrix v ═ v1,v2,...,vh]，Λ＝diag(λ1,λ2,...,λh)；
4.7) calculating an independent principal element s according to the following calculation formula:
s＝W·z；
4.8) inputting the characteristics extracted by the KICA space into a convolution twin network with the same structure and parameters for model training.
Further preferably, in step 5), the test data in the test set is input into the multiple convolution twin network diagnostic model for preliminary diagnosis, then multiple evidence bodies are formed, and the fault type is finally obtained by utilizing improved DS evidence theory fusion, which can be seen in fig. 2 specifically as follows:
5.1) inputting test data in the test set into the multiple convolution twin network diagnostic model trained in the step 4), and constructing a BPA function according to the obtained classification result, wherein the expression is as follows:
Figure BDA0003366736240000101
wherein m isi(EM) Is EMBPA, EMAs a type of failure, AiFor fault diagnosis accuracy of the ith convolution twin neural network, PiMThe classification result belongs to E when the ith convolution twin neural network is used for carrying out diagnosis classification on the sampleMThe probability of (d);
5.2) carrying out evidence correction by using a Langmuir distance function, wherein the correction expression is as follows:
Figure BDA0003366736240000102
wherein d isij(L) is the Langmuir distance of the two evidences, DiAs evidence body miTotal distance from other evidential bodies, Rrel(mi) Evaluating the reliability of the evidence;
5.3) performing evidence fusion according to a DS synthesis rule to finally obtain the fault type of the power transformer, wherein the synthesis rule is as follows:
Figure BDA0003366736240000103
Figure BDA0003366736240000104
wherein K is a conflict factor of the conflict degree of the two evidences.
Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.
It is to be understood that the present invention is not limited to what has been described above, and that various modifications and changes may be made without departing from the scope thereof. The scope of the invention is limited only by the appended claims.

Claims (8)

