CN113985156A - Intelligent fault identification method based on transformer voiceprint big data - Google Patents
Intelligent fault identification method based on transformer voiceprint big data Download PDFInfo
- Publication number
- CN113985156A CN113985156A CN202111043620.XA CN202111043620A CN113985156A CN 113985156 A CN113985156 A CN 113985156A CN 202111043620 A CN202111043620 A CN 202111043620A CN 113985156 A CN113985156 A CN 113985156A
- Authority
- CN
- China
- Prior art keywords
- transformer
- voiceprint
- data
- working state
- state
- 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
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000001514 detection method Methods 0.000 claims abstract description 43
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000007781 pre-processing Methods 0.000 claims abstract description 16
- 230000007704 transition Effects 0.000 claims abstract description 14
- 239000013598 vector Substances 0.000 claims description 50
- 230000004913 activation Effects 0.000 claims description 33
- 238000005070 sampling Methods 0.000 claims description 33
- 238000012544 monitoring process Methods 0.000 claims description 21
- 238000011176 pooling Methods 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 18
- 239000000463 material Substances 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 8
- 230000007797 corrosion Effects 0.000 claims description 8
- 238000005260 corrosion Methods 0.000 claims description 8
- 239000003921 oil Substances 0.000 claims description 7
- 229910000976 Electrical steel Inorganic materials 0.000 claims description 3
- 238000004891 communication Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000004804 winding Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 10
- 238000000547 structure data Methods 0.000 abstract 1
- 230000006870 function Effects 0.000 description 24
- 238000004458 analytical method Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000013508 migration Methods 0.000 description 2
- 230000005012 migration Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004883 computer application Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000004868 gas analysis Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/08—Locating faults in cables, transmission lines, or 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
-
- 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
- Y04S10/52—Outage or fault management, e.g. fault detection or location
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Acoustics & Sound (AREA)
- Human Computer Interaction (AREA)
- Evolutionary Computation (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Signal Processing (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Testing Electric Properties And Detecting Electric Faults (AREA)
Abstract
The invention discloses an intelligent fault identification method based on transformer voiceprint big data, which is used for solving the technical problems that the number of transformer equipment in a power supply and distribution network is large, the fault loss is huge, and the fault detection and prediction cannot be timely and effectively realized. The method comprises the following steps: the method comprises the steps that voiceprint data of a transformer which operates in real time and basic structure data of the transformer are subjected to preprocessing operation such as classification and combination, voiceprint database information is converted into state data of a multi-dimensional space, the state data of the multi-dimensional space are further fitted into curves of multi-section state transition, the curves are used as input training samples in an AlexNet convolutional neural network, a convolutional neural network model is obtained after training is conducted, the real-time online working state of the transformer is calculated and output through the convolutional neural network model, and various fault states are judged. The invention realizes the on-line voiceprint fault detection of the transformer based on the deep learning network, and effectively improves the accuracy and the real-time performance of the fault prediction of the transformer.
Description
Technical Field
The invention relates to the field of data analysis and computer application, in particular to an intelligent fault identification method based on transformer voiceprint big data.
Background
The transformers are important equipment in power transmission, transformation, power supply and distribution systems, are very large in quantity, and have more than ten million transformers in China, and millions of transformers need to be updated and maintained every year, and the safe, reliable and stable operation of the transformers is the basis of the safe, reliable and stable operation of the power supply network. In recent years, with the rapid development of national economy of China, higher requirements are put forward on the quality and the quantity of power supply and distribution networks. In the traditional transformer fault detection method, manual data sampling and online parameter sampling are combined, and maintenance personnel judge the running state of the transformer according to the sampled data. However, manual judgment generally only combines the sampling data at the current moment, so that predictability is lacked, faults cannot be effectively predicted, and inestimable loss may be caused. Therefore, it is very important to introduce computer online detection and auxiliary analysis. Deep learning has become a very critical research field in machine learning and has achieved many subversive research results. The method has very important practical significance for analyzing equipment faults by adopting a deep learning method aiming at the prediction model of the slow fault change and the lack of fault development transformation of the transformer and the like.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides an intelligent fault identification method based on transformer voiceprint big data, and solves the technical problems that the number of transformer equipment in the existing power supply and distribution network is large, the fault loss is huge, and the fault detection and prediction cannot be timely and effectively realized.
