CN117574264A - Transformer fault diagnosis method and system based on knowledge constraint neural network - Google Patents
Transformer fault diagnosis method and system based on knowledge constraint neural network Download PDFInfo
- Publication number
- CN117574264A CN117574264A CN202311552523.2A CN202311552523A CN117574264A CN 117574264 A CN117574264 A CN 117574264A CN 202311552523 A CN202311552523 A CN 202311552523A CN 117574264 A CN117574264 A CN 117574264A
- Authority
- CN
- China
- Prior art keywords
- neural network
- transformer
- knowledge
- information
- modal
- 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
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 151
- 238000003745 diagnosis Methods 0.000 title claims abstract description 70
- 238000000034 method Methods 0.000 title claims abstract description 63
- 230000007246 mechanism Effects 0.000 claims abstract description 77
- 230000004927 fusion Effects 0.000 claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000012512 characterization method Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 39
- 238000004590 computer program Methods 0.000 claims description 21
- 238000012545 processing Methods 0.000 claims description 14
- 238000003860 storage Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 239000013598 vector Substances 0.000 claims description 8
- 238000009826 distribution Methods 0.000 claims description 6
- 230000003287 optical effect Effects 0.000 claims description 6
- 230000008278 dynamic mechanism Effects 0.000 claims description 4
- 238000001228 spectrum Methods 0.000 claims description 4
- 238000009825 accumulation Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- 230000001502 supplementing effect Effects 0.000 claims description 2
- 238000004422 calculation algorithm Methods 0.000 description 16
- 238000010586 diagram Methods 0.000 description 8
- 238000004519 manufacturing process Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 6
- 238000005457 optimization Methods 0.000 description 6
- 238000005070 sampling Methods 0.000 description 6
- 238000007689 inspection Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 241000287127 Passeridae Species 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 230000003449 preventive effect Effects 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004587 chromatography analysis Methods 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000011810 insulating material Substances 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000012925 reference material Substances 0.000 description 1
- 238000004171 remote diagnosis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Probability & Statistics with Applications (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Testing Electric Properties And Detecting Electric Faults (AREA)
Abstract
A transformer fault diagnosis method and system based on knowledge constraint neural network, the method includes extracting multiple kinds of transformer physical mode information data, and inputting the transformer physical mode information data differential combination into the neural network; dynamically acquiring characteristic information of different transformer data samples, dynamically evaluating the modal information of the different samples by using the modal confidence coefficient, fusing the neural network through a unified multi-modal fusion frame to obtain a fused neural network, and dynamically fusing the characteristic information and the modal information of the different samples through the fused neural network; extracting the mechanism knowledge of the transformer, carrying out mathematical characterization on the mechanism knowledge, and carrying out embedded with the fused neural network to obtain a knowledge constraint neural network; and performing active early warning training on the knowledge constraint neural network, and performing fault diagnosis by using the trained knowledge constraint neural network. The invention improves the recognition accuracy of rare fault cases while ensuring the recognition accuracy of the historical common fault phenomenon.
Description
Technical Field
The invention belongs to the technical field of transformer fault diagnosis, and particularly relates to a transformer fault diagnosis method and system based on a knowledge constraint neural network.
Background
The artificial intelligence technology based on data driving can realize real-time monitoring and remote diagnosis of the performance and state of the transformer. By collecting a large amount of sensor data and operational records, potential problems can be detected in time, equipment faults can be predicted, and timely alarms can be provided, so that the risks of power failure and equipment damage are reduced. Second, the technique can effectively process large-scale data, identifying complex patterns and trends. This helps the utility to better understand the operation of the equipment, discover hidden failure modes, and take preventive maintenance measures to reduce maintenance costs and equipment downtime.
While data-driven based artificial intelligence techniques have significant advantages in the field of transformer fault diagnosis, there are also potential drawbacks, one of which is that for certain specific fault types, models may be faced with problems of inaccurate diagnosis, especially when the device is in rare abnormal conditions. This is because relying solely on a data-driven model may not adequately understand the complex operating principles and failure mechanisms of a transformer, resulting in a model that is not effective in identifying rare or novel failure modes; in addition, the transformer fault sample data are fewer, and the parameter fitting of the guaranteed model is difficult to meet.
In the prior art, the literature "study on a fault diagnosis method of an oil-immersed transformer based on deep learning, li Kunpeng" applies a convolutional neural network in the deep learning to the fault diagnosis of the transformer on the basis of analyzing the existing fault diagnosis method of the oil-immersed transformer, establishes the mapping between input feature vectors and output fault categories, and further plays the advantages of the deep learning in the fault diagnosis of the transformer. The transformer fault diagnosis model based on the convolutional neural network is constructed, and the accuracy of fault diagnosis is improved. The genetic algorithm is introduced into the model, and the weight and the bias factor of the neural network are subjected to parameter optimization, so that the problem of instability of the model is solved. The literature 'transformer voiceprint recognition technology based on FISVDD and GRU [ J ]. Wang Rong, li, sun Zheng and the like ]. High voltage technology 2022,48 (11): 4546-4556' in order to reduce the false alarm rate of a voiceprint recognition algorithm in a low signal-to-noise ratio environment, a transformer mechanical fault voiceprint recognition method based on Fast Incremental Support Vector Data Description (FISVDD) and a gating cycle unit (GRU) is provided. The samples passing through the class 1 algorithm are identified by separating strange classes using FISVDD as the class 1 algorithm and GRU as the class 2 classification algorithm. Compared with the traditional closed set identification algorithm, the FISVDD greatly reduces training time; compared with the traditional machine learning algorithm, the GRU model improves the fault identification accuracy and the noise resistance identification capability. However, the fault diagnosis method based on deep learning belongs to a black box model, the diagnosis process is unclear, and the interpretation of the diagnosis result is poor. The GRU is an end-to-end diagnosis method, no physical mechanism is considered in the diagnosis process, the reliability and generalization of the model are poor, and the accuracy of the diagnosis result is affected. The above method belongs to a data driving method, and if input data is affected by a change of operating conditions of a transformer or other interference factors, the algorithm may become unstable, resulting in a significant reduction of model performance.
