CN116524200A - High-voltage circuit breaker fault diagnosis method based on image recognition - Google Patents
High-voltage circuit breaker fault diagnosis method based on image recognition Download PDFInfo
- Publication number
- CN116524200A CN116524200A CN202310497416.8A CN202310497416A CN116524200A CN 116524200 A CN116524200 A CN 116524200A CN 202310497416 A CN202310497416 A CN 202310497416A CN 116524200 A CN116524200 A CN 116524200A
- Authority
- CN
- China
- Prior art keywords
- fault diagnosis
- current
- circuit breaker
- voltage circuit
- image
- 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
- 238000003745 diagnosis Methods 0.000 title claims abstract description 109
- 238000000034 method Methods 0.000 title claims abstract description 80
- 238000012549 training Methods 0.000 claims abstract description 61
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 27
- 238000012544 monitoring process Methods 0.000 claims abstract description 12
- 238000007781 pre-processing Methods 0.000 claims abstract description 12
- 238000013507 mapping Methods 0.000 claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 11
- 239000010410 layer Substances 0.000 claims description 88
- 239000013598 vector Substances 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 18
- 238000012360 testing method Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 14
- 238000011176 pooling Methods 0.000 claims description 14
- 238000010586 diagram Methods 0.000 claims description 10
- 238000000354 decomposition reaction Methods 0.000 claims description 8
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000013461 design Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000012986 modification Methods 0.000 claims description 4
- 230000004048 modification Effects 0.000 claims description 4
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 3
- 238000003780 insertion Methods 0.000 claims description 3
- 230000037431 insertion Effects 0.000 claims description 3
- 239000002356 single layer Substances 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 description 10
- 238000000605 extraction Methods 0.000 description 5
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical group [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 4
- 230000004913 activation Effects 0.000 description 4
- 238000002405 diagnostic procedure Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000005728 strengthening Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/327—Testing of circuit interrupters, switches or circuit-breakers
- G01R31/3271—Testing of circuit interrupters, switches or circuit-breakers of high voltage or medium voltage devices
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
- G01R19/0092—Arrangements for measuring currents or voltages or for indicating presence or sign thereof measuring current only
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a high-voltage circuit breaker fault diagnosis method based on image recognition, which relates to the technical field of on-line monitoring and fault diagnosis of electrical equipment, and comprises the following steps: collecting current history data of an operating coil of the high-voltage circuit breaker, preprocessing, converting a one-dimensional time sequence of the preprocessed current history data into a two-dimensional tensor by using a mapping function, and drawing a current image with uniform pixel size; processing the current image to obtain a current gray image, and combining a historical data tag to form a high-voltage circuit breaker fault sample to obtain a high-voltage circuit breaker fault sample library; training a pre-established two-dimensional convolutional neural network model based on a high-voltage circuit breaker fault sample to obtain a fault diagnosis model meeting the precision requirement; and on-line monitoring the current of the operating coil of the high-voltage circuit breaker, converting the current into an on-line current gray level image after data preprocessing, and inputting the on-line current gray level image into a trained fault diagnosis model to obtain a fault diagnosis result.
Description
Technical Field
The invention relates to the technical field of on-line monitoring and fault diagnosis of electrical equipment, in particular to a fault diagnosis method of a high-voltage circuit breaker based on image recognition.
Background
The high-voltage circuit breaker is an important control and protection device in the power system, and the safe and stable operation of the high-voltage circuit breaker plays a vital role in ensuring the safety and reliability of the power system. It mainly plays two roles in the power grid: first, the control action, according to the power grid operation needs, can utilize high voltage circuit breaker to put a part of power equipment or circuit into or withdraw from operation. Secondly, under the protection effect, when the power line or the equipment breaks down, the high-voltage circuit breaker can rapidly cut off the fault part from the power grid, and normal operation of the fault-free part in the power grid is ensured. If the high-voltage circuit breaker fails and cannot reliably act, an accident of a power system is caused, and the circuit and equipment heat is caused by light short-circuit current; if the system is crashed, a large-scale and long-time power failure accident is caused, and the life safety of operation and maintenance personnel is endangered. Therefore, the method has very important significance in ensuring the safe and reliable operation of the circuit breaker, and researches are conducted around the reliability of the high-voltage circuit breaker, so that the normal operation of the high-voltage circuit breaker can be ensured under various working conditions, and the possibility of the occurrence of faults of the circuit breaker is reduced.
At present, aiming at the on-line monitoring and fault diagnosis of the high-voltage circuit breaker, a plurality of research results are already available at home and abroad. In order to accurately identify various breaker fault types, researchers sequentially put forward fault diagnosis methods of signals based on vibration (a high-voltage breaker fault diagnosis method, a high-voltage breaker fault diagnosis system and a high-voltage breaker fault diagnosis device, CN 112083328A), sound (a multi-feature optimization fusion high-voltage breaker fault diagnosis method, CN 112255538A), a travel-time curve (a fuzzy clustering-based support vector machine high-voltage breaker fault diagnosis method, CN 103345639A) and the like, wherein the fault diagnosis method has a complex feature extraction process, and a processing algorithm for on-line monitoring signals is complex.