The technical scheme of the invention is realized as follows:
an intelligent fault identification method based on transformer voiceprint big data comprises the following steps:
the method comprises the following steps: acquiring a monitoring sample of the physical parameters of the body of the transformer, and preprocessing the monitoring sample of the physical parameters of the body to obtain a body characteristic vector X;
step two: acquiring a monitoring sample of the operation parameter of the transformer, and preprocessing the monitoring sample of the operation parameter to obtain a detection characteristic vector Z;
step three: combining the body characteristic vector X detection characteristic vector Z and the working state of the transformer to be used as a characteristic vector Y';
step four: determining a frame interval to construct a density image of the feature vector Y ', and constructing a working state kernel according to the density image of the feature vector Y';
step five: aiming at the working state core, setting a time slice for the working state transition of the transformer according to the speed of parameter change when the working state of the transformer fails, and fitting a density image of the working state transition of the transformer in the time slice into a fitting curve; the linearity of the fitting curve is one of a straight line, a quadratic curve or a cubic curve;
step six: combining the working states of the transformers with different time periods respectively, processing the combined working states in the operation modes from the first step to the fifth step to obtain fitting curves of the working state transition of the M transformers, marking the M fitting curves by using the working states respectively, and combining the marked M fitting curves with the working states to form a data set, wherein the data set comprises a training set and a test set;
step seven: inputting the training set into an AlexNet convolutional neural network for training to obtain an AlexNet convolutional neural network model;
step eight: and inputting the test set into the AlexNet convolutional neural network model for detection to verify the detection rate, and simultaneously transferring data in the test set into a training set to update the AlexNet convolutional neural network model in real time, so that the detection rate of the AlexNet convolutional neural network model is improved.
Preferably, the physical parameters of the transformer body comprise transformer material, and the transformer material refers to silicon steel sheet dimension material, winding mode, insulating part material or transformer oil; the transformer oil comprises H2、CH4、C2H6、C2H4And C2H2And fifthly, gas.
Preferably, the method for preprocessing the monitoring sample of the bulk physical parameter comprises the following steps:
all parameters included by the ontology physical parameters are combined to construct a feature vector Y ═ Y1,y2,…,yn) Wherein, y1,y2,…,ynRespectively are direct values or serial numbers of parameters in the physical parameters of the body;
converting absolute content of parameters in the physical parameters of the body into relative content:
where i is 1,2, …, n, j is 2,3, …, n, i.e. the feature vector Y is converted into a relative feature vector X (X is ═ 2,3, …, n)1,x2,…,xn)。
Preferably, the operating parameters of the transformer include voiceprint data, a primary side electrical parameter and a secondary side electrical parameter of the transformer; the primary side electrical parameter and the secondary side electrical parameter both comprise current, voltage, active power, reactive power and temperature of key points at two sides of the transformer.
Preferably, the method for preprocessing the monitoring sample of the operation parameter comprises the following steps:
acquiring, denoising, filtering and smoothing the voiceprint data of the transformer by using transformer voiceprint detection equipment; then, performing fast Fourier change on the processed voiceprint data, and extracting three frequency signals with the maximum local power as characteristic parameters of the voiceprint data, wherein the three frequency signals are respectively the characteristic frequency 1, the intensity of the characteristic frequency 1, the duration of the characteristic frequency 1, the characteristic frequency 2, the intensity of the characteristic frequency 2, the duration of the characteristic frequency 2, the characteristic frequency 3, the intensity of the characteristic frequency 3 and the duration of the characteristic frequency 3;
dividing the current, the voltage, the active power and the reactive power at two sides of the transformer, which are included by the secondary side electrical parameters, by the transformer capacity respectively to serve as relative sampling values;
combining the characteristic parameters and the relative sampling values of the voiceprint data to construct a characteristic vector Z, which is expressed as: z ═ Z1,z2,…,z13)。
Preferably, the working state of the transformer comprises a fault state, a normal working state, an early warning state and an alarm state; the fault state comprises a high-temperature overheat state, a medium-temperature overheat state, a low-temperature overheat state, a high-energy discharge state, a low-energy discharge state and a partial discharge state.
Preferably, the method for constructing the density image of the feature vector Y' is as follows: converting the characteristic vector of the transformer into a density image of a multidimensional space with the sampling data as a coordinate axis, wherein the pixel interval of the density image of the multidimensional space represents the resolution of each sampling data, the frame interval represents the sampling time period of certain data determined according to needs, all the sampling data in the time period are used as a density image frame, the coordinate of an image point represents the sampling value of each related quantity, and the occurrence frequency of the sampling value is used as the value of the image point; and displaying the operation data of a plurality of transformers of the same type in all operation periods in the same graph to form a full-state space operation graph of the transformer operation, namely a density image of the characteristic vector Y'.
Preferably, the method for constructing the working state kernel according to the density image of the feature vector Y' includes:
a) taking the area with the maximum density in the density image of the extracted feature vector Y' as a working state kernel contour;
b) filling some small holes of the nuclear outline in the working state by using an eight-neighborhood expansion and corrosion method;
c) carrying out corrosion operation on the filled working state nuclear contour, and cutting other states adhered to the working state nuclear contour;
d) and c, performing expansion operation on the working state kernel contour in the step c to obtain a working state kernel.
Preferably, the structure of the AlexNet convolutional neural network comprises five convolutional layers, three fully-connected layers and seven activation function layers; the first convolution layer is connected with the first activation function layer, the first activation function layer is connected with the first maximum pooling layer, the first maximum pooling layer is connected with the second convolution layer, the second convolution layer is connected with the second activation function layer, the second activation function layer is connected with the second maximum pooling layer, the second maximum pooling layer is connected with the third convolution layer, the third convolution layer is connected with the third activation function layer, the third activation function layer is connected with the fourth convolution layer, the fourth convolution layer is connected with the fourth activation function layer, the fourth activation function layer is connected with the fifth convolution layer, the fifth convolution layer is connected with the sixth activation function layer, the sixth activation function layer is connected with the third maximum pooling layer, the third maximum pooling layer is connected with the first full-link layer, the first full-link layer is connected with the second full-link layer, and the second full-link layer is connected with the third full-link layer, the third full link layer is connected with the softmax classifier.