For example, the high-voltage technology 1-9[2023-09-13] aims at solving the problems that the sparrow searching algorithm is seriously homogenized and the classification effect is poor due to unbalance of the transformer fault sample, and provides a transformer fault diagnosis method for optimizing a support vector machine and improving a synthetic few oversampling technology by using the sparrow searching algorithm. Denoising the data set by using a Tomek Link, and introducing a center offset weight so as to obtain a transformer fault data set after balancing treatment; based on the idea of variation, constructing a transformer fault diagnosis model of the VSSA-SVM; the diagnosis is carried out by combining DGA data of 413 oil immersed transformers, and the result shows that the method can accurately judge the fault state of the transformer, effectively solves the problem of low classification precision caused by unbalanced fault data, and has certain engineering practical value. However, in the prior art, the Tomek Link is used to denoise the data set, and the center offset weight is introduced, so that the transformer fault data set after the balancing processing is obtained, and if a few samples contain abnormal samples, a homogeneous oversampling weight is allocated to all the few samples, so that redundant samples and noise labels are generated. Meanwhile, the calculation complexity of using the SVM algorithm is high, and particularly for a large-scale data set and a high-dimensional data set, the calculation time and the calculation space are large. In addition, the training process of the SVM algorithm requires multiple iterations, and the calculation complexity is increased; the SVM algorithm is sensitive to scaling of the data, which may result in bias in the classification result if the data is not normalized.
Disclosure of Invention
The invention aims to solve the problems in the prior art, and provides a transformer fault diagnosis method and system based on a knowledge constraint neural network, which can overcome the defects of a pure data driving method by combining transformer fault knowledge with a deep learning model, ensure the accuracy of common fault phenomenon identification, and realize accurate identification of special fault cases even under the condition of poor data quality or incomplete data without using a large amount of data.
In order to achieve the above purpose, the present invention has the following technical scheme:
in a first aspect, a method for diagnosing a transformer fault based on a knowledge constraint neural network is provided, which is characterized by comprising the following steps:
extracting physical mode information data of multiple transformers, and differentially and jointly inputting the physical mode information data of the multiple transformers into a neural network;
dynamically acquiring characteristic information of different transformer data samples, dynamically evaluating the modal information of the different transformer data samples by using the modal confidence, fusing the neural network which is differentially and jointly input with the physical modal information data of the multiple transformers through a unified multi-modal fusion frame to obtain a fused neural network, and dynamically fusing the characteristic information and the modal information of the different transformer data samples through the fused neural network;
Extracting the mechanism knowledge of the transformer, carrying out mathematical characterization on the mechanism knowledge of the transformer, and embedding the mechanism knowledge of the transformer and a fusion neural network fused with characteristic information and modal information of different transformer data samples to obtain a knowledge constraint neural network;
and performing active early warning training on the knowledge constraint neural network, and performing fault diagnosis by using the trained knowledge constraint neural network.
As a preferable scheme, the step of extracting physical mode information data of multiple transformers, and differentially and jointly inputting the physical mode information data of the multiple transformers into the neural network, wherein the physical mode information data comprises oil chromatogram, optical signals, voiceprint signals, vibration signals and text data; and adding first-order, second-order or multi-order differential signals corresponding to the information data of each physical mode as one type of variables input by the neural network.
As a preferred scheme, in the step of dynamically obtaining the characteristic information of different transformer data samples, dynamically evaluating the modal information of the different transformer data samples by using the modal confidence, fusing the neural network which is differentially and jointly input with the plurality of types of transformer physical modal information data through a unified multi-modal fusion framework to obtain a fused neural network, and dynamically fusing the characteristic information and the modal information of the different transformer data samples through the fused neural network, the method comprises the following steps of:
The modal feature vectors x of different transformer data samples m Input encoder network E m Obtaining feature information vectors Where σ () represents the activation function, information of features is combined using gating strategy, i.e., x m And w is equal to m Dot product gets characteristic information->Thereby preserving information characteristics and suppressingMaking non-information features;
use l 1 The regular loss function improves feature sparsity, and the expression is as follows:
f when using modal confidence to dynamically evaluate modal information for different transformer data samples m Is a modal classifier g m In order to be a trusted regression network,to remove f of the last full tie layer m Network, feature information->Input to f m Wherein a single-mode classification result is obtained->Tag y based on feature information k Obtaining a standard modal confidence TCP which represents the modal information amount; characteristic information +.>Input to->In which the result is input to g m Obtain modality confidence->For the purpose of model output->With TCP m Equal, use l 2 The regular loss function and the cross entropy loss are used for measuring the accuracy of the single-mode classification, and the expression is as follows:
by means ofThe size of each mode weight is distributed, and the larger the value is, the higher the weight is; thenWherein h is multi-mode fusion output, h 1 To h M The weight assigned to each mode is input into a classifier network f to obtain the final classification p k The accuracy of the final classification is measured using cross entropy loss:
the loss function of the fused neural network is as follows:
wherein lambda is 1 Weights, lambda, assigned to regular loss functions 2 Weights assigned to cross entropy loss;
the fused neural network has robustness to data of dynamic changes of characteristic information and modal information.