On-line monitoring and fault diagnosis of high-voltage circuit breakers based on coil current signals are also receiving continuous attention, and fault diagnosis is generally realized by collecting coil current time sequence signals of the circuit breakers and extracting current time sequence characteristics, such as a fault diagnosis method of the high-voltage circuit breakers of a transformer substation based on the coil current signals, CN114298079A; plum fly, mei Jun, zheng Jianyong. Robust diagnosis method for breaker failure based on KPCA-SVM [ J ]. Protect of electrician, 2014,29 (S1): 50-58; huang Xinbo, hu Xiaowen, zhu Yongcan high voltage circuit breaker fault diagnosis based on convolutional neural network algorithm [ J ]. Power Automation device, 2018,38 (05): 136-140+147; guan Yonggang, yang Yuanwei, zhong Jianying. Methods for diagnosing mechanical faults in high voltage circuit breakers [ J ]. High voltage electrical apparatus, 2018,54 (07): 10-19. As shown in fig. 1, the current timing characteristic of the operating coil includes a current peak value, a current peak value time and a current duration time, and since only key characteristic points are concerned, the current information is not fully excavated by the current timing characteristic extraction method, the time course of continuous change of waveforms is ignored, and the characteristic differences possibly existing in the current waveforms under certain conditions, such as the situation of multiple peaks, are not considered, so that the real state of the high-voltage circuit breaker cannot be accurately identified, and the accuracy and generalization performance of a fault diagnosis model are greatly affected. In addition, the extracted time sequence feature has lower dimensionality, can not effectively represent current data containing tens of thousands of sampling points, and has serious information loss.
Disclosure of Invention
In order to solve the defects in the background art, the invention aims to provide a fault diagnosis method for a high-voltage circuit breaker based on image recognition.
The aim of the invention can be achieved by the following technical scheme: a fault diagnosis method of a high-voltage circuit breaker based on image recognition comprises the following steps:
collecting current history data of an operating coil of the high-voltage circuit breaker, preprocessing, converting a one-dimensional time sequence of the preprocessed current history data into a two-dimensional tensor by using a mapping function, and drawing a current image with uniform pixel size;
processing the current image to obtain a current gray image, and combining a historical data tag to form a high-voltage circuit breaker fault sample to obtain a high-voltage circuit breaker fault sample library;
training a pre-established two-dimensional convolutional neural network model based on a high-voltage circuit breaker fault sample to obtain a fault diagnosis model meeting the precision requirement;
and on-line monitoring the current of the operating coil of the high-voltage circuit breaker, converting the current into an on-line current gray level image after data preprocessing, and inputting the on-line current gray level image into a trained fault diagnosis model to obtain a fault diagnosis result.
Preferably, the preprocessing includes denoising the wavelet packet to eliminate high-frequency noise and burrs, and expanding the number of samples by a data enhancement method.
Preferably, the data enhancement method expands the data volume by generating samples close to the current raw data, including time-shifting the current signal, adding random noise.
Preferably, the current image is subjected to amplitude normalization and image graying to obtain a current gray image.
Preferably, the training process of the fault diagnosis model is as follows:
dividing the high-voltage breaker fault samples in the high-voltage breaker fault sample library into a training set and a testing set by using a ten-fold intersection method to respectively obtain ten groups of different data sets, training a two-dimensional convolutional neural network model in sequence, analyzing the performance of a trained network, and adjusting network structure parameters by combining training targets until a network structure meeting the performance requirements is obtained, wherein the network structure and the trained parameters are the trained fault diagnosis model.
Preferably, in the training fault diagnosis model, the pixel sparse feature of the input current gray level image is considered when the structural parameters of the convolutional neural network are initialized, a convolutional kernel with smaller dimension and smaller output channel number are selected in the convolutional layer, a larger pooling area is arranged in the pooling layer, the number of convolutional layers and the number of full connection layers are reduced under the condition that the precision of the fault diagnosis model meets the requirement, so that the parameter number and the complexity of the fault diagnosis model are reduced, the design targets of high fault diagnosis accuracy and light network structure are considered, and a regularization layer is added before the full connection layer so as to prevent the fault diagnosis model from being over fitted.
Preferably, the operating coil current history data is as follows:
wherein I (n) is the time series of the operating coil current signal, I n For time t n Corresponding current value;
decomposing the current history data of the operation coil by utilizing wavelet packet transformation, wherein the wavelet packet decomposition process is expressed as follows:
wherein: i (n) is an acquired operating coil current signal, L is a low-pass filter, H is a high-pass filter, an approximation a is a wavelet coefficient generated by a larger scale factor and represents a low-frequency component of the signal, and a detail d is a wavelet coefficient generated by a smaller scale factor and represents a high-frequency component of the signal;
the Sqtwolog criterion is selected to calculate the magnitude of the noise reduction threshold of the operating coil current history data, formulated as:
wherein:
preferably, performing wavelet reconstruction, calculating the wavelet reconstruction according to the original approximation coefficients of the layer number j and the modification detail coefficients of 1 to j, including: first, the approximation a after the last layer decomposition is utilized j And detail value d j Respectively performing zero insertion at intervals, and then respectively connecting with a filter L 1 And H 1 The convolution operation is carried out to obtain a j-1 layer approximation value a j-1 Repeating the process to obtain the original signal a 0 =i (n). Expressed by the formula:
preferably, the training process of the fault diagnosis model includes the following steps:
initializing a convolutional neural network hyper-parameter, extracting a current gray image sample and a target output vector thereof, namely a sample label, from a training set, sequentially calculating from a front layer to a rear layer to obtain the output of the convolutional neural network, wherein the two-dimensional discrete convolution of a single-layer convolutional layer on an input tensor is as shown in formula (10):
wherein k is the number of input tensor channels, l is the number of output tensor channels, and the length and width of the output characteristic diagram are as shown in formula (11):
wherein s is the moving step length of the convolution kernel, and k is the size of the convolution kernel;
then calculate the cross entropy loss of convolutional neural network output and target vector according to equation (12):
wherein p (x) i ) Q (x) i ) Sample label probability distribution output for neural network forward operation;
updating parameters of each layer in the network by utilizing the gradient descent principle according to the formula (13) until the loss is lower than a set threshold value or the iterative training times reach a preset value, wherein the accuracy of the training set reaches a stable value,
F(n+1)＝F(n)±u·||H(p,q)|| (13)
wherein F represents parameters of each layer in the network, and u is an iteration factor;
finally, whether the losses on the training set and the testing set are converged or not is checked, whether the accuracy meets the requirement of fault diagnosis precision or not, if not, the accuracy is more than 90%, the super parameters of the convolutional neural network are reset, the layer number is adjusted, the above steps are repeated until the evaluation index of the fault diagnosis model meets the requirement, and the training of the fault diagnosis model is completed after the requirement is met.