An intelligent fault identification system based on transformer voiceprint big data comprises a voiceprint detection module, an instrument detection module and a monitoring center; the voiceprint detection module and the instrument detection module are connected with the monitoring center through the communication module; a data processing module and a storage module are arranged in the monitoring center, and the data processing module is connected with the storage module; the data processing module is connected with the voiceprint detection module and the instrument detection module respectively, an AlexNet convolutional neural network model is embedded in the data processing module, and the real-time online working state of the transformer is calculated and output by using the AlexNet convolutional neural network model.
Compared with the prior art, the invention has the following beneficial effects: the invention provides multivariable multi-period sampling data preprocessing, and uses a large amount of multi-period sampling data to train a deep learning network, so as to obtain a transformer fault prediction model to predict the fault state of the transformer, realize the on-line voiceprint fault detection of the transformer based on the deep learning network, and effectively improve the accuracy and the real-time performance of the fault prediction of the transformer.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a "density image" of a certain sampling period.
Fig. 3 is a full-time "density image".
FIG. 4 is a curve fit for continuous state space migration.
Fig. 5 is a structural diagram of a deep learning network.
Fig. 6 is a structural diagram of a transformer state monitoring system based on a deep learning training model.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without inventive effort based on the embodiments of the present invention, are within the scope of the present invention.
the method comprises the following steps: acquiring a monitoring sample of the physical parameters of the body of the transformer, and preprocessing the monitoring sample of the physical parameters of the body to obtain a body characteristic vector X; the physical parameters of the transformer body comprise transformer material, wherein the transformer material refers to silicon steel sheet dimension material, winding mode, insulating part material or transformer oil; the transformer oil comprises H2、CH4、C2H6、C2H4And C2H2And fifthly, gas. Experience shows that early faults inside the transformer can be found in time by using the parameters, and accident potential is eliminated. By utilizing the dissolved gas analysis technology, the parameters of the transformer can be accurately measured, and the running state of the transformer body can be effectively reflected; in order to facilitate analysis and operation, all parameters included by the ontology physical parameters are combined to construct a feature vector Y ═ Y1,y2,…,yn) Wherein, y1,y2,…,ynRespectively, the direct numerical value or the number of the parameter in the physical parameter of the body.
In addition, because the parameters of the transformers with different capacities and voltage levels have larger difference, the absolute content of the parameters needs to be converted into relative content so as to reduce errors, and a specific formula for converting the absolute content of the parameters in the physical parameters of the body into the relative content is as follows:
where i is 1,2, …, n, j is 2,3, …, n, i.e. the feature vector Y is converted into a relative feature vector X (X is ═ 2,3, …, n)1,x2,…,xn). Therefore, the parameter difference of the transformers with different capacities and voltage levels can be eliminated by describing the state of the transformer body by using the relative characteristic vector X.
Step two: acquiring a monitoring sample of the operation parameter of the transformer, and preprocessing the monitoring sample of the operation parameter to obtain a detection characteristic vector Z; the operation parameters of the transformer comprise voiceprint data, primary side electrical parameters and secondary side electrical parameters of the transformer; the primary side electrical parameter and the secondary side electrical parameter comprise current, voltage, active power, reactive power and temperature of key points at two sides of the transformer; and obtaining the information such as dielectric loss, capacitance and the like by operating and processing the digital signal.
The method for preprocessing the monitoring sample of the operation parameter comprises the following steps:
acquiring, denoising, filtering and smoothing the voiceprint data of the transformer by using transformer voiceprint detection equipment; and then, carrying out fast Fourier change on the processed voiceprint data, and extracting three frequency signals with the maximum local power as characteristic parameters of the voiceprint data, wherein the three frequency signals are respectively the characteristic frequency 1, the intensity of the characteristic frequency 1, the duration of the characteristic frequency 1, the characteristic frequency 2, the intensity of the characteristic frequency 2, the duration of the characteristic frequency 2, the characteristic frequency 3, the intensity of the characteristic frequency 3 and the duration of the characteristic frequency 3.
When the voiceprint data and the electrical parameter data are abnormally fluctuated, the system automatically gives an alarm to remind maintenance or safety supervision personnel to timely deal with problems and avoid accidents. Similarly, in order to facilitate analysis and operation and eliminate data redundancy, the primary electricity measurement parameters are ignored, and the current, the voltage, the active power and the reactive power of the two sides of the transformer, which are included in the secondary side electricity parameters, are respectively divided by the capacity of the transformer to serve as relative sampling values, so that the difference of the capacity of the transformer is eliminated.
Combining the characteristic parameters and the relative sampling values of the voiceprint data to construct a characteristic vector Z, which is expressed as: z ═ Z1,z2,…,z13) Which is used to describe the electrical operating state of the transformer. In addition, the temperature of key points of the transformer is used as a parameter to describe the working state of the transformer.