As a preferred scheme, in the step of extracting the mechanism knowledge of the transformer, performing mathematical characterization on the mechanism knowledge of the transformer, and performing embedded with a fused neural network fused with characteristic information and modal information of different transformer data samples to obtain a knowledge constraint neural network:
creating a dynamic mechanism knowledge database, and gradually supplementing the mechanism knowledge database with experience accumulation;
the fault category outputted by the fused neural network fused with the characteristic information and the modal information of the data samples of the different transformers corresponds to u m M=1, 2, …, M, where M is the fault type and the corresponding probabilities are P respectively m Taking the fault probability as an input quantity, if the transformer mechanism knowledge n corresponds to the fault m, the probability P is calculated m Proceeding withKnowledge embedding processing, the expression is as follows:
P′ m ＝P m +α n ReLU(T n -τ n )
wherein alpha is n For scaling the coefficients, reLU () represents controlling the mechanism feature quantity, T n For the characteristic quantity of different knowledge mechanisms, τ n Is a threshold value corresponding to the characteristic quantity of different knowledge mechanisms. When the model is in a normal value, the model is degenerated into a traditional model, and when the mechanism characteristic quantity exceeds a threshold value, the input probability is influenced, so that the mechanism knowledge of the transformer is introduced into a neural network; thereafter P' m The input is normalized and weighted to the improved softmax function with temperature parameters, and the sharpness of the output probability distribution is adjusted; the expression of the softmax function with temperature parameters is as follows:
wherein T is a temperature parameter, and the smoothness of the model output probability distribution is controlled by adjusting the temperature parameter T.
As a preferred scheme, in the step of performing active early warning training on the knowledge constraint neural network and performing fault diagnosis by using the trained knowledge constraint neural network, a Focal Loss is adopted as a Loss function of active early warning, and the expression of the Focal Loss function is as follows:
FL(P′ m,avg )＝-α m *(1-P′ m,avg ) γ *log(P′ m,avg ))
in the formula, FL (P' m,avg ) Focal Loss function, P 'for fault m' m,avg The fault probability alpha after the fault m is subjected to knowledge embedded processing and normalized and weighted by softmax function n And gamma is an adjustment factor.
As a preferable scheme, in the step of performing active early warning training on the knowledge constraint neural network and performing fault diagnosis by using the trained knowledge constraint neural network, the learning rate is adaptively adjusted by adopting RMSProp under the condition of considering the historical gradient information of each parameter;
RMSProp updates parameters using the update rules of the following formula:
wherein E g 2 ] t Is the moving average of the square gradient at time step t, w t Is the weight of time step t, η is the initial learning rate, and e is a constant for stable calculation.
As a preferable scheme, in the step of performing active early warning training on the knowledge constraint neural network and performing fault diagnosis by using the trained knowledge constraint neural network, an early warning threshold is set, dynamic parameter updating is performed on the trained knowledge constraint neural network in combination with the actual situation of the site, and transformer fault diagnosis active early warning is performed through the knowledge constraint neural network after the dynamic parameter updating.
In a second aspect, a transformer fault diagnosis system based on a knowledge constraint neural network is provided, including:
the combined physical signal input module is used for extracting the physical mode information data of various transformers and differentially and jointly inputting the physical mode information data of various transformers into the neural network;
The multi-mode fusion module is used for dynamically acquiring the characteristic information of different transformer data samples, dynamically evaluating the mode information of the different transformer data samples by utilizing the mode confidence, fusing the neural network which is differentially and jointly input with the data of the physical mode information of the multiple transformers through a unified multi-mode fusion frame to obtain a fused neural network, and dynamically fusing the characteristic information and the mode information of the different transformer data samples through the fused neural network;
the mechanism knowledge embedding module is used for extracting the mechanism knowledge of the transformer, carrying out mathematical characterization on the mechanism knowledge of the transformer, and embedding the mechanism knowledge into a fused neural network fused with characteristic information and modal information of different transformer data samples to obtain a knowledge constraint neural network;
and the fault diagnosis module is used for carrying out active early warning training on the knowledge constraint neural network and carrying out fault diagnosis by utilizing the trained knowledge constraint neural network.
In a third aspect, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the transformer fault diagnosis method based on the knowledge constraint neural network when executing the computer program.
In a fourth aspect, a computer readable storage medium is provided, where a computer program is stored, where the computer program, when executed by a processor, implements the transformer fault diagnosis method based on the knowledge constraint neural network.
Compared with the prior art, the invention has at least the following beneficial effects:
most of the existing transformer fault diagnosis methods are pure data driving methods, however, in an actual production environment, the transformer fault cases are few, fault data are difficult to collect in time, the transformer data set has serious problems of insufficient negative samples, unbalanced data sets and the like, so that the existing fault diagnosis model is difficult to respond to equipment faults in an actual operation environment in time and accurately, and the problems of low precision, poor reliability and the like exist. According to the invention, the transformer overhaul knowledge is embedded into the neural network decision layer to realize data-knowledge hybrid driving, so that the recognition accuracy of the rare fault cases is greatly improved while the recognition accuracy of the historical common fault phenomena is ensured. And the multi-mode monitoring data of the power transformer are effectively fused, and finally reliable health state assessment and fault diagnosis of the transformer are realized. The invention can be applied to a remote monitoring and intelligent inspection system of the transformer, assists a first-line operation inspection personnel in analyzing the operation state of the transformer, and purposefully designates an operation inspection strategy to improve the operation safety and stability of a core main device of a power grid.
It will be appreciated that the advantages of the second to fourth aspects may be found in the relevant description of the first aspect and are not repeated here.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required for the embodiments or the description of the prior art will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a transformer fault diagnosis method based on a knowledge constraint neural network in an embodiment of the invention;
FIG. 2 is a schematic diagram of the principle of obtaining a loss function of each part by fusing a neural network according to the embodiment of the invention;
FIG. 3 is a block diagram of a transformer fault diagnosis system based on a knowledge constraint neural network in an embodiment of the invention;
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system configurations, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.
In addition, in the description of the present application and the appended claims, the terms "first," "second," "third," and the like are used merely to distinguish between descriptions and are not to be construed as indicating or implying relative importance.