The invention has the beneficial effects that:
1. the fault diagnosis method overcomes the defects of the existing fault diagnosis method based on the current signal of the operation coil, does not need a complex feature vector extraction process, directly identifies the current image features of the operation coil through computer vision, can fully excavate the information in the current signal of the operation coil, further realizes the effective diagnosis of various fault types of the high-voltage circuit breaker, and has extremely high diagnosis accuracy;
2. according to the data enhancement method in the fault diagnosis method, sampling noise and sampling delay caused by field interference are considered, the number of current image samples is increased by simulating current samples with small interference, a fault sample library of the high-voltage circuit breaker with a large number of complete operation coil current image samples is constructed, the problems that the number of on-line monitoring data samples of the circuit breaker is small and the data samples are difficult to acquire in actual application are effectively solved, the experimental cost caused by acquisition of a large number of samples and the influence on the service life of the circuit breaker are reduced, the requirements of a deep learning model on a large number of data samples are met, and the generalization performance of the model is improved;
3. compared with a classical AlexNet image recognition network, the network structure of the fault diagnosis model in the fault diagnosis method has the advantages of less network parameter quantity, simpler and lighter network structure and suitability for actual application of fault diagnosis of the high-voltage circuit breaker.
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, and it will be obvious to those skilled in the art that other drawings can be obtained according to these drawings without inventive effort;
FIG. 1 is a schematic flow chart of a fault diagnosis method of the present invention;
FIG. 2 is a diagram of a prior art method for extracting current characteristics of an operating coil;
FIG. 3 is a schematic diagram of the fault location of the high voltage circuit breaker in the diagnostic method of the present invention;
FIG. 4 is a flow chart of a pre-processing of operating coil current data in a diagnostic method of the present invention;
FIG. 5 is a schematic diagram of a data enhancement method in a diagnostic method of the present invention;
FIG. 6 is a schematic structural diagram of a fault diagnosis model in the diagnosis method of the present invention;
FIG. 7 is a training flow chart of a fault diagnosis model in the diagnosis method of the present invention;
FIG. 8 is a first training process of a fault diagnosis model in the diagnosis method of the present invention;
FIG. 9 is a spatial distribution diagram of characteristic vectors of gray images of test set current in the diagnostic method of the present invention;
fig. 10 is an on-line fault diagnosis process in the diagnosis method of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
FIG. 2 is a graph showing typical closing current timing characteristics of a high voltage circuit breaker operating coil, t 0 、t 1 、t 2 、t 3 、t 4 Respectively a start time, a first peak time, a second peak time, a third peak time and a final time, I 1 、I 2 、I 3 The peak currents are respectively a first peak current, a second peak current and a third peak current. The traditional fault diagnosis method based on the current signal of the operating coil can realize the state identification of the high-voltage circuit breaker by extracting the current waveform characteristics, but the fault diagnosis precision is not high enough due to the fact that a plurality of useful information in the current waveform is ignored.
The high-voltage circuit breaker fault diagnosis method based on image recognition is shown in fig. 1, and comprises the steps of historical data offline training and online fault diagnosis, wherein the historical data offline training comprises the following three steps: the method comprises the steps of firstly carrying out wavelet packet denoising on current historical data of a breaker operating coil to eliminate high-frequency noise and burrs, expanding the number of samples through a data enhancement method, then converting a one-dimensional time sequence of the current data into a two-dimensional tensor by utilizing a mapping function, drawing current images with uniform pixel sizes, forming high-voltage breaker fault samples together with labels of the current images, constructing a breaker fault sample library by a plurality of breaker fault samples, dividing a training set and a testing set by utilizing a ten-fold intersection method to respectively obtain ten groups of different data sets, sequentially training a two-dimensional convolutional neural network model, analyzing the performance of a trained network, and adjusting network structure parameters by combining training targets until a network structure meeting performance requirements is obtained, wherein the network structure and the trained parameters are the fault diagnosis model obtained through training. On-line fault diagnosis is carried out by on-line monitoring of current signals of an operating coil of a circuit breaker, data preprocessing and function mapping are carried out to convert the current signals into on-line current gray-scale images with specified sizes, finally, the on-line current gray-scale images are input into a fault diagnosis model obtained through off-line training, the characteristics of the current gray-scale images are visually identified by a computer, and classification labels of the current gray-scale images are calculated to obtain fault diagnosis results.