Step three: combining the body characteristic vector X detection characteristic vector Z and the working state of the transformer to be used as a characteristic vector Y'; the working state of the transformer can be described by two vectors X, Z and oil temperature and other parameters. The working state of the transformer comprises a fault state, a normal working state, an early warning state and an alarming state; the fault state comprises a high-temperature overheat (higher than 700 ℃), a medium-temperature overheat (lower than 700 ℃ and higher than 300 ℃), a low-temperature overheat (lower than 300 ℃), a high-energy discharge state, a low-energy discharge state and a partial discharge state. The two states of the early warning state and the alarming state refer to the rest space of the transformer state except the transformer fault state, and the purpose of ensuring the complete sealing of the transformer state space is achieved.
Step four: determining a frame interval to construct a density image of the feature vector Y ', and constructing a working state kernel according to the density image of the feature vector Y'; in order to perform spatial data analysis on the working state of the transformer by using an AlexNet deep convolution neural network algorithm, the data needs to be preprocessed, and the three steps are as follows: extracting a density image of a working state core of the transformer, checking 7 large-state cores, and performing curve fitting on a continuous state space migration image developed by fault generation. The method for constructing the density image of the feature vector Y' comprises the following steps: converting the characteristic vector of the transformer into a density image of a multidimensional space with the sampling data as a coordinate axis, wherein the pixel interval of the density image of the multidimensional space represents the resolution of each sampling data, the frame interval (monitoring period) represents the sampling time period of certain data determined according to needs, all the sampling data in the time period are used as a density image frame, the coordinate of an image point represents the sampling value of each related quantity, and the occurrence frequency of the sampling value is used as the value of the image point; for example (X)1, X2) are two parameters to be sampled, a function can be used in the X1-X2 coordinate system
The value of (b) represents a function of the number of occurrences of the sampling result (x1, x2), also referred to as the density, and the number of sampling values that do not occur in the x1-x2 coordinate space, i.e., the residual space points
The value is 0, i.e. the density is 0. The "density image" of the sampled values of the two parameters over a certain time in the x1-x2 coordinate system is shown (x1, x 2). By the above conversion, the transformer fault detection problem is converted into a spatial cluster of multidimensional spatial "density images". In addition, the sampling data can be continuously acquired in the power grid real-time monitoring system, and the acquisition parameters of the transformers of the same type can be mutually universal after the processing, so that massive multidimensional space density image data can be obtained in the power grid monitoring system. And displaying the operation data of a plurality of transformers of the same type in all operation periods in the same graph to form a full-state space operation graph of the transformer operation, namely a density image of the characteristic vector Y'. As shown in fig. 2 and 3.
The first step of data preprocessing is the extraction of the transformer operating state kernel. The invention mainly takes the interest of the working state kernel part of the transformer, so the working state kernel is extracted from the original working state image to be analyzed, and the working state data of the transformer can be effectively filtered. In the process of extracting the working state kernel, a method of image contour corrosion is adopted to remove other interference data adhered on the working state kernel.
The method for constructing the working state kernel according to the density image of the feature vector Y' comprises the following steps:
a) taking the area with the maximum density in the density image of the extracted feature vector Y' as a working state kernel contour;
b) filling some small holes of the nuclear outline in the working state by using an eight-neighborhood expansion and corrosion method;
c) carrying out corrosion operation on the filled working state nuclear contour, and cutting other states adhered to the working state nuclear contour;
d) and c, performing expansion operation on the working state kernel contour in the step c to obtain a working state kernel.
Wherein, the erosion operation and the expansion operation in the steps c and d have the same operator.
The second step of data preprocessing is to check 6 large failing state cores out of 7 state cores. After the preliminary working state kernel is obtained, the number of the fault state kernels must be checked, that is, the limit value of each parameter is checked, if the limit value of each parameter is contained in the fault state kernel, the fault state kernel remains unchanged, and if the limit values of some parameters are outside the fault state kernel, the fault state kernel is readjusted to contain the limit values of the parameters, so that the description image of the fault state kernel is ensured to meet the actual requirement. The adjustment method of the fault state kernel comprises the methods of region combination, expansion corrosion, boundary fitting of point-surface images and the like.
After 7 fault state kernels are obtained, the state images except the normal state and the 6 large fault state kernels are classified into an early warning state and an alarming state, and are divided into a multi-stage early warning state and an alarming state according to actual needs, so that more states can be generated.
Step five: aiming at the working state core, setting a time slice for the working state transition of the transformer according to the speed of parameter change when the working state of the transformer fails, and fitting a density image of the working state transition of the transformer in the time slice into a fitting curve; the fitting curve is a straight line, a quadratic curve or a cubic curve. After 6 fault state kernels, normal state kernels, multi-stage early warning state kernels and alarm state kernels are obtained, time slices for transformer working state transition can be set according to the speed of parameter change when the transformer working state fault occurs, transformer working state transition images in the time slices are fitted into a simple curve, the curve can be fitted into a straight line, a quadratic curve or a cubic curve as required, as shown in fig. 4, the curve is used as the minimum description of state transition, and the curve is used as the input parameter of the deep learning network.