1. A fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network is characterized by comprising the following steps:
s1: acquiring vibration information of the power transformer, concentration information of dissolved gas in oil, infrared information of the power transformer, voltage information and current information in real time and acquiring data;
s2: preprocessing the acquired power transformer data, eliminating abnormal and redundant data, and constructing a data dictionary to realize standardized integrated classification of multi-source heterogeneous data;
s3: and after extracting principal component data characteristics of abnormal state data in the multi-source heterogeneous data subjected to standardized integrated classification, inputting the extracted principal component data characteristics into a trained multiple convolution twin network model, preliminarily diagnosing, forming a multi-evidence body, fusing by utilizing an improved DS evidence theory, and finally obtaining a fault type and early warning.
2. The fault diagnosis and early warning method based on the multi-source data fusion and convolution twin neural network of claim 1, wherein in step S2, the data dictionary is constructed to realize standardized integrated classification of multi-source heterogeneous data, specifically:
s201: representing the preprocessed power transformer data by adopting a quintuple (epsilon NAME, pi INFORMATION, omega CONTENT, gamma CONT-MATION, tau IDENTIFY), wherein epsilon NAME is a source data NAME, pi INFORMATION is source data marking INFORMATION, omega CONTENT is source data CONTENT, gamma CONT-MATION is source data CONTENT marking INFORMATION, and tau IDENTIFY is identification data;
s202: extracting identification data information in power transformer data expressed by quintuple and defining identification base
Figure FDA0003366736230000011
If Cr≥Cr，baseIf the data is successfully identified, adding the data into a data type space corresponding to the data dictionary, otherwise, abandoning the data;
wherein, taunew，kRepresenting newly acquired identification data; tau iskThe identification data of the data type corresponding to the data dictionary is represented; cr，baseThe identification reference value is shown.
3. The fault diagnosis and early warning method based on the multi-source data fusion and the convolution twin neural network as claimed in claim 1, wherein in step S3, the training process of the multi-source data fusion and convolution twin neural network model is as follows:
1) acquiring vibration information of the power transformer, concentration information of dissolved gas in oil, infrared information of the power transformer, voltage information and current information and acquiring data;
2) preprocessing the acquired power transformer data, eliminating abnormal and redundant data, and constructing a data dictionary to realize standardized integrated classification of multi-source heterogeneous data;
3) extracting principal component data characteristics of abnormal state data in the multi-source heterogeneous data classified by the standardized integration, and dividing a training set and a test set;
4) inputting training data in a training set into a multiple convolution twin network for model training;
5) and inputting the test data in the test set into a multiple convolution twin network diagnosis model for preliminary diagnosis, then forming a multiple evidence body, and finally obtaining the fault type by utilizing the improved DS evidence theory fusion.
4. The fault diagnosis and early warning method based on the multi-source data fusion and convolution twin neural network according to claim 3, wherein in step 2), the data dictionary is built to realize standardized integrated classification of multi-source heterogeneous data, and specifically:
2.1) representing the preprocessed power transformer data by quintuple (epsilon NAME, pi INFORMATION, omega CONTENT, gamma CONT-MATION and tau IDENTIFY), wherein epsilon NAME is a source data NAME, pi INFORMATION is source data marking INFORMATION, omega CONTENT is source data CONTENT, gamma CONT-MATION is source data CONTENT marking INFORMATION, and tau IDENTIFY is identification data;
2.2) extracting identification data information in the power transformer data expressed by quintuple and defining an identification base
Figure FDA0003366736230000021
If Cr≥Cr，baseIf the data is successfully identified, adding the data into a data type space corresponding to the data dictionary, otherwise, abandoning the data;
wherein, taunew，kRepresenting newly acquired identification data; tau iskThe identification data of the data type corresponding to the data dictionary is represented; cr，baseThe identification reference value is shown.
5. The fault diagnosis and early warning method based on the multi-source data fusion and convolution twin neural network according to claim 3, characterized in that, step 3) extracts principal component data features of abnormal state data in the multi-source heterogeneous data classified by standardized integration, and divides a training set and a test set, specifically:
3.1) transforming the acquired data into a matrix X, wherein the size of the matrix X is n multiplied by m, n is the number of samples, and m is the number of variables;
3.2) inputting the matrix x into a PCA space model, reducing high-dimensional data into low-dimensional data, and reserving main characteristic information of the original data, wherein the PCA space model is as follows:
Figure FDA0003366736230000022
wherein, tiAs a principal component vector, piIs a load vector;
3.3) dividing the obtained labeled principal component vectors into a training set and a testing set, wherein the training set accounts for 80 percent, and the testing set accounts for 20 percent.
6. The fault diagnosis and early warning method based on the multi-source data fusion and convolution twin neural network according to claim 3, wherein step 4) inputs training data in a training set into the multi-source convolution twin neural network for model training, specifically:
4.1) initializing convolution twin network parameters and structures, converting input data into a 3 x 3 matrix, then inputting convolution layers with the number of convolution kernels of 16 and the size of 2 x 2 into the network, wherein the expression is as follows:
yi＝f(xi*ωi+bi)
wherein, yiFor the output of convolutional layer i, f () is the activation function ReLU, xiFor input data, ωiAs weights of convolution kernels, biIs a bias vector.
4.2) inputting the convolution layer output to a pooling layer with the size of 2 multiplied by 2, adopting a maximum pooling mode, and having the expression:
Figure FDA0003366736230000031
wherein, ymaxOutputting data for the pooling layer, wherein R is a pooling area;
4.3) inputting output data of the pooling layer into a convolution layer with the size of 2 multiplied by 2 and the number of convolution kernels of 32, inputting the output data of the convolution layer into a maximum pooling layer, and then inputting the output data of the maximum pooling layer into a full-connection layer, wherein the expression is as follows:
y′i＝g(y′iω′i+b′i)
wherein, y'iOutputting data for full connectivity layer, ω'i，b′iRespectively, the weights and bias vectors of the fully connected layers.
4.4) judging the error between the loss function and the expected target, if the error meets the requirement, finishing the model training, wherein the loss function expression is as follows:
Figure FDA0003366736230000032
L(W，(Y，X1，X2)i)＝(1-Y)LG(Ew(X1，X2)i)+T
T＝YL1(Ew(X1，X2)i)
where P is the total input sample, Y is whether the two input samples are the same fault, LGFor two samples being loss functions of the same type, L1The same type of loss function is used for both samples.
7. The fault diagnosis and early warning method for the multi-source data fusion and convolution twin neural network according to claim 6, wherein step 4) inputs training data in a training set into the multi-source convolution twin network for model training, and further comprises the following steps:
4.5) if the fault type of the input sample is difficult to judge, inputting the original input data into a KICA space, mapping the data to a high-dimensional space for principal component calculation, wherein the calculation method is as follows:
Figure FDA0003366736230000041
Figure FDA0003366736230000042
Figure FDA0003366736230000043
where K is the kernel matrix, ftr() Calculating trace;
4.6) mapping data to a high-dimensional feature space:
Figure FDA0003366736230000044
wherein the feature matrix v ═ v1，v2，...，vh]，Λ＝diag(λ1，λ2，...，λh)；
4.7) calculating an independent principal element s according to the following calculation formula:
s＝W·z；
4.8) inputting the characteristics extracted by the KICA space into a convolution twin network with the same structure and parameters for model training.
8. The fault diagnosis and early warning method for the multi-source data fusion and convolution twin neural network according to claim 3, characterized in that step 5) inputs test data in a test set into a multi-convolution twin network diagnosis model for preliminary diagnosis, then forms a multi-evidence body, and finally obtains a fault type by improving DS evidence theory fusion, specifically:
5.1) inputting test data in the test set into the multiple convolution twin network diagnostic model trained in the step 4), and constructing a BPA function according to the obtained classification result, wherein the expression is as follows:
Figure FDA0003366736230000045
wherein m isi(EM) Is EMBPA, EMAs a type of failure, AiFor fault diagnosis accuracy of the ith convolution twin neural network, PiMThe classification result belongs to E when the ith convolution twin neural network is used for carrying out diagnosis classification on the sampleMThe probability of (d);
5.2) carrying out evidence correction by using a Langmuir distance function, wherein the correction expression is as follows:
Figure FDA0003366736230000046
wherein d isij(L) is the Langmuir distance of the two evidences, DiAs evidence body miTotal distance from other evidential bodies, Rrel(mi) Evaluating the reliability of the evidence;
5.3) performing evidence fusion according to a DS synthesis rule to finally obtain the fault type of the power transformer, wherein the synthesis rule is as follows:
Figure FDA0003366736230000051
Figure FDA0003366736230000052
wherein K is a conflict factor of the conflict degree of the two evidences.
CN202111390122.2A 2021-11-22 2021-11-22 Fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network Pending CN114118251A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111390122.2A CN114118251A (en) 2021-11-22 2021-11-22 Fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111390122.2A CN114118251A (en) 2021-11-22 2021-11-22 Fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network