The fault diagnosis method comprises the steps of off-line training of historical data and on-line fault diagnosis, wherein the off-line training of the historical data comprises the following steps:
step 1: and (5) preprocessing data. Fig. 3 shows a 40.5kV three-phase vacuum circuit breaker for a new energy station, and uses a spring operating mechanism, and through field investigation and literature investigation, common fault types of the circuit breaker include 7 types of operating coil iron core blocking, closing spring loosening, opening spring loosening, transmission shaft screw loosening, higher control voltage, lower control voltage, operating coil aging and the like. The data labels in normal and various fault conditions are detailed in table 1, the various condition labels are converted into single thermal codes when the convolutional neural network is trained, and the common fault occurrence positions of the circuit breaker are also marked in fig. 3.
Table 1 breaker failure type and status tag
As shown in fig. 4, which is a flowchart for converting current data of an operation coil into a current gray image, for current history data of the operation coil, the influence of high-frequency noise and burrs on waveforms is reduced by denoising with a wavelet packet, the number of samples is increased by a data enhancement method, finally, a one-dimensional time sequence of the current data is converted into a two-dimensional tensor by a mapping function, the current gray image with uniform pixel size is drawn, and the current gray image is obtained by amplitude normalization and image graying.
The operating coil current history data may be expressed as:
first, wavelet packet denoising is performed on current data. The Daubechies wavelet is selected as a wavelet base, the decomposition layer number of the wavelet packet is 3, the acquired operation coil current signal is decomposed by utilizing wavelet packet transformation, and the wavelet packet decomposition process is expressed as follows by a formula:
wherein: i (n) is an acquired operating coil current signal, L is a low-pass filter, H is a high-pass filter, an approximation a is a wavelet coefficient generated by a larger scale factor and represents a low-frequency component of the signal, and a detail d is a wavelet coefficient generated by a smaller scale factor and represents a high-frequency component of the signal.
The detail coefficient threshold is selected, a fixed threshold is selected for each layer from 1 to j, and a threshold is applied to the detail coefficient, and a threshold method is generally used for the processing of the wavelet coefficient, and the method is divided into a soft threshold method and a hard threshold method, wherein the hard threshold method is shown as a formula (3), w is an original wavelet coefficient, lambda is a threshold in a threshold function, if the absolute value of the coefficient is not less than lambda, the value is kept unchanged, and when the absolute value is less than a given threshold, the wavelet coefficient is set to 0.
The soft threshold method is represented by the formula (4), in which when the absolute value of the coefficient after the decomposition of the original signal is not less than the threshold value, the coefficient is changed to the original coefficient minus λ, and when the coefficient is less than the threshold value, the wavelet coefficient is changed to 0.
The selection of the threshold is very important to the noise reduction effect of the current signal, the useful information contained in the current signal can be removed by an excessive threshold, and the noise can be removed incompletely due to an excessive threshold, so that the signal is distorted. The patent selects the sqtwo criterion to calculate the magnitude of the noise reduction threshold, expressed in terms of the disclosure:
wherein:
and carrying out wavelet reconstruction, and calculating the wavelet reconstruction according to the original approximation coefficient of the layer number j and the modification detail coefficients of 1 to j. The method specifically comprises the following steps: first, the approximation a after the last layer decomposition is utilized j And detail value d j Respectively performing zero insertion at intervals, and then respectively connecting with a filter L 1 And H 1 The convolution operation is carried out to obtain a j-1 layer approximation value a j-1 Repeating the process to obtain the original signal a 0 =i (n). Expressed by the formula:
and then, carrying out data enhancement on the denoised data. For the current sequence I (n), there are total 8 states of normal state and 7 fault states, each of which has 25 sets of off-line history data, so that the operating coil current history data has 400 samples in total, wherein the opening current and the closing current each have 200 samples. For each type of current sequence I (n), the sampling time is 150ms, the sampling frequency is 125kHz, the problems of sampling noise and sampling delay caused by field interference are considered, the sampling noise and the sampling delay are respectively translated forward by 0.5ms, 1ms, 1.5ms and 2ms in time sequence, noise is added, the current signals after the noise is added respectively reach the signal to noise ratios of 40dB, 45dB, 50dB and 55dB, current samples with small interference are simulated, 8 groups of enhanced current data are obtained, 3600 samples are contained in a constructed fault sample library of the high-voltage circuit breaker, and the number of the closed current samples is equal to that of the open current samples, as shown in a schematic diagram of a data enhancement method in FIG. 5.
Step 2: converting the one-dimensional time series of the current data into a two-dimensional tensor by using the formula (7), wherein the dimension is (18750),
in the formula, Δi=i max -I min N is the current time series length, taking n=18750.
Pair I by using (8) img Performing dimension scaling, converting into two-dimensional tensors with the sizes (300 ),
I img '(u,v)＝mean(I img (kR:(k+1)R,jR:(j+1)R))u,v,k,j∈[0,1,…N] (8)
where M is the scaled tensor dimension, taking m=300, R is the image scaling, r= [ N/M ]. Mapping of the one-dimensional time series of current data to the current image can be accomplished using equations (7) and (8).
Drawing a current image with uniform pixel size, carrying out amplitude normalization processing on the current image by using a formula (9) to eliminate the influence of the amplitude on the current image characteristics of the operating coil,
and carrying out graying treatment, reducing the number of current image channels, obtaining an operation coil current gray image, and constructing a high-voltage breaker fault sample by the current gray image and a label thereof, wherein a plurality of breaker fault samples are constructed into a breaker fault sample library. In the step 1 and the step 2, wavelet packet denoising reduces the influence of high-frequency noise and burrs on the characteristics of the current image, data enhancement expands the number of samples of the current image, function mapping converts a one-dimensional time sequence of the current data into a two-dimensional tensor, and graying reduces the number of channels of the current image, thereby being beneficial to simplifying the structure of a deep learning network and reducing the operation amount.