Step six: in order to increase the number of training sets, the working states of the transformer are respectively combined with different time periods, namely, one group of working states corresponds to a plurality of time periods, the working states can be divided into data related to faults and data unrelated to the faults, the data are processed through the operation modes from the first step to the fifth step to obtain fitting curves of M transformer working state transitions, the M fitting curves are respectively marked by the working states, the marked M fitting curves are combined with the working states to serve as a data set, and the data set comprises a training set and a test set.
Step seven: inputting the training set into an AlexNet convolutional neural network for training to obtain an AlexNet convolutional neural network model; the AlexNet is a classical open source convolution neural network, and compared with dozens of layers, dozens of layers and even hundreds of layers of the moving rules today, the AlexNet has a very simple network structure; the AlexNet convolutional neural network structure comprises five convolutional layers, three full-connection layers and seven activation function layers, and has nearly sixty million free parameters. The first convolution layer is connected with the first activation function layer, the first activation function layer is connected with the first maximum pooling layer, the first maximum pooling layer is connected with the second convolution layer, the second convolution layer is connected with the second activation function layer, the second activation function layer is connected with the second maximum pooling layer, the second maximum pooling layer is connected with the third convolution layer, the third convolution layer is connected with the third activation function layer, the third activation function layer is connected with the fourth convolution layer, the fourth convolution layer is connected with the fourth activation function layer, the fourth activation function layer is connected with the fifth convolution layer, the fifth convolution layer is connected with the sixth activation function layer, the sixth activation function layer is connected with the third maximum pooling layer, the third maximum pooling layer is connected with the first full-link layer, the first full-link layer is connected with the second full-link layer, and the second full-link layer is connected with the third full-link layer, the third full link layer is connected with the softmax classifier. By utilizing the tensoflow architecture, an AlexNet network is easy to realize, and can be defined into a single Python class for convenient reference. The Alexnet network is realized by adopting the existing universal tenserflow architecture, two independent learning processes are adopted due to the two GPU servers, and the data processing flow is shown in figure 5.
The initial learning rate was set to 0.01 and the iteration was performed with a decay factor of 0.001. The maximum number of iterations is set to 100000, the momentum coefficient is set to 0.9, and the decay weight is 0.0005. After training, a neural network model is obtained, and the causal relationship between the transformer working state transition fitting curve and one of the fault states is judged.
Step eight: and inputting the test set into the AlexNet convolutional neural network model for detection to verify the detection rate, and simultaneously transferring data in the test set into a training set to update the AlexNet convolutional neural network model in real time, so that the detection rate of the AlexNet convolutional neural network model is improved.
The learning data set can adopt transformer operation data of a power bureau within 5-10 years as basic data, a fitting curve related to a fault and a fitting curve unrelated to the fault are obtained after preprocessing and serve as a training data sample set, the data are learned and analyzed through a deep learning network, and a prediction model is obtained through training.
In embodiment 2, after a transformer state prediction model is obtained by using a deep learning algorithm, an intelligent fault identification system based on transformer voiceprint big data can be built, as shown in fig. 6, the intelligent fault identification system comprises a voiceprint detection module, an instrument detection module and a monitoring center; the voiceprint detection module and the instrument detection module are connected with the monitoring center through the communication module; a data processing module and a storage module are arranged in the monitoring center, and the data processing module is connected with the storage module; the data processing module is connected with the voiceprint detection module and the instrument detection module respectively, an AlexNet convolutional neural network model is embedded in the data processing module, the real-time online working state of the transformer is calculated and output by using the AlexNet convolutional neural network model, various fault states are judged, and various fault processing instructions are given. The system is tried in a certain power bureau at present, and the trial result shows that the accuracy and the real-time performance of the fault prediction of the transformer are effectively improved, and the requirement of actual use is basically met.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (10)
1. An intelligent fault identification method based on transformer voiceprint big data is characterized by comprising the following steps:
the method comprises the following steps: acquiring a monitoring sample of the physical parameters of the body of the transformer, and preprocessing the monitoring sample of the physical parameters of the body to obtain a body characteristic vector X;
step two: acquiring a monitoring sample of the operation parameter of the transformer, and preprocessing the monitoring sample of the operation parameter to obtain a detection characteristic vector Z;
step three: combining the body characteristic vector X detection characteristic vector Z and the working state of the transformer to be used as a characteristic vector Y';
step four: determining a frame interval to construct a density image of the feature vector Y ', and constructing a working state kernel according to the density image of the feature vector Y';
step five: aiming at the working state core, setting a time slice for the working state transition of the transformer according to the speed of parameter change when the working state of the transformer fails, and fitting a density image of the working state transition of the transformer in the time slice into a fitting curve; the linearity of the fitting curve is one of a straight line, a quadratic curve or a cubic curve;
step six: combining the working states of the transformers with different time periods respectively, processing the combined working states in the operation modes from the first step to the fifth step to obtain fitting curves of the working state transition of the M transformers, marking the M fitting curves by using the working states respectively, and combining the marked M fitting curves with the working states to form a data set, wherein the data set comprises a training set and a test set;
step seven: inputting the training set into an AlexNet convolutional neural network for training to obtain an AlexNet convolutional neural network model;
step eight: and inputting the test set into the AlexNet convolutional neural network model for detection to verify the detection rate, and simultaneously transferring data in the test set into a training set to update the AlexNet convolutional neural network model in real time, so that the detection rate of the AlexNet convolutional neural network model is improved.