The transformer is core junction equipment of the power system, and has important research significance for maintaining long-term stable operation. The operating state of the transformer is directly related to the stable level of the power system, and once a fault occurs, a great economic loss and a plurality of adverse effects are caused to the society. Therefore, the fault cause of the transformer is accurately diagnosed, potential faults are timely found, the risk of faults is reduced through effective preventive measures, hidden trouble of the faults is eliminated, and the power transformer can be ensured to stably operate for a long time.
The existing transformer fault diagnosis methods are mostly pure data driving methods, and have the problems of too few input modes, few fault cases, unbalanced data sets, insufficient data samples and the like, so that a fault judgment model is not comprehensively learned, and the occurrence risk of faults of equipment cannot be accurately judged when rare potential fault phenomena occur, so that the problems of low precision, poor reliability and the like exist. In recent years, knowledge-data fusion driving has remarkable effects in improving the interpretability and robustness of artificial intelligence technology, and provides technical reference for fault diagnosis and analysis of power transformation equipment. The invention provides auxiliary input information which can be used for training a deep learning model by using transformer fault knowledge. Incorporating physical properties into model training may increase the interpretability and interpretability of the model. And the working principle of the transformer can be better understood by the model, and meanwhile, the robustness of the transformer fault diagnosis system is improved. Even if the data quality is poor or the data is incomplete, the model can still rely on the information of the physical characteristics to diagnose, so that the dependence on high-quality data is reduced. By combining the transformer fault knowledge with the deep learning model, the defects of a pure data driving method can be overcome, and the reliability and applicability of transformer fault diagnosis are improved.
The transformer fault diagnosis method based on the knowledge constraint neural network comprises a physical mode information data differential joint input strategy, a data-driven transformer joint active early warning network model construction strategy and an overhaul knowledge embedded strategy. The technical route proposed by the transformer fault diagnosis method of the embodiment of the invention is as follows:
firstly, inputting the data difference of the physical mode information of a plurality of transformers into a neural network in a combined way, extracting and abstracting the information of a plurality of physical modes such as oil chromatography, optical signals, voiceprint signals, vibration signals, text data and the like, and calculating corresponding first-order, second-order and even multi-order differential signals; secondly, constructing a multi-mode network model and realizing dynamic fusion. The method comprises the steps of dynamically acquiring characteristic information of different transformer data samples by using a sparse gating strategy, dynamically evaluating the modal information of the different transformer data samples by using a modal confidence coefficient, and dynamically fusing the characteristic information and the modal information by using a unified multi-modal fusion frame; then, extracting the mechanism knowledge of the transformer, creating a dynamic mechanism knowledge database, realizing the mechanism knowledge characterization of equipment, and finishing the embedment of the overhauling mechanism knowledge of the transformer equipment; and finally, training and reasoning tasks of the transformer active early warning model of the knowledge constraint neural network are completed, and efficient early warning of transformer fault diagnosis is realized. The transformer active early warning method for creating the knowledge constraint neural network solves the problems of low accuracy, poor robustness and Fan Huaxing weakness of the traditional method, and solves the problems of few current fault cases, unbalanced data sets and insufficient data samples.
Referring to fig. 1, a transformer fault diagnosis method based on a knowledge constraint neural network according to an embodiment of the present invention includes:
s1, extracting physical mode information data of multiple transformers, and differentially and jointly inputting the physical mode information data of the multiple transformers into a neural network;
s2, dynamically acquiring characteristic information of different transformer data samples, dynamically evaluating the modal information of the different transformer data samples by using the modal confidence, fusing the neural network which is subjected to differential joint input of various transformer physical modal information data through a unified multi-modal fusion frame to obtain a fused neural network, and dynamically fusing the characteristic information and the modal information of the different transformer data samples through the fused neural network;
s3, extracting mechanism knowledge of the transformer, carrying out mathematical characterization on the mechanism knowledge of the transformer, and embedding the mechanism knowledge of the transformer and a fusion neural network fused with characteristic information and modal information of different transformer data samples to obtain a knowledge constraint neural network;
s4, performing active early warning training on the knowledge constraint neural network, and performing fault diagnosis by using the trained knowledge constraint neural network.
In one possible implementation, the physical mode information data in step S1 includes data such as oil color spectrum, optical signal, voiceprint signal, vibration signal and text, and the data is input into the neural network to be trained according to the mode type. Because potential faults of the transformer often occur along with abnormality of certain physical quantity change rules, according to feedback of first-line transformer inspection personnel and contents recorded in the rules of analysis and judgment of dissolved gas in transformer oil in the power industry standard: the "gas growth rate (gas production rate)" is directly related to the fault energy, the temperature at the fault point, the fault-related range, and the like. The gas production rate is also dependent on the type of equipment, the load conditions and the volume of insulating material used and its degree of personalization. Loss of gas should also be considered when judging severe equipment failure conditions. It should be noted that the gas generation time may be only a certain period of time within the two detection periods, and thus the calculated value of the gas generation rate may be smaller than the actual value "calculated based on:
Absolute gas production rate, i.e. the average value of a certain gas produced per operating day, is calculated as follows:
wherein, gamma a For absolute gas production rate, C i,2 For measuring the concentration of a gas in the oil by sampling for the second time, C i,1 For the first sampling, a certain gas concentration in the oil is measured, Δt is the actual running time in the sub-sampling interval, m is the equipment oil quantity, ρ is the oil density.
The relative gas production rate, i.e., the percentage of the increase in gas content per month of operation (or converted to month) relative to the original value, is calculated as follows:
wherein, gamma τ For relative gas production rate, C i,2 For measuring the concentration of a gas in the oil by sampling for the second time, C i,1 For the first sampling a certain gas concentration in the oil is measured, Δt is the actual run time in the sub-sampling interval.