Step 3: training a two-dimensional convolutional neural network model, analyzing the performance of a trained network, and adjusting network structure parameters by combining training targets until a network structure meeting the performance requirement is obtained, wherein the network structure and the parameters obtained through training are training to obtain a fault diagnosis model.
Firstly, dividing a training set and a test set by using a ten-fold intersection method based on the high-voltage breaker fault sample library constructed in the step 2, randomly dividing the fault sample library into 10 parts without repeated sampling, selecting 1 part as the test set each time, taking the rest 9 parts as the training set for model training, and repeating 10 times, wherein each subset has a chance to be taken as the test set, and the rest subset is taken as the training set. The current waveform fault sample library is divided into a training set and a testing set by using a ten-fold cross method, the training set comprises 1620 training samples and 180 testing samples, the ten-fold cross verification is used for evaluating the prediction performance of the model, particularly the performance of the trained model on new data, the over fitting can be reduced to a certain extent, and as much effective information as possible can be obtained from limited data.
The two-dimensional convolutional neural network model is then trained using the fault samples. The process of training a convolutional neural network model based on high voltage breaker failure samples is described in detail below.
Fig. 6 is a flow chart of the training of the fault diagnosis model. Firstly, initializing a convolutional neural network super-parameter based on a network structure shown in fig. 7, setting a learning rate to 0.001, setting iteration times to 70 in the example, setting batch processing numbers to 5, selecting an Adam optimizer by an optimizer, and selecting a cross entropy loss function by a loss function; then, extracting a current gray image sample and a target output vector thereof, namely a sample label, from the training set, sequentially calculating from a front layer to a rear layer to obtain the output of the convolutional neural network, wherein the two-dimensional discrete convolution of the single-layer convolutional layer on the input tensor is as shown in a formula (10):
where k is the number of input tensor channels and l is the number of output tensor channels. The length and width of the output feature map are as shown in formula (11):
where s is the convolution kernel movement step length and k is the convolution kernel size
Then calculate the cross entropy loss of convolutional neural network output and target vector according to equation (12):
wherein p (x) i ) Q (x) i ) And (5) a sample label probability distribution which is output by the neural network forward operation.
And (3) updating parameters of each layer in the network by utilizing a gradient descent principle according to a formula (13) until the loss is lower than a set threshold value or the iterative training times reach a preset value, and at the moment, the accuracy of the training set reaches a stable value.
F(n+1)＝F(n)±u·||H(p,q)|| (13)
Wherein F represents parameters of each layer in the network, and u is an iteration factor.
Finally, checking whether the accuracy rate of the training set and the testing set meets the fault diagnosis precision requirement, if not, resetting the super-parameters of the convolutional neural network, adjusting the layer number, and repeating the steps until the model evaluation index meets the requirement. So far, once model training is completed, the evaluation index of the model is calculated, the data set divided by the ten-fold intersection method is used for training the model for 9 times, and the average value of the evaluation index of the model for 10 times is calculated as the final evaluation index of the model.
The optimal convolutional neural network structure is selected through a comparison experiment, 4 convolutional neural network models are generated by adjusting the feature extractor structure on the basis of not changing the full-connection layer structure, performances of the models are compared, the design targets of high accuracy of fault diagnosis and light weight of the network structure are considered, the fault diagnosis model shown in the figure 7 is selected, the network structure is the fault diagnosis model of the embodiment, the model is composed of 3 layers of convolutional layers, 3 layers of activation function layers, 2 layers of pooling layers and 2 layers of full-connection layers, output of the upper layer serves as input of the lower layer, four dimensions of convolutional layer parameters respectively represent the number of input channels, the number of output channels, the size of convolutional cores and the moving step length of the convolutional cores, two dimensions of pooling layer parameters respectively represent pooling size and step length, and two dimensions of the full-connection layer parameters respectively represent the number of nodes of the input layer and the number of nodes of the output layer. The convolution layer realizes two-dimensional discrete convolution of an input image and a convolution kernel, the main task is to extract advanced features hidden in a current gray image, an activation function layer is usually a nonlinear unit and is used for processing linear activation response output by the convolution layer, nonlinear characterization capability of a network model on complex features is improved, a pooling layer replaces the value of the network at the position by total feature statistics of adjacent output values through a pooling function, the parameter quantity can be greatly reduced under the condition that input data is kept approximately unchanged, meanwhile, pooling operation also provides translational and rotation invariance to a certain extent for the input current waveform gray image, robustness of the whole neural network is enhanced after the pooling layer is added, the problem of overfitting is effectively avoided, all neurons between two adjacent layers in the fully-connected layer are fully connected, and input information of the previous layer can be more carefully utilized.
The network structure of the fault diagnosis model of the high-voltage circuit breaker designed in the embodiment is shown in a table 2, wherein four dimensions of the output tensor of the convolution layer and the pooling layer in the table respectively represent the batch processing number, the channel number, the length and the width of the feature map, and two dimensions of the output tensor of the full-connection layer respectively represent the batch processing number and the feature vector dimension. Compared with an AlexNet classical deep learning network, the network parameter quantity is less, the network structure is simpler and lighter, and the method is more suitable for the practical application of fault diagnosis of the high-voltage circuit breaker.