2. The intelligent fault identification method based on the transformer voiceprint big data is characterized in that physical parameters of a body of the transformer comprise a transformer material, and the transformer material refers to a silicon steel sheet size material, a winding mode, an insulating part material or transformer oil; the transformer oil comprises H2、CH4、C2H6、C2H4And C2H2And fifthly, gas.
3. The intelligent fault identification method based on the transformer voiceprint big data according to claim 2, wherein the method for preprocessing the monitoring sample of the body physical parameter comprises the following steps:
all parameters included by the ontology physical parameters are combined to construct a feature vector Y ═ Y1,y2,…,yn) Wherein, y1,y2,…,ynRespectively are direct values or serial numbers of parameters in the physical parameters of the body;
converting absolute content of parameters in the physical parameters of the body into relative content:
where i is 1,2, …, n, j is 2,3, …, n, i.e. the feature vector Y is converted into a relative feature vector X (X is ═ 2,3, …, n)1,x2,…,xn)。
4. The intelligent fault identification method based on the transformer voiceprint big data is characterized in that the operation parameters of the transformer comprise the voiceprint data, the primary side electrical parameters and the secondary side electrical parameters of the transformer; the primary side electrical parameter and the secondary side electrical parameter both comprise current, voltage, active power, reactive power and temperature of key points at two sides of the transformer.
5. The intelligent fault identification method based on the transformer voiceprint big data according to claim 4, wherein the method for preprocessing the monitoring samples of the operation parameters comprises the following steps:
acquiring, denoising, filtering and smoothing the voiceprint data of the transformer by using transformer voiceprint detection equipment; then, performing fast Fourier change on the processed voiceprint data, and extracting three frequency signals with the maximum local power as characteristic parameters of the voiceprint data, wherein the three frequency signals are respectively the characteristic frequency 1, the intensity of the characteristic frequency 1, the duration of the characteristic frequency 1, the characteristic frequency 2, the intensity of the characteristic frequency 2, the duration of the characteristic frequency 2, the characteristic frequency 3, the intensity of the characteristic frequency 3 and the duration of the characteristic frequency 3;
dividing the current, the voltage, the active power and the reactive power at two sides of the transformer, which are included by the secondary side electrical parameters, by the transformer capacity respectively to serve as relative sampling values;
combining the characteristic parameters and the relative sampling values of the voiceprint data to construct a characteristic vector Z, which is expressed as: z ═ Z1,z2,…,z13)。
6. The intelligent fault identification method based on the transformer voiceprint big data is characterized in that the working state of the transformer comprises a fault state, a normal working state, an early warning state and an alarming state; the fault state comprises a high-temperature overheat state, a medium-temperature overheat state, a low-temperature overheat state, a high-energy discharge state, a low-energy discharge state and a partial discharge state.
7. The intelligent fault identification method based on the transformer voiceprint big data according to claim 1, wherein the method for constructing the density image of the feature vector Y' comprises the following steps: converting the characteristic vector of the transformer into a density image of a multidimensional space with the sampling data as a coordinate axis, wherein the pixel interval of the density image of the multidimensional space represents the resolution of each sampling data, the frame interval represents the sampling time period of certain data determined according to needs, all the sampling data in the time period are used as a density image frame, the coordinate of an image point represents the sampling value of each related quantity, and the occurrence frequency of the sampling value is used as the value of the image point; and displaying the operation data of a plurality of transformers of the same type in all operation periods in the same graph to form a full-state space operation graph of the transformer operation, namely a density image of the characteristic vector Y'.
8. The intelligent fault identification method based on the transformer voiceprint big data according to claim 6, wherein the method for constructing the working state kernel according to the density image of the feature vector Y' comprises the following steps:
a) taking the area with the maximum density in the density image of the extracted feature vector Y' as a working state kernel contour;
b) filling some small holes of the nuclear outline in the working state by using an eight-neighborhood expansion and corrosion method;
c) carrying out corrosion operation on the filled working state nuclear contour, and cutting other states adhered to the working state nuclear contour;
d) and c, performing expansion operation on the working state kernel contour in the step c to obtain a working state kernel.
9. The intelligent fault identification method based on the transformer voiceprint big data is characterized in that the AlexNet convolutional neural network comprises five convolutional layers, three full-connection layers and seven activation function layers; the first convolution layer is connected with the first activation function layer, the first activation function layer is connected with the first maximum pooling layer, the first maximum pooling layer is connected with the second convolution layer, the second convolution layer is connected with the second activation function layer, the second activation function layer is connected with the second maximum pooling layer, the second maximum pooling layer is connected with the third convolution layer, the third convolution layer is connected with the third activation function layer, the third activation function layer is connected with the fourth convolution layer, the fourth convolution layer is connected with the fourth activation function layer, the fourth activation function layer is connected with the fifth convolution layer, the fifth convolution layer is connected with the sixth activation function layer, the sixth activation function layer is connected with the third maximum pooling layer, the third maximum pooling layer is connected with the first full-link layer, the first full-link layer is connected with the second full-link layer, and the second full-link layer is connected with the third full-link layer, the third full link layer is connected with the softmax classifier.