The embodiment of the invention adds first-order, second-order or multi-order differential signals corresponding to the information data of each physical mode as one type of variable input by the neural network, for example, the following steps:
as one type of input, the rate of change of the physical quantity is represented, where x is the physical input data, m represents the m-th modal input, t is the current time, and i represents the added i-th differential signal. The subsequent calculation will be x m+i Incorporated into x m Is a kind of medium.
In one possible implementation, referring to fig. 2, in step S2, first, feature information of different transformer data samples is dynamically acquired using a sparse gating strategy; secondly, dynamically evaluating the modal information of different transformer data samples by using the modal confidence coefficient; and finally, dynamically fusing the characteristic information and the modal information by using a unified multi-modal fusion framework, so that the model has robustness on the data of the dynamic change of the characteristic information and the modal information. The loss function of the fused neural network is as follows:
Wherein lambda is 1 Weights, lambda, assigned to regular loss functions 2 Weights assigned to cross entropy loss; by combining modal feature vectors x of different transformer data samples m Input encoder network E m Obtaining feature information vectors Where σ () represents the activation function, information of features is combined using gating strategy, i.e., x m And w is equal to m Dot product gets characteristic information->Thereby protectingInformation features are reserved and non-information features are restrained;
use l 1 The regular loss function improves feature sparsity, and the expression is as follows:
f when using modal confidence to dynamically evaluate modal information for different transformer data samples m Is a modal classifier g m In order to be a trusted regression network,to remove f of the last full tie layer m Network, feature information->Input to f m Wherein a single-mode classification result is obtained->Tag y based on feature information k Obtaining a standard modal confidence TCP which represents the modal information amount; characteristic information +.>Input to->In which the result is input to g m Obtain modality confidence->For the purpose of model output->With TCP m Equal, use l 2 The regular loss function and the cross entropy loss are used for measuring the accuracy of the single-mode classification, and the expression is as follows:
When (when)When the value of (2) is low, meaning that the classification is uncertain, the amount of information of the corresponding modality is low and vice versa. By means ofThe size of each mode weight is distributed, and the larger the value is, the higher the weight is; then->Wherein h is multi-mode fusion output, h 1 To h M The weight assigned to each mode is input into a classifier network f to obtain the final classification p k The accuracy of the final classification is measured using cross entropy loss:
in a possible implementation manner, step S3 summarizes and summarizes the mechanism knowledge inside the transformer device by summarizing and generalizing the overhaul experience of the transformer device, then, characterizing the mechanism knowledge, converting the mechanism knowledge into a mathematical form that can be understood by the neural network model, and finally, embedding the mathematical form of the mechanism knowledge of the device and the neural network model after the processing in step S2, so as to realize the active early warning of the transformer of the knowledge-constrained neural network, which specifically comprises the following steps:
first, device mechanism knowledge extraction is performed. According to important reference materials such as first-line transformer inspection personnel feedback and power industry standards of analysis and judgment of dissolved gas in transformer oil, power transformer maintenance guide and the like, the mechanism knowledge of the core transformer shown in the table 1 can be generalized and extracted, wherein a dynamic mechanism knowledge database can be created, and the mechanism knowledge database can be gradually supplemented and completed along with experience accumulation.
TABLE 1 mechanism knowledge database
Then, the knowledge characterization of the mechanism of the device is performed. With the continuous perfection of the mechanism knowledge database, the mechanism knowledge in the database needs to be characterized so as to be embedded into the neural network model.
For the above mechanism knowledge, the embodiment of the present invention performs knowledge characterization according to the following rule.
Let the input core grounding current be I a Mechanism 1 is characterized by:
if|I a |>300 mA-iron core grounding
Set H 2 ，CH 4 ，C 2 H 4 Equal gas growth rate of V X Where X is the type of gas possible, mechanism 2 is characterized by:
mechanism 3 is characterized by:
mechanism 4 is characterized by:
wherein τ 2 ,τ 3 ,τ 4 As super parameters, further parameter adjustment should be performed after subsequent model training and debugging.
Finally, the knowledge of the mechanism of the equipment is embedded. Given the mathematical characterization of the mechanism knowledge, further consideration is required to how the mechanism knowledge is embedded with the neural network. By introducing the knowledge embedded module, the mechanism knowledge and the neural network are embedded, and the active early warning of the combined driving transformer based on the knowledge constraint neural network is realized.
Improvement of the neural network after the processing of the step S2After the probability values of different faults are finally output by the network, a knowledge embedded module is added, and the characteristic quantities of different knowledge mechanisms are unified as T n N=1, 2,3,4, the corresponding threshold value is unified as τ n N=1, 2,3,4, where τ 1 =300, the remainder being hyper-parameters. Thus, the mechanistic knowledge can be unified as:
T n >τ n ,n＝1,2,3,4
step S2, the fault class of the fused neural network output fused with the characteristic information and the modal information of different transformer data samples corresponds to u m M=1, 2, …, M, where M is the fault type and the corresponding probabilities are P respectively m The fault probability is input into the embedded module as an input quantity, and if the transformer mechanism knowledge n corresponds to the fault m, the probability P is calculated m And carrying out knowledge embedded processing, wherein the expression is as follows:
P′ m ＝P m +α n ReLU(T n -τ n )
wherein alpha is n For scaling the coefficients, reLU (·) represents controlling the mechanism feature, T n For the characteristic quantity of different knowledge mechanisms, τ n Is a threshold value corresponding to the characteristic quantity of different knowledge mechanisms. When the model is at a normal value, the model is degraded into a traditional model, however, when the mechanism characteristic quantity exceeds a threshold value, the input probability is influenced, and then the mechanism knowledge of the transformer is introduced into the neural network.
Thereafter P' m The input to the modified softmax function with temperature parameters is normalized and weighted, which adjusts the sharpness of the output probability distribution, thereby providing more control in training the neural network.