Table 2 high-voltage circuit breaker fault diagnosis model network structure
As can be seen from fig. 7 and table 2, the fault diagnosis model network makes the following improvements based on the AlexNet classical deep learning network structure:
(1) The design targets of high accuracy of fault diagnosis and light network structure are considered by reducing the number of convolution layers and convolution kernels, the characteristics of an input current gray level image are extracted by using the convolution layers and global maximum pooling layers to replace full-connection layers, and the design targets of network strengthening capability are realized, so that the parameter number is reduced, overfitting is avoided, and the network strengthening capability is enhanced;
(2) And an active layer ReLU is added after each convolution layer, so that the description capability of the network to the nonlinear mapping relation is enhanced, the convergence is easy, and compared with the Sigmoid active layer, the method is simpler in calculation and lower in calculation cost.
The performance of the fault diagnosis model obtained by 10 times of training in the example is listed in table 3, fig. 8 is a first model training process, and it can be seen that after the model is trained for 70 times in an iterative manner, the accuracy of the training set, the accuracy of the test set and the cross entropy loss all tend to be stable, the accuracy of the training set reaches 100% in 10 times of training processes, the model convergence is good, the average value of the accuracy of the test set is 99.62%, no fitting phenomenon occurs, and the fault diagnosis process of 180 test set samples is completed only by about 0.7s, so that the requirement of on-line fault diagnosis of the circuit breaker is met.
TABLE 3 Performance of fault diagnosis models
In order to better analyze the performance of the fault diagnosis model, the high-order features extracted by the model are mined, and the high-dimensional output of a convolution layer in the model is subjected to dimension reduction by adopting a t-SNE algorithm, so that the high-dimensional data is visualized. As shown in fig. 9, which shows a spatial distribution diagram of feature vectors of a current gray image after the first model training is completed, the dimension 1 and the dimension 2 in the diagram have no practical physical significance, it can be seen that the high-order feature vectors of the current gray image after the feature extraction of the convolution layer become distinguishable, the feature vectors of the current gray image belonging to the same state are gathered in the space, the distance between the feature vectors of different states is larger, the division boundary is obvious, which indicates that the fault diagnosis model can extract the high-order features for identifying the current gray image of different states, and the feature extraction method is learned.
The online fault diagnosis comprises the following steps:
step 1: on-line monitoring a high-voltage breaker operating coil current signal, and removing high-frequency noise and burrs through wavelet packet denoising;
step 2: converting the one-dimensional time sequence of the current data into a two-dimensional tensor by using a mapping function, drawing a current image with uniform pixel size, and carrying out amplitude normalization and image graying on the current image to obtain an online current gray image with specified pixel size;
step 3: inputting the online current gray level image into a fault diagnosis model obtained through offline training, visually identifying the characteristics of the online current gray level image by using a computer, and calculating a classification label of the online current gray level image to obtain a fault diagnosis result.
And (3) monitoring current data of an operating coil of the circuit breaker in a certain closing action on line, preprocessing the data and mapping functions, and converting the current data into a current gray level image. For convenience of explanation, the number of batch processing is set to be 1, for example, a current gray image input to a fault diagnosis model in fig. 10 is on-line data of a circuit breaker under the condition of iron core jamming fault, the current gray image can be equivalently regarded as a tensor with one dimension (1,1,300,300), a first layer of convolution layer has 8 convolution kernels, 8 feature images with the dimension (1,1,99,99) can be output after discrete convolution operation, after the activation layer, a value smaller than zero in the tensor is set to be zero, after the maximum pooling layer is passed, a value of a region in the feature image is replaced by a maximum value in the region, 8 feature images with the dimension (1,1,30,30) are output, and the calculation processes of the second layer and the third layer of convolution layers are the same.
The third layer convolution layer outputs 32 feature images with the dimensions of (1, 3), the feature images are flattened into a one-dimensional vector containing 288 elements, the vector can be regarded as a high-order feature extracted by the convolution layer from an input current gray image, the conversion from the current image feature to the vector is realized, the vector is input into a full-connection layer to carry out linear operation, a one-dimensional vector containing 10 elements is output, after the vector passes through a softmax classifier, the value of the output vector is between 0 and 1, the value size represents the probability that an input current waveform belongs to one of 0-7 classification labels, for example, the output vector is [99.1487,17.7252, -54.4929,54.8098,43.3669, -23.4523, -40.6012, -74.0036,40.1601,46.9580], the output vector after the softmax classifier is [1,0,0,0,0,0,0,0,0,0], the probability that the input current gray image belongs to a label 0, namely, the iron core is maximum, and therefore, the output fault diagnosis result is iron core jam.
In the description of the present specification, the descriptions of the terms "one embodiment," "example," "specific example," and the like, mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The foregoing has shown and described the basic principles, principal features, and advantages of the present disclosure. It will be understood by those skilled in the art that the present disclosure is not limited to the embodiments described above, which have been described in the foregoing and description merely illustrates the principles of the disclosure, and that various changes and modifications may be made therein without departing from the spirit and scope of the disclosure, which is defined in the appended claims.
Claims (9)
1. The fault diagnosis method of the high-voltage circuit breaker based on image recognition is characterized by comprising the following steps of:
collecting current history data of an operating coil of the high-voltage circuit breaker, preprocessing, converting a one-dimensional time sequence of the preprocessed current history data into a two-dimensional tensor by using a mapping function, and drawing a current image with uniform pixel size;
processing the current image to obtain a current gray image, and combining a historical data tag to form a high-voltage circuit breaker fault sample to obtain a high-voltage circuit breaker fault sample library;
training a pre-established two-dimensional convolutional neural network model based on a high-voltage circuit breaker fault sample to obtain a fault diagnosis model meeting the precision requirement;
and on-line monitoring the current of the operating coil of the high-voltage circuit breaker, converting the current into an on-line current gray level image after data preprocessing, and inputting the on-line current gray level image into a trained fault diagnosis model to obtain a fault diagnosis result.