10. The intelligent fault identification system based on the transformer voiceprint big data as described in any one of 1-9 is characterized by comprising a voiceprint detection module, an instrument detection module and a monitoring center; the voiceprint detection module and the instrument detection module are connected with the monitoring center through the communication module; a data processing module and a storage module are arranged in the monitoring center, and the data processing module is connected with the storage module; the data processing module is connected with the voiceprint detection module and the instrument detection module respectively, an AlexNet convolutional neural network model is embedded in the data processing module, and the real-time online working state of the transformer is calculated and output by using the AlexNet convolutional neural network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111043620.XA CN113985156A (en) | 2021-09-07 | 2021-09-07 | Intelligent fault identification method based on transformer voiceprint big data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111043620.XA CN113985156A (en) | 2021-09-07 | 2021-09-07 | Intelligent fault identification method based on transformer voiceprint big data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113985156A true CN113985156A (en) | 2022-01-28 |
Family
ID=79735410
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111043620.XA Pending CN113985156A (en) | 2021-09-07 | 2021-09-07 | Intelligent fault identification method based on transformer voiceprint big data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113985156A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114927141A (en) * | 2022-07-19 | 2022-08-19 | 中国人民解放军海军工程大学 | Method and system for detecting abnormal underwater acoustic signals |
| CN114997664A (en) * | 2022-06-10 | 2022-09-02 | 江苏前景瑞信科技发展有限公司 | Transformer abnormity early warning analysis method and system based on convolutional neural network |
| CN117630543A (en) * | 2023-11-30 | 2024-03-01 | 国网湖北省电力有限公司超高压公司 | Transformer fault state supervision method and system based on Internet of things data |
| CN118116732A (en) * | 2024-04-18 | 2024-05-31 | 泰安鑫杰机械有限公司 | Multi-shaft high-speed machining system and method for transformer winding insulating part |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010197124A (en) * | 2009-02-24 | 2010-09-09 | Tokyo Electric Power Co Inc:The | Apparatus, method and program for detecting abnormal noise |
| JP2017106893A (en) * | 2015-11-30 | 2017-06-15 | ユカインダストリーズ株式会社 | Method and device for diagnosing abnormality and deterioration in transformer |
| CN108717554A (en) * | 2018-05-22 | 2018-10-30 | 复旦大学附属肿瘤医院 | A kind of thyroid tumors histopathologic slide image classification method and its device |
| CN109815537A (en) * | 2018-12-19 | 2019-05-28 | 清华大学 | A kind of high-throughput material simulation calculation optimization method based on time prediction |
| CN110376455A (en) * | 2019-06-26 | 2019-10-25 | 深圳供电局有限公司 | Transformer working state detection method and device, computer equipment and storage medium |
| CN110415709A (en) * | 2019-06-26 | 2019-11-05 | 深圳供电局有限公司 | Transformer working state identification method based on voiceprint identification model |
| CN110617982A (en) * | 2019-09-19 | 2019-12-27 | 苏州时辰智能机电设备有限公司 | Rotating machinery equipment fault identification method based on voiceprint signals |
| CN110634493A (en) * | 2019-09-09 | 2019-12-31 | 国网湖南省电力有限公司 | Transformer state identification method, system and medium based on voiceprint image characteristics |
| CN111652927A (en) * | 2020-05-11 | 2020-09-11 | 广东亿云付科技有限公司 | CNN-based cancer cell multi-scale scaling positioning detection method |
| CN112201260A (en) * | 2020-09-07 | 2021-01-08 | 北京科技大学 | Transformer running state online detection method based on voiceprint recognition |
| CN112395959A (en) * | 2020-10-30 | 2021-02-23 | 天合云能源互联网技术(杭州)有限公司 | Power transformer fault prediction and diagnosis method and system based on audio features |
| CN113253156A (en) * | 2021-05-17 | 2021-08-13 | 国网江苏省电力有限公司检修分公司 | Sound monitoring-based latent defect diagnosis method for transformer |
-
2021
- 2021-09-07 CN CN202111043620.XA patent/CN113985156A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010197124A (en) * | 2009-02-24 | 2010-09-09 | Tokyo Electric Power Co Inc:The | Apparatus, method and program for detecting abnormal noise |
| JP2017106893A (en) * | 2015-11-30 | 2017-06-15 | ユカインダストリーズ株式会社 | Method and device for diagnosing abnormality and deterioration in transformer |
| CN108717554A (en) * | 2018-05-22 | 2018-10-30 | 复旦大学附属肿瘤医院 | A kind of thyroid tumors histopathologic slide image classification method and its device |
| CN109815537A (en) * | 2018-12-19 | 2019-05-28 | 清华大学 | A kind of high-throughput material simulation calculation optimization method based on time prediction |
