CN110174610A - A method of obtaining A.C. contactor electric life based on convolutional neural networks - Google Patents
A method of obtaining A.C. contactor electric life based on convolutional neural networks Download PDFInfo
- Publication number
- CN110174610A CN110174610A CN201910414443.8A CN201910414443A CN110174610A CN 110174610 A CN110174610 A CN 110174610A CN 201910414443 A CN201910414443 A CN 201910414443A CN 110174610 A CN110174610 A CN 110174610A
- Authority
- CN
- China
- Prior art keywords
- contactor
- data
- breaking arc
- neural networks
- convolutional neural
- 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.)
- Granted
Links
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
- G01R31/3277—Testing of circuit interrupters, switches or circuit-breakers of low voltage devices, e.g. domestic or industrial devices, such as motor protections, relays, rotation switches
- G01R31/3278—Testing of circuit interrupters, switches or circuit-breakers of low voltage devices, e.g. domestic or industrial devices, such as motor protections, relays, rotation switches of relays, solenoids or reed switches
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- 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
Abstract
The present invention provides a kind of method that A.C. contactor electric life is obtained based on convolutional neural networks, including obtaining the breaking arc experimental data of A.C. contactor and the contact qualitative data of the front and back of breaking arc experiment each time;The contact mass loss data of breaking arc experiment each time are calculated, and handles and obtains breaking arc discrete sample, contact mass loss data and breaking arc discrete sample are further randomly divided into training set and test set by a certain percentage;Construct the A.C. contactor Endurance Prediction model returned based on convolutional neural networks;Prediction model is trained and is tested, by comparing, obtains trained prediction model;The current breaking arc data for obtaining A.C. contactor in the trained prediction model of importing, export A.C. contactor electric life.Implement the present invention, A.C. contactor electric life Accurate Prediction can be realized in the data of breaking operation before being not required to existing on-off operation method, improves reliability and resource utilization.
Description
Technical field
The present invention relates to A.C. contactor detection technique fields, more particularly to one kind to be obtained based on convolutional neural networks
The method of A.C. contactor electric life.
Background technique
A.C. contactor in low-voltage electrical apparatus has the characteristics that on-off line operation is frequent, is applied to extensively, in large quantities electrical
In control system.Electromagnetic AC contactor is the research object of this paper, and this contactor usually has service life, once make
The limit is reached with the service life, A.C. contactor can not just work on.
Mechanical life and electrical endurance are two macrotaxonomies of A.C. contactor service life, and electrical endurance is much small
In mechanical life, if electric life can extend, very big economic benefit can be brought to electric system.In order to study ac contactor
The electric life of device can predict electric life by certain means, provide a kind of means for the extension of electric life.
Currently, predicted using the method for on-off operation counting how many times come A.C. contactor electric life, in order to
Guaranteeing safety, a lesser region is conservatively arranged in maximum number of operations, although this prediction technique is simple and easy,
But it there is a problem that reliability is not high low with resource utilization, be unable to satisfy the modern industrial society being constantly progressive to industry
The requirement of product high-precision, high-environmental.
Summary of the invention
The technical problem to be solved by the embodiment of the invention is that providing one kind obtains friendship based on convolutional neural networks
The method for flowing contactor electric life, does not need the data of breaking operation before on-off operation method in the prior art
It realizes the Accurate Prediction of A.C. contactor electric life, improves reliability and resource utilization.
In order to solve the above-mentioned technical problem, the embodiment of the invention provides one kind, and friendship is obtained based on convolutional neural networks
The method for flowing contactor electric life, the described method comprises the following steps:
Obtain the breaking arc experimental data of A.C. contactor and the contact quality of the front and back of breaking arc experiment each time
Data;
According to the contact qualitative data of accessed A.C. contactor breaking arc experiment each time front and back, calculate
The contact mass loss data of breaking arc experiment each time, and the breaking arc of accessed A.C. contactor is tested
Data are processed into breaking arc discrete sample, and further by calculated contact mass loss data of institute and handled
To breaking arc discrete sample be randomly divided into training set and test set by a certain percentage;
Using breaking arc discrete point sample as the aspect of model, contact mass loss data are model label, construct and are based on
The A.C. contactor Endurance Prediction model that convolutional neural networks return;
According to the training set and the test set, the A.C. contactor Endurance Prediction model is instructed respectively
Practice and test, and by comparing the A.C. contactor Endurance Prediction model contact gross mass in training and test respectively
The square mean error amount of label and convolutional neural networks output is lost, obtains trained A.C. contactor Endurance Prediction
Model;
The current breaking arc data of A.C. contactor are obtained, and by the current disjunction of accessed A.C. contactor
Arc data imports in obtained trained A.C. contactor Endurance Prediction model, the trained ac contactor
The result of device Endurance Prediction model output is A.C. contactor electric life.