2. The image recognition-based high voltage circuit breaker fault diagnosis method according to claim 1, wherein the preprocessing process comprises the steps of removing high frequency noise and burrs through wavelet packet denoising, and expanding the number of samples through a data enhancement method.
3. The image recognition-based high voltage circuit breaker fault diagnosis method according to claim 2, wherein the data enhancement method expands the data amount by generating samples close to the current raw data, including time-shifting the current signal and adding random noise.
4. The method for diagnosing faults of the high voltage circuit breaker based on image recognition according to claim 1, wherein the current image is subjected to amplitude normalization and image graying to obtain a current gray image.
5. The high-voltage circuit breaker fault diagnosis method based on image recognition according to claim 1, wherein the training process of the fault diagnosis model is as follows:
dividing the high-voltage breaker fault samples in the high-voltage breaker fault sample library into a training set and a testing set by using a ten-fold intersection method to respectively obtain ten groups of different data sets, training a two-dimensional convolutional neural network model in sequence, analyzing the performance of a trained network, and adjusting network structure parameters by combining training targets until a network structure meeting the performance requirements is obtained, wherein the network structure and the trained parameters are the trained fault diagnosis model.
6. The method for diagnosing the fault of the high-voltage circuit breaker based on image recognition according to claim 5, wherein in the trained fault diagnosis model, pixel sparse characteristics of an input current gray level image are considered when structural parameters of a convolutional neural network are initialized, a convolution kernel with smaller dimension and smaller output channel number are selected in a convolution layer, a larger pooling area is arranged in a pooling layer, the number of convolution layers and the number of full connection layers are reduced under the condition that the precision of the fault diagnosis model meets the requirement, the parameter quantity and the complexity of the fault diagnosis model are reduced, the design targets of high accuracy of fault diagnosis and light weight of a network structure are considered, and a regularization layer is added before the full connection layer so as to prevent the fault diagnosis model from being fitted excessively.
7. The image recognition-based high voltage circuit breaker fault diagnosis method according to claim 1, wherein the operating coil current history data is as follows:
wherein I (n) is the time series of the operating coil current signal, I n For time t n Corresponding current value;
decomposing an operation coil current signal acquired by using wavelet packet transformation, wherein the wavelet packet decomposition process is expressed as follows:
wherein: i (n) is an acquired operating coil current signal, L is a low-pass filter, H is a high-pass filter, an approximation a is a wavelet coefficient generated by a larger scale factor and represents a low-frequency component of the signal, and a detail d is a wavelet coefficient generated by a smaller scale factor and represents a high-frequency component of the signal;
the Sqtwolog criterion is selected to calculate the magnitude of the noise reduction threshold of the operating coil current history data, formulated as:
wherein:
8. the image recognition-based high voltage circuit breaker fault diagnosis method according to claim 1, wherein performing wavelet reconstruction, calculating wavelet reconstruction according to the original approximation coefficient of the layer number j and the modification detail coefficients of 1 to j, comprises: first, the approximation a after the last layer decomposition is utilized j And detail value d j Respectively performing zero insertion at intervals, and then respectively connecting with a filter L 1 And H 1 The convolution operation is carried out to obtain a j-1 layer approximation value a j-1 Repeating the process to obtain the original signal a 0 =i (n), expressed by the formula:
9. the image recognition-based high voltage circuit breaker fault diagnosis method according to claim 6, wherein the training process of the fault diagnosis model comprises the following steps:
initializing a convolutional neural network hyper-parameter, extracting a current gray image sample and a target output vector thereof, namely a sample label, from a training set, sequentially calculating from a front layer to a rear layer to obtain the output of the convolutional neural network, wherein the two-dimensional discrete convolution of a single-layer convolutional layer on an input tensor is as shown in formula (10):
wherein k is the number of input tensor channels, l is the number of output tensor channels, and the length and width of the output characteristic diagram are as shown in formula (11):
wherein s is the moving step length of the convolution kernel, and k is the size of the convolution kernel;
then calculate the cross entropy loss of convolutional neural network output and target vector according to equation (12):
wherein p (x) i ) Q (x) i ) Sample label probability distribution output for neural network forward operation;
updating parameters of each layer in the network by utilizing a gradient descent principle according to a formula (13) until the loss is lower than a set threshold value or the iterative training times reach a preset value, and the accuracy of the training set reaches a stable value at the moment;
F(n+1)＝F(n)±u·H(p,q)(13)
wherein F represents parameters of each layer in the network, and u is an iteration factor;