Publications (1)

Publication Number Publication Date
CN114118251A true CN114118251A (en) 2022-03-01

Family

ID=80439872

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111390122.2A Pending CN114118251A (en) 2021-11-22 2021-11-22 Fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network

Country Status (1)

Country Link
CN (1) CN114118251A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115114965A (en) * 2022-08-29 2022-09-27 中国长江三峡集团有限公司 Wind turbine generator gearbox fault diagnosis model, method, equipment and storage medium
CN115293057A (en) * 2022-10-10 2022-11-04 深圳先进技术研究院 Wind driven generator fault prediction method based on multi-source heterogeneous data
CN115598582A (en) * 2022-12-14 2023-01-13 国网福建省电力有限公司（Cn） Distributed mutual inductor online monitoring method, system, equipment and storage medium
CN116341396A (en) * 2023-05-30 2023-06-27 青岛理工大学 Complex equipment digital twin modeling method based on multi-source data fusion
CN116449255A (en) * 2023-03-09 2023-07-18 国网浙江省电力有限公司嘉兴供电公司 Fault detection system and method for box-type transformer
CN117972535A (en) * 2024-03-29 2024-05-03 华电电力科学研究院有限公司 Wind turbine generator set key component fault early warning method, device, equipment and storage medium

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115114965A (en) * 2022-08-29 2022-09-27 中国长江三峡集团有限公司 Wind turbine generator gearbox fault diagnosis model, method, equipment and storage medium
CN115293057A (en) * 2022-10-10 2022-11-04 深圳先进技术研究院 Wind driven generator fault prediction method based on multi-source heterogeneous data
CN115293057B (en) * 2022-10-10 2022-12-20 深圳先进技术研究院 Wind driven generator fault prediction method based on multi-source heterogeneous data
WO2024077753A1 (en) * 2022-10-10 2024-04-18 深圳先进技术研究院 Wind turbine fault prediction method based on multi-source heterogeneous data
CN115598582A (en) * 2022-12-14 2023-01-13 国网福建省电力有限公司（Cn） Distributed mutual inductor online monitoring method, system, equipment and storage medium
CN116449255A (en) * 2023-03-09 2023-07-18 国网浙江省电力有限公司嘉兴供电公司 Fault detection system and method for box-type transformer
CN116449255B (en) * 2023-03-09 2023-12-22 国网浙江省电力有限公司嘉兴供电公司 Fault detection system and method for box-type transformer
CN116341396A (en) * 2023-05-30 2023-06-27 青岛理工大学 Complex equipment digital twin modeling method based on multi-source data fusion
CN116341396B (en) * 2023-05-30 2023-08-11 青岛理工大学 Complex equipment digital twin modeling method based on multi-source data fusion
CN117972535A (en) * 2024-03-29 2024-05-03 华电电力科学研究院有限公司 Wind turbine generator set key component fault early warning method, device, equipment and storage medium

Similar Documents

Publication Publication Date Title
CN114118251A (en) Fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network
CN112286751B (en) Intelligent diagnosis system and method for high-end equipment fault based on edge cloud cooperation
CN109635677B (en) Compound fault diagnosis method and device based on multi-label classification convolutional neural network
CN111273125A (en) RST-CNN-based power cable channel fault diagnosis method
CN113191429A (en) Power transformer bushing fault diagnosis method and device
CN111680875A (en) Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model
CN110851654A (en) Industrial equipment fault detection and classification method based on tensor data dimension reduction
CN113608140A (en) Battery fault diagnosis method and system
CN116738339A (en) Multi-classification deep learning recognition detection method for small-sample electric signals
CN114118138A (en) Bearing composite fault diagnosis method based on multi-label field self-adaptive model
CN115510950A (en) Aircraft telemetry data anomaly detection method and system based on time convolution network
CN116842379A (en) Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models
CN117351659B (en) Hydrogeological disaster monitoring device and monitoring method
CN113486926B (en) Automatic change pier equipment anomaly detection system
CN112763215B (en) Multi-working-condition online fault diagnosis method based on modular federal deep learning
Tamilselvan et al. Multi-sensor health diagnosis using deep belief network based state classification
CN114912637B (en) Human-computer object knowledge graph manufacturing production line operation and maintenance decision method and system and storage medium
CN116343016A (en) Multi-angle sonar image target classification method based on lightweight convolution network
CN115730255A (en) Motor fault diagnosis and analysis method based on transfer learning and multi-source information fusion
CN114926702A (en) Small sample image classification method based on depth attention measurement
CN114841222A (en) Power distribution system, high-voltage circuit breaker fault detection model and method
CN113254468A (en) Fault query and reasoning method for certain type of equipment
Fu et al. Multi-scale Convolutional Neural Networks Based on Self-attention And Residual Network for Industrial Equipment Fault Diagnosis
CN117591813B (en) Complex equipment fault diagnosis method and system based on multidimensional features
CN114118140B (en) Multi-view intelligent fault diagnosis method and system for explosion-proof motor bearing

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