| CN110376455A (en) * | 2019-06-26 | 2019-10-25 | 深圳供电局有限公司 | Transformer working state detection method and device, computer equipment and storage medium |
| CN110415709A (en) * | 2019-06-26 | 2019-11-05 | 深圳供电局有限公司 | Transformer working state identification method based on voiceprint identification model |
| CN110634493A (en) * | 2019-09-09 | 2019-12-31 | 国网湖南省电力有限公司 | Transformer state identification method, system and medium based on voiceprint image characteristics |
| CN110617982A (en) * | 2019-09-19 | 2019-12-27 | 苏州时辰智能机电设备有限公司 | Rotating machinery equipment fault identification method based on voiceprint signals |
| CN111652927A (en) * | 2020-05-11 | 2020-09-11 | 广东亿云付科技有限公司 | CNN-based cancer cell multi-scale scaling positioning detection method |
| CN112201260A (en) * | 2020-09-07 | 2021-01-08 | 北京科技大学 | Transformer running state online detection method based on voiceprint recognition |
| CN112395959A (en) * | 2020-10-30 | 2021-02-23 | 天合云能源互联网技术(杭州)有限公司 | Power transformer fault prediction and diagnosis method and system based on audio features |
| CN113253156A (en) * | 2021-05-17 | 2021-08-13 | 国网江苏省电力有限公司检修分公司 | Sound monitoring-based latent defect diagnosis method for transformer |
Non-Patent Citations (1)
| Title |
|---|
| 等国锋 等: "基于深度学习的变压器在线故障检测", 计算机测量与控制, vol. 28, no. 9, pages 65 - 68 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114997664A (en) * | 2022-06-10 | 2022-09-02 | 江苏前景瑞信科技发展有限公司 | Transformer abnormity early warning analysis method and system based on convolutional neural network |
| CN114927141A (en) * | 2022-07-19 | 2022-08-19 | 中国人民解放军海军工程大学 | Method and system for detecting abnormal underwater acoustic signals |
| CN117630543A (en) * | 2023-11-30 | 2024-03-01 | 国网湖北省电力有限公司超高压公司 | Transformer fault state supervision method and system based on Internet of things data |
| CN118116732A (en) * | 2024-04-18 | 2024-05-31 | 泰安鑫杰机械有限公司 | Multi-shaft high-speed machining system and method for transformer winding insulating part |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113985156A (en) | Intelligent fault identification method based on transformer voiceprint big data | |
| CN108320043B (en) | Power distribution network equipment state diagnosis and prediction method based on electric power big data | |
| CN110262450B (en) | Fault prediction method for cooperative analysis of multiple fault characteristics of steam turbine | |
| CN105718958B (en) | Circuit breaker failure diagnostic method based on linear discriminant analysis and support vector machines | |
| CN116610998A (en) | Switch cabinet fault diagnosis method and system based on multi-mode data fusion | |
| Xu et al. | An unknown fault identification method based on PSO-SVDD in the IoT environment | |
| CN116562114A (en) | Power transformer fault diagnosis method based on graph convolution neural network | |
| CN113611568A (en) | Vacuum circuit breaker based on genetic convolution depth network | |
| CN114487643A (en) | On-spot handing-over of extra-high voltage GIL equipment is accepted and is synthesized test platform | |
| CN116070140B (en) | Power distribution substation safe operation state monitoring system and method | |
| CN108761263A (en) | A kind of fault diagnosis system based on evidence theory | |
| CN115564075B (en) | Main and auxiliary integrated fault collaborative diagnosis method and system for urban power grid | |
| CN114298413B (en) | Hydroelectric generating set runout trend prediction method | |
| CN107622251A (en) | A kind of aircraft fuel pump signal degradation feature extracting method and device | |
| Yang et al. | Trend Analysis and Fault Diagnosis of Equipment State Based on Transformer Operation Mechanism Modeling | |
| Khalyasmaa et al. | Fuzzy inference algorithms for power equipment state assessment | |
| Guo et al. | Construction of Power Equipment Running Status Monitoring System Based on Infrared Temperature Measurement Technology and Big Data Algorithm | |
| Bai et al. | Abnormal Detection Scheme of Substation Equipment based on Intelligent Fusion Terminal | |
| Shuo-jie et al. | Relationship Extraction of Bushing Failure from Chinese Corpus Based on BERT-FC Model | |
| Dunwen et al. | A trainsient voltage stability evaluation model based on morphological similarity distance online calculation | |
| Hou et al. | A hybrid deep learning model approach for performance index prediction of mechanical equipment | |
| Cui et al. | Automatic Early Warning Technology of Automatic Pipeline Fault based on Cluster Analysis | |
| Shanyuan et al. | Research on transformer condition evaluation method based on association rule set pair analysis theory | |
| Long et al. | Substation Automation Equipment State Evaluation Method Based on Comprehensive Analysis of Interval Value Fuzzy Soft Sets | |
| Yang et al. | Converter valve state evaluation method based on hierarchical graded fuzzy theory |
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 |