finally, checking whether the accuracy rate on the training set and the testing set meets the requirement of fault diagnosis precision, if not, resetting the super parameters of the convolutional neural network, adjusting the layer number, repeating the steps until the evaluation index of the fault diagnosis model meets the requirement, and finishing the training of the fault diagnosis model after the requirement is met.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310497416.8A CN116524200A (en) | 2023-05-05 | 2023-05-05 | High-voltage circuit breaker fault diagnosis method based on image recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310497416.8A CN116524200A (en) | 2023-05-05 | 2023-05-05 | High-voltage circuit breaker fault diagnosis method based on image recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116524200A true CN116524200A (en) | 2023-08-01 |
Family
ID=87391768
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310497416.8A Pending CN116524200A (en) | 2023-05-05 | 2023-05-05 | High-voltage circuit breaker fault diagnosis method based on image recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116524200A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117114420A (en) * | 2023-10-17 | 2023-11-24 | 南京启泰控股集团有限公司 | Image recognition-based industrial and trade safety accident risk management and control system and method |
| CN117169657A (en) * | 2023-11-03 | 2023-12-05 | 山东和兑智能科技有限公司 | High-voltage cable state monitoring method and system based on artificial intelligence |
| CN117235583A (en) * | 2023-11-13 | 2023-12-15 | 国网浙江省电力有限公司双创中心 | Monitoring method and system for GIS breaker actuating mechanism |
| CN117498262A (en) * | 2023-10-31 | 2024-02-02 | 神州技测(深圳)科技有限公司 | High-voltage direct-current electronic load switch protection circuit |
| CN117595133A (en) * | 2023-10-30 | 2024-02-23 | 江苏双汇电力发展股份有限公司 | Direct-acting type inflatable high-voltage switch cabinet and working method thereof |
| CN117872124A (en) * | 2024-03-12 | 2024-04-12 | 杭州布雷科电气有限公司 | Off-line testing system and method for circuit breaker based on neural network |
-
2023
- 2023-05-05 CN CN202310497416.8A patent/CN116524200A/en active Pending
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117114420A (en) * | 2023-10-17 | 2023-11-24 | 南京启泰控股集团有限公司 | Image recognition-based industrial and trade safety accident risk management and control system and method |
| CN117114420B (en) * | 2023-10-17 | 2024-01-05 | 南京启泰控股集团有限公司 | Image recognition-based industrial and trade safety accident risk management and control system and method |
| CN117595133A (en) * | 2023-10-30 | 2024-02-23 | 江苏双汇电力发展股份有限公司 | Direct-acting type inflatable high-voltage switch cabinet and working method thereof |
| CN117595133B (en) * | 2023-10-30 | 2024-05-17 | 江苏双汇电力发展股份有限公司 | Direct-acting type inflatable high-voltage switch cabinet and working method thereof |
| CN117498262A (en) * | 2023-10-31 | 2024-02-02 | 神州技测(深圳)科技有限公司 | High-voltage direct-current electronic load switch protection circuit |
| CN117169657A (en) * | 2023-11-03 | 2023-12-05 | 山东和兑智能科技有限公司 | High-voltage cable state monitoring method and system based on artificial intelligence |
| CN117169657B (en) * | 2023-11-03 | 2024-01-12 | 山东和兑智能科技有限公司 | High-voltage cable state monitoring method and system based on artificial intelligence |
| CN117235583A (en) * | 2023-11-13 | 2023-12-15 | 国网浙江省电力有限公司双创中心 | Monitoring method and system for GIS breaker actuating mechanism |
| CN117235583B (en) * | 2023-11-13 | 2024-01-30 | 国网浙江省电力有限公司双创中心 | Monitoring method and system for GIS breaker actuating mechanism |
| CN117872124A (en) * | 2024-03-12 | 2024-04-12 | 杭州布雷科电气有限公司 | Off-line testing system and method for circuit breaker based on neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116524200A (en) | High-voltage circuit breaker fault diagnosis method based on image recognition | |
| CN110929763B (en) | Multi-source data fusion-based mechanical fault diagnosis method for medium-voltage vacuum circuit breaker | |
| CN104181460B (en) | A kind of permanent magnetic actuator vacuum circuit breaker method for diagnosing faults based on Multi-information acquisition | |
| CN105891707A (en) | Opening-closing fault diagnosis method for air circuit breaker based on vibration signals | |
| Yang et al. | Condition evaluation for opening damper of spring operated high-voltage circuit breaker using vibration time-frequency image | |
| CN110829417A (en) | Electric power system transient stability prediction method based on LSTM double-structure model | |
| CN109284672A (en) | A kind of Mechanical Failure of HV Circuit Breaker diagnostic method based on PSO-Kmeans algorithm | |
| CN113780060A (en) | High-voltage switch cabinet situation sensing method based on multi-mode deep learning | |
| CN111915101A (en) | Complex equipment fault prediction method and system based on LPP-HMM method | |
| CN110412458A (en) | A kind of OLTC drive failure diagnostic method | |
| CN109948194A (en) | A kind of high-voltage circuitbreaker mechanical defect integrated study diagnostic method | |
| CN117471252A (en) | DC fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM | |
| CN111079647A (en) | Circuit breaker defect identification method | |
| CN114239653A (en) | KPCA-MIV-LSTM-based fault arc multi-domain identification method | |
| CN113611568A (en) | Vacuum circuit breaker based on genetic convolution depth network | |
| CN117607672A (en) | Intelligent monitoring method and system for GIS circuit breaker | |
| CN117520809A (en) | Transformer fault diagnosis method based on EEMD-KPCA-CNN-BiLSTM | |
| CN116559728A (en) | Power transmission line fault diagnosis method based on wavelet transformation-multilayer perceptron | |
| CN116298725A (en) | Fault arc detection method, system and storage medium | |
| CN114157552B (en) | Distribution network fault detection method based on twin time sequence diagram network | |
| CN115564075A (en) | Main and auxiliary integrated fault collaborative diagnosis method and system for urban power grid | |
| CN114186590A (en) | Power distribution network single-phase earth fault identification method based on wavelet and deep learning | |
| CN115130500A (en) | Motor fault diagnosis system and method based on SDAE algorithm | |
| Hong et al. | Machine learning techniques for power system application: fault analysis | |
| CN114021424A (en) | PCA-CNN-LVQ-based voltage sag source identification method |
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 |