The expression of the softmax function with temperature parameters is as follows:
where T is a temperature parameter, typically a positive number. By adjusting the temperature parameter T, the smoothness of the probability distribution of the model output can be controlled. A larger T value will make the probability of all categories more uniform, while a smaller T value will result in the model more confident selection of one or a few categories.
In one possible implementation, step S4 takes Focal Loss as the Loss function of the active pre-warning. The Focal Loss function focuses on the transformer data samples which are difficult to classify by reducing the weight of the transformer data samples which are easy to classify, so that the model performance is improved, and the problem of unbalanced fault classification is solved. The expression of the Focal Loss function is as follows:
FL(P′ m,avg )＝-α m *(1-P′ m,avg ) γ *log(P′ m,avg ))
in the formula, FL (P' m,avg ) Focal Loss function, P 'for fault m' m,avg The fault probability alpha after the fault m is subjected to knowledge embedded processing and normalized and weighted by softmax function m And gamma is an adjustment factor.
And adopting RMSProp as an optimization method of active early warning.
RMSProp (Root Mean Square Propagation) is an optimization algorithm for training neural networks, which aims to solve the problem of learning rate adjustment in gradient descent algorithms. RMSProp improves convergence rate and stability of neural network training by adaptively adjusting the learning rate of each parameter.
By introducing the concept of adaptive learning rate, the learning rate is adaptively adjusted while taking into account the historical gradient information of each parameter. Specifically, RMSProp updates parameters using the update rules of the following formula:
wherein E g 2 ] t Is the moving average of the square gradient at time step t, w t Is the weight of time step t, η is the initial learning rate, and e is a constant for stable calculation.
According to the embodiment of the invention, the training work of the transformer active early warning model of the knowledge constraint neural network is completed after the network parameters are initialized by combining the neural network models established in the steps S1 to S3. Aiming at the existing fault types in the fault sample data set, the fusion neural network in the step S2 can have a good early warning result, but the problems of fewer fault cases, unbalanced data sets, insufficient data samples and the like still exist in the fault sample data set at present. The mechanism knowledge is embedded into the neural network through the step S3, so that the neural network can not only early warn the existing fault types in the fault data set, but also early warn the fault types in which the fault sample data set does not exist or the sample is insufficient through the mechanism knowledge embedded into the neural network.
In a possible implementation manner, step S4 sets an early warning threshold, and performs dynamic parameter update on the trained knowledge constraint neural network in combination with the actual situation of the site, and performs active early warning on transformer fault diagnosis through the knowledge constraint neural network after the dynamic parameter update.
The model parameter adjustment optimization method in the active early warning can be replaced by other adjustment methods, such as an RMSProp optimization algorithm, an L-BFGS optimization algorithm and the like.
The transformer fault diagnosis method based on the knowledge constraint neural network mainly comprises an input strategy of data difference combination of physical mode information, a construction strategy of a transformer combination active early warning network model based on data driving and an embedded strategy of mechanism knowledge. The active early warning criterion expansion of the transformer is realized based on the high-order difference processing data, the high-order difference of the information data such as the transformer oil color spectrum, the temperature and the like is calculated, then the oil color spectrum, the temperature and the difference information are input into the neural network in a combined way, the low-order sensing characteristic is supplemented, and the active early warning criterion expansion of the transformer is realized. Meanwhile, the transformer equipment overhaul knowledge neural network constraint is realized based on mechanism embedment, firstly, transformer overhaul guide rule knowledge is extracted, the knowledge is converted into a mathematical model, and then the mathematical model is embedded into a decision layer of the neural network, so that the constraint of the transformer equipment overhaul knowledge on the neural network is realized.
Referring to fig. 3, a transformer fault diagnosis system based on a knowledge constraint neural network according to an embodiment of the present invention includes:
the combined physical signal input module is used for extracting the physical mode information data of various transformers and differentially and jointly inputting the physical mode information data of various transformers into the neural network;
the multi-mode fusion module is used for dynamically acquiring the characteristic information of different transformer data samples, dynamically evaluating the mode information of the different transformer data samples by utilizing the mode confidence, fusing the neural network which is differentially and jointly input with the data of the physical mode information of the multiple transformers through a unified multi-mode fusion frame to obtain a fused neural network, and dynamically fusing the characteristic information and the mode information of the different transformer data samples through the fused neural network;
the mechanism knowledge embedding module is used for extracting the mechanism knowledge of the transformer, carrying out mathematical characterization on the mechanism knowledge of the transformer, and embedding the mechanism knowledge into a fused neural network fused with characteristic information and modal information of different transformer data samples to obtain a knowledge constraint neural network;
and the fault diagnosis module is used for carrying out active early warning training on the knowledge constraint neural network and carrying out fault diagnosis by utilizing the trained knowledge constraint neural network.
Another embodiment of the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the method for diagnosing a transformer fault based on a knowledge constraint neural network when executing the computer program.
Another embodiment of the present invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for diagnosing a transformer fault based on a knowledge-constrained neural network.
The computer program comprises computer program code which may be in source code form, object code form, executable file or in some intermediate form, etc. The computer readable storage medium may include: any entity or device, medium, usb disk, removable hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier wave signals, telecommunications signals, software distribution media, and the like capable of carrying the computer program code. It should be noted that the computer readable medium contains content that can be appropriately scaled according to the requirements of jurisdictions in which such content is subject to legislation and patent practice, such as in certain jurisdictions in which such content is subject to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals. For convenience of description, the foregoing disclosure shows only those parts relevant to the embodiments of the present invention, and specific technical details are not disclosed, but reference is made to the method parts of the embodiments of the present invention. The computer readable storage medium is non-transitory and can be stored in a storage device formed by various electronic devices, and can implement the execution procedure described in the method according to the embodiment of the present invention.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flowchart and/or block of the flowchart illustrations and/or block diagrams, and combinations of flowcharts and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical aspects of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those of ordinary skill in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which is intended to be covered by the claims.