Wherein, the contact mass loss data of the experiment of breaking arc each time are real according to breaking arc each time
The contact qualitative data for testing front and back obtains the contact mass loss of the front and back of breaking arc experiment each time, and further uses line
Property interpolation method processing each time breaking arc experiment front and back contact mass loss obtained from.
Wherein, the breaking arc Data Processing in Experiment by accessed A.C. contactor is discrete at breaking arc
The specific steps of sample include:
In the breaking arc experimental data of accessed A.C. contactor, sampled point is determined, and use linear interpolation
Method expands the number of the sampled point, and further normalization exists the sample point data scope limitation after the linear interpolation
[0,1] in;
According to after the linear interpolation sampled point and its corresponding data area, pass through preset data waveform software
Extract one-dimensional discrete data, and the one-dimensional discrete data group synthesizing one-dimensional matrix that will be extracted;Wherein, described group
The one-dimensional matrix of synthesis is reassembled as Two-Dimensional Moment in convolutional neural networks in the A.C. contactor Endurance Prediction model automatically
Battle array.
Wherein, described using breaking arc discrete point sample as the aspect of model, contact mass loss data are model label,
Constructing the specific steps of A.C. contactor Endurance Prediction model returned based on convolutional neural networks includes:
Convolutional neural networks are established using deep learning frame TensorFlow, establish the A.C. contactor electric life
Prediction model, using breaking arc discrete point sample as the aspect of model, contact mass loss data are read respectively as model label
Enter in prediction model, defines batch function module, data read module, CNN construction module, model-evaluation index module, training
And test module;Wherein, there are two convolution pond alternating layer and two full articulamentums for the convolutional neural networks.
The implementation of the embodiments of the present invention has the following beneficial effects:
The present invention passes through convolutional neural networks model prediction using the current-voltage waveform of breaking arc as input feature vector
The mass loss of contact not only overcomes the variation at starting the arc phase angle to the tremendous influence of arc waveform, can also be for each time
Breaking operation predicts electric life in real time, without the data of breaking operation before, greatly simplifies the complexity of experimentation,
Improve Efficiency.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will to embodiment or
Attached drawing needed to be used in the description of the prior art is briefly described, it should be apparent that, the accompanying drawings in the following description is only
Some embodiments of the present invention, for those of ordinary skill in the art, without any creative labor,
It obtains other drawings based on these drawings and still falls within scope of the invention.
Fig. 1 is the method provided in an embodiment of the present invention that A.C. contactor electric life is obtained based on convolutional neural networks
Flow chart;
Fig. 2 is the method provided in an embodiment of the present invention that A.C. contactor electric life is obtained based on convolutional neural networks
Contact gross mass Dissipation change tendency chart in application scenarios;
Fig. 3 is the method provided in an embodiment of the present invention that A.C. contactor electric life is obtained based on convolutional neural networks
Data waveform in application scenarios analyzes software interface;
Fig. 4 is the method provided in an embodiment of the present invention that A.C. contactor electric life is obtained based on convolutional neural networks
The arc waveform figure of a cycle in application scenarios;
Fig. 5 is the method provided in an embodiment of the present invention that A.C. contactor electric life is obtained based on convolutional neural networks
Arc waveform figure after linear interpolation in application scenarios;
Fig. 6 is the method provided in an embodiment of the present invention that A.C. contactor electric life is obtained based on convolutional neural networks
Arc waveform figure after normalization in application scenarios;
Fig. 7 is the method provided in an embodiment of the present invention that A.C. contactor electric life is obtained based on convolutional neural networks
Contact mass loss trend chart after linear interpolation in application scenarios;
Fig. 8 is the method provided in an embodiment of the present invention that A.C. contactor electric life is obtained based on convolutional neural networks
A.C. contactor Endurance Prediction model parameter figure in application scenarios;
Fig. 9 is the method provided in an embodiment of the present invention that A.C. contactor electric life is obtained based on convolutional neural networks
The A.C. contactor Endurance Prediction model structure of foundation in application scenarios;
Figure 10 is the method provided in an embodiment of the present invention that A.C. contactor electric life is obtained based on convolutional neural networks
The result after the execution of A.C. contactor Endurance Prediction model in application scenarios.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, the present invention is made into one below in conjunction with attached drawing
Step ground detailed description.