Claims (10)
1. The transformer fault diagnosis method based on the knowledge constraint neural network is characterized by comprising the following steps of:
extracting physical mode information data of multiple transformers, and differentially and jointly inputting the physical mode information data of the multiple transformers into a neural network;
dynamically acquiring characteristic information of different transformer data samples, dynamically evaluating the modal information of the different transformer data samples by using the modal confidence, fusing the neural network which is differentially and jointly input with the physical modal information data of the multiple transformers through a unified multi-modal fusion frame to obtain a fused neural network, and dynamically fusing the characteristic information and the modal information of the different transformer data samples through the fused neural network;
extracting the mechanism knowledge of the transformer, carrying out mathematical characterization on the mechanism knowledge of the transformer, and embedding the mechanism knowledge of the transformer and a fusion neural network fused with characteristic information and modal information of different transformer data samples to obtain a knowledge constraint neural network;
and performing active early warning training on the knowledge constraint neural network, and performing fault diagnosis by using the trained knowledge constraint neural network.
2. The transformer fault diagnosis method based on the knowledge constraint neural network according to claim 1, wherein the step of extracting a plurality of transformer physical mode information data, wherein the plurality of transformer physical mode information data are differentially and jointly input into the neural network, and the physical mode information data include oil color spectrum, optical signals, voiceprint signals, vibration signals and text data; and adding first-order, second-order or multi-order differential signals corresponding to the information data of each physical mode as one type of variables input by the neural network.
3. The transformer fault diagnosis method based on knowledge constraint neural network according to claim 1, wherein in the step of dynamically obtaining characteristic information of different transformer data samples, dynamically evaluating the modal information of the different transformer data samples by using a modal confidence, fusing the neural network, which is differentially and jointly input with a plurality of kinds of transformer physical modal information data, through a unified multi-modal fusion framework to obtain a fused neural network, and dynamically fusing the characteristic information and the modal information of the different transformer data samples through the fused neural network, the method is characterized in that:
the modal feature vectors x of different transformer data samples m Input encoder network E m Obtaining feature information vectors Where σ () represents the activation function, information of features is combined using gating strategy, i.e., x m And w is equal to m Dot product gets characteristic information->Thereby preserving information features and suppressing non-information features;
use l 1 The regular loss function improves feature sparsity, and the expression is as follows:
f when using modal confidence to dynamically evaluate modal information for different transformer data samples m Is a modal classifier g m In order to be a trusted regression network,to remove f of the last full tie layer m Network, feature information->Input to f m Wherein a single-mode classification result is obtained->Tag y based on feature information k Obtaining a standard modal confidence TCP which represents the modal information amount; characteristic information +.>Input to->In which the result is input to g m Obtain modality confidence->For the purpose of model output->With TCP m Equal, use l 2 The regular loss function and the cross entropy loss are used for measuring the accuracy of the single-mode classification, and the expression is as follows:
by means ofThe size of each mode weight is distributed, and the larger the value is, the higher the weight is; thenWherein h is multi-mode fusion output, h 1 To h M The weight assigned to each mode is input into a classifier network f to obtain the final classification p k The accuracy of the final classification is measured using cross entropy loss:
the loss function of the fused neural network is as follows:
wherein lambda is 1 Weights, lambda, assigned to regular loss functions 2 Weights assigned to cross entropy loss;
the fused neural network has robustness to data of dynamic changes of characteristic information and modal information.
4. The transformer fault diagnosis method based on the knowledge constraint neural network according to claim 1, wherein in the step of extracting the transformer mechanism knowledge, the transformer mechanism knowledge is mathematically represented and embedded with a fusion neural network which fuses characteristic information and modal information of different transformer data samples, so as to obtain the knowledge constraint neural network, wherein the method comprises the steps of:
Creating a dynamic mechanism knowledge database, and gradually supplementing the mechanism knowledge database with experience accumulation;
the fault category outputted by the fused neural network fused with the characteristic information and the modal information of the data samples of the different transformers corresponds to u m M=1, 2,..m, where M is the fault type and the corresponding probabilities are P, respectively m Taking the fault probability as an input quantity, if the transformer mechanism knowledge n corresponds to the fault m, the probability P is calculated m And carrying out knowledge embedded processing, wherein the expression is as follows:
P′ m ＝P m +α nReLU (T n -τ n )
wherein alpha is n For scaling the coefficients, reLU (·) represents controlling the mechanism feature, T n For the characteristic quantity of different knowledge mechanisms, τ n A threshold value corresponding to the characteristic quantity of different knowledge mechanisms; when the model is in a normal value, the model is degenerated into a traditional model, and when the mechanism characteristic quantity exceeds a threshold value, the input probability is influenced, so that the mechanism knowledge of the transformer is introduced into a neural network; thereafter P' m The input is normalized and weighted to the improved softmax function with temperature parameters, and the sharpness of the output probability distribution is adjusted; the expression of the softmax function with temperature parameters is as follows:
wherein T is a temperature parameter, and the smoothness of the model output probability distribution is controlled by adjusting the temperature parameter T.
5. The transformer fault diagnosis method based on the knowledge constraint neural network according to claim 4, wherein in the step of performing active early warning training on the knowledge constraint neural network and performing fault diagnosis by using the trained knowledge constraint neural network, focalLoss is adopted as a loss function of active early warning, and the expression of the FocalLoss loss function is as follows:
FL(P′ m，avg )＝-α m *(1-P′ m，avg ) γ *log(P′ m，avg ))
in the formula, FL (P' m，avg ) Focal Loss function, P 'for fault m' m，avg The fault probability alpha after the fault m is subjected to knowledge embedded processing and normalized and weighted by softmax function m And gamma is an adjustment factor.
6. The transformer fault diagnosis method based on the knowledge constraint neural network according to claim 5, wherein in the step of performing active early warning training on the knowledge constraint neural network and performing fault diagnosis by using the trained knowledge constraint neural network, RMSProp is adopted to adaptively adjust the learning rate under the condition of considering the historical gradient information of each parameter;
RMSProp updates parameters using the update rules of the following formula:
wherein E g 2 ] t Is the moving average of the square gradient at time step t, w t Is the weight of time step r, η is the initial learning rate, and e is a constant for stable calculation.
7. The transformer fault diagnosis method based on the knowledge constraint neural network according to claim 6, wherein in the step of performing active early warning training on the knowledge constraint neural network and performing fault diagnosis by using the trained knowledge constraint neural network, an early warning threshold is set, dynamic parameter updating is performed on the trained knowledge constraint neural network in combination with on-site actual conditions, and active early warning of transformer fault diagnosis is performed by using the knowledge constraint neural network after the dynamic parameter updating.
8. A transformer fault diagnosis system based on a knowledge constrained neural network, comprising:
the combined physical signal input module is used for extracting the physical mode information data of various transformers and differentially and jointly inputting the physical mode information data of various transformers into the neural network;
the multi-mode fusion module is used for dynamically acquiring the characteristic information of different transformer data samples, dynamically evaluating the mode information of the different transformer data samples by utilizing the mode confidence, fusing the neural network which is differentially and jointly input with the data of the physical mode information of the multiple transformers through a unified multi-mode fusion frame to obtain a fused neural network, and dynamically fusing the characteristic information and the mode information of the different transformer data samples through the fused neural network;
The mechanism knowledge embedding module is used for extracting the mechanism knowledge of the transformer, carrying out mathematical characterization on the mechanism knowledge of the transformer, and embedding the mechanism knowledge into a fused neural network fused with characteristic information and modal information of different transformer data samples to obtain a knowledge constraint neural network;
and the fault diagnosis module is used for carrying out active early warning training on the knowledge constraint neural network and carrying out fault diagnosis by utilizing the trained knowledge constraint neural network.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized by: the processor, when executing the computer program, implements the transformer fault diagnosis method based on the knowledge constraint neural network as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that: the computer program, when executed by a processor, implements the transformer fault diagnosis method based on the knowledge constraint neural network as claimed in any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311552523.2A CN117574264A (en) | 2023-11-20 | 2023-11-20 | Transformer fault diagnosis method and system based on knowledge constraint neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311552523.2A CN117574264A (en) | 2023-11-20 | 2023-11-20 | Transformer fault diagnosis method and system based on knowledge constraint neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117574264A true CN117574264A (en) | 2024-02-20 |
Family
ID=89883869
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311552523.2A Pending CN117574264A (en) | 2023-11-20 | 2023-11-20 | Transformer fault diagnosis method and system based on knowledge constraint neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117574264A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117851897A (en) * | 2024-03-08 | 2024-04-09 | 国网山西省电力公司晋城供电公司 | Multi-dimensional feature fusion oil immersed transformer online fault diagnosis method |
-
2023
- 2023-11-20 CN CN202311552523.2A patent/CN117574264A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117851897A (en) * | 2024-03-08 | 2024-04-09 | 国网山西省电力公司晋城供电公司 | Multi-dimensional feature fusion oil immersed transformer online fault diagnosis method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11474163B2 (en) | Power transformer winding fault positioning method based on deep convolutional neural network integrated with visual identification | |
| CN109459993B (en) | Online adaptive fault monitoring and diagnosing method for process industrial process | |
| CN110763997A (en) | Early fault early warning method for synchronous motor stator | |
| CN117574264A (en) | Transformer fault diagnosis method and system based on knowledge constraint neural network | |
| CN113656938A (en) | Monitoring method, control device and storage medium for distribution network transformer | |
| Xu et al. | Quantile regression neural network‐based fault detection scheme for wind turbines with application to monitoring a bearing | |
| CN110012009B (en) | Internet of things intrusion detection method based on combination of decision tree and self-similarity model | |
| CN113420509A (en) | Wind turbine state evaluation method and device and storage medium | |
| CN111291918B (en) | Rotating machine degradation trend prediction method based on stationary subspace exogenous vector autoregression | |
| CN115481726A (en) | Industrial robot complete machine health assessment method and system | |
| CN115221233A (en) | Transformer substation multi-class live detection data anomaly detection method based on deep learning | |
| CN113742993A (en) | Method, device, equipment and storage medium for predicting life loss of dry-type transformer | |
| CN110956112B (en) | Novel high-reliability slewing bearing service life assessment method | |
| CN117350897A (en) | Method and system for predicting conversion power by considering air humidity | |
| CN116596301A (en) | Cable multi-state variable evaluation model design method and device based on zero-order learning method | |
| CN116896093A (en) | Online analysis and optimization method for grid-connected oscillation stability of wind farm | |
| CN113341347B (en) | Dynamic fault detection method for distribution transformer based on AOELM | |
| CN116706863A (en) | Transient stability optimization method for seabed direct current system | |
| CN116108385A (en) | Power equipment fault type diagnosis method and system, terminal and storage medium | |
| CN115578084A (en) | Wind turbine generator set frequency converter fault early warning method based on deep convolution self-encoder | |
| CN113887571A (en) | Electronic equipment fault prediction method for improving SVR algorithm | |
| CN115456041A (en) | Equipment fault early warning method and device, computing equipment and storage medium | |
| CN113886592A (en) | Quality detection method for operation and maintenance data of power information communication system | |
| CN110457906B (en) | Intelligent alarm method for network security event | |
| CN114046816A (en) | Sensor signal fault diagnosis method based on lightweight gradient lifting decision tree |
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 |