As shown in Figure 1, the one kind provided obtains ac contactor based on convolutional neural networks in the embodiment of the present invention
The method of device electric life, the described method comprises the following steps:
Step S1, it obtains the breaking arc experimental data of A.C. contactor and breaking arc tests front and back each time
Contact qualitative data;
Specifically, acquiring breaking arc data using electrical endurance system under AC-4 experiment condition, obtaining
6000 groups of breaking arc experimental datas, every to carry out 600 experiments, the quality for all contacts of weighing will carry out altogether 11 titles
Weight, including once testing the qualitative data before not starting.
Step S2, according to accessed A.C. contactor each time breaking arc experiment front and back contact qualitative data,
The contact mass loss data of breaking arc experiment each time are calculated, and the disjunction of accessed A.C. contactor is electric
Arc Data Processing in Experiment is at breaking arc discrete sample, and further by the calculated contact mass loss data of institute and an institute
It handles obtained breaking arc discrete sample and is randomly divided into training set and test set by a certain percentage;
Specifically, being pre-processed to breaking arc data and contact qualitative data.Firstly, breaking arc is real each time
The contact mass loss data tested are divided each time according to the contact qualitative data of the experiment of breaking arc each time front and back
Power off arc experiment front and back contact mass loss, and further using linear interpolation method processing each time breaking arc experiment before
Obtained from contact mass loss afterwards;Secondly, in the breaking arc experimental data of accessed A.C. contactor, really
Determine sampled point, and expands the number of the sampled point using linear interpolation method, and further normalization will be after the linear interpolation
Sample point data scope limitation in [0,1];According to after linear interpolation sampled point and its corresponding data area, pass through
Preset data waveform software extracts one-dimensional discrete data, and the one-dimensional discrete data extracted are combined into one
Tie up matrix;Wherein, one-dimensional matrix can in A.C. contactor Endurance Prediction model in step s3 in convolutional neural networks from
It is dynamic to be reassembled as two-dimensional matrix.
Step S3, using breaking arc discrete point sample as the aspect of model, contact mass loss data are model label, structure
Build out the A.C. contactor Endurance Prediction model returned based on convolutional neural networks;
Specifically, realizing convolutional neural networks using deep learning frame TensorFlow, the A.C. contactor electric longevity is established
Prediction model is ordered, using breaking arc discrete point sample as the aspect of model, contact mass loss data are as model label, respectively
It reads in prediction model, defines batch function module, data read module, convolutional neural networks CNN construction module, model and comment
Valence Index module, training and test module;Wherein, there are two convolution pond alternating layers and two to connect entirely for convolutional neural networks
Connect layer.
In one example, the convolution kernel of 3 × 3 sizes, pond window are used in A.C. contactor Endurance Prediction model
Size be 2 × 2, nonlinear activation function ReLU is just used after each convolution operation, from input in terms of output,
The port number of convolutional neural networks CNN model from 6 becomes 32,32 and becomes 64,64 becoming 1024, final output only have 1 it is logical
Road.The eigenmatrix elongation that pond layer 2 exports is become into one-dimensional matrix, then is input in full articulamentum 1, in full articulamentum 1
Output end has used Dropout method, and the method is given up by probability the partial nerve in network model in the training stage at random
Member.
The specific structure of convolutional neural networks is as follows:
Input layer
≤ 1 convolutional layer 1_1 (3x3x64)
≤ 2 Relu layers of nonlinear responses
≤ 3 convolutional layer 1_2 (3x3x64)
≤ 4 Relu layers of nonlinear responses
≤ 5 pond layers (2x2/2)
≤ 6 convolutional layer 2_1 (3x3x128)
≤ 7 Relu layers of nonlinear responses
≤ 8 convolutional layer 2_2 (3x3x128)
≤ 9 Relu layers of nonlinear responses
≤ 10 pond layers (2x2/2)
≤ 11 convolutional layer 3_1 (3x3x256)
≤ 12 Relu layers of nonlinear responses
≤ 13 convolutional layer 3_2 (3x3x256)
≤ 14 global average pond layers
≤ 15 full articulamentum (256x100) s
≤ 16 Relu layers of nonlinear responses
≤ 17 full articulamentums (100x2)
≤ 14 warp lamination D1 (4x4x256)
≤ 19 convolutional layer D1_1 (3x3x256)
≤ 20 convolutional layer D1_2 (3x3x256)
≤ 21 warp lamination D2 (4x4x128)
≤ 22 convolutional layer D2_1 (3x3x128)
≤ 23 convolutional layer D2_2 (3x3x128)
≤ 24 convolutional layer D2_3 (3x3x2)
Wherein, the number before symbol "≤" is current layer number, and the subsequent number of symbol "≤" is the input number of plies;
It in bracket is convolution layer parameter behind convolutional layer and warp lamination, the wherein product of two multipliers before the convolution layer parameter
For convolution kernel size, which is port number;It is pond layer parameter in bracket behind the layer of pond,
In the products of two multipliers before the pond layer parameter be Chi Huahe size, which is step-length;Entirely
Layer parameter is connected to be complete in bracket behind articulamentum, it is to baby here that wherein the full articulamentum behindness parameter, which is the classification of output,
Whether youngster, which kicks, is checked, so being one two classification;Non-thread response layer is by a nonlinear activation primitive ReLU structure
At.
Step S4, according to the training set and the test set, respectively to the A.C. contactor Endurance Prediction model
It is trained and tests, and by comparing the A.C. contactor Endurance Prediction model contact in training and test respectively
The square mean error amount of label and convolutional neural networks output is lost in gross mass, obtains the trained A.C. contactor electric longevity
Order prediction model;
Specifically, the sample using training set is trained A.C. contactor Endurance Prediction model, using test set
Sample A.C. contactor Endurance Prediction model is tested, pass through the mean square error for asking physical tags and CNN model to export
Difference (MSE) assesses the error amount of contact mass loss, i.e., is trained using the back-propagation algorithm of error, until
Convolutional neural networks convergence, obtains trained A.C. contactor Endurance Prediction model.
Step S5, the current breaking arc data of A.C. contactor, and working as accessed A.C. contactor are obtained
Preceding breaking arc data import in obtained trained A.C. contactor Endurance Prediction model, the trained friendship
The result for flowing the output of contactor Endurance Prediction model is A.C. contactor electric life.
Specifically, obtaining A.C. contactor actually breaking arc data to be measured, it is sent into trained A.C. contactor electricity
In Life Prediction Model, the result of trained A.C. contactor Endurance Prediction model output is A.C. contactor at this time
Electric life.
As shown in Fig. 2 to Figure 10, A.C. contactor electricity is obtained based on convolutional neural networks in the embodiment of the present invention
The application scenarios of the method in service life are described further:
This method acquires 6000 groups of breaking arc experimental datas and contact mass loss number under AC-4 experiment condition
According to.For experimental data by the realization acquisition (not shown) of electric life control system, this system includes control feature, data acquisition
System and load system three parts, wherein the start stop operation of control feature control A.C. contactor, load system are used to adjust
Resistive and inductive load guarantees that experiment current value is 6 times of load current value, and data collection system is collected for saving
Breaking arc data.
Before life experiment starts, four ports of experimental system are shorted, since supply voltage is constant, pass through calculating
And impedance and induction reactance value needed for adjusting experiment, the actual current for allowing experiment to lead to are 6 times, i.e. 480A of rated current 80A,
Power factor is 0.35, and phase voltage when load system is powered is 386V, then the impedance value R and induction reactance of computational load system
Value L.The calculating process of R and L is as follows:
It can obtain:
Then R and L is finely tuned, guarantees that every phase actual current is in 480A or more on A.C. contactor, error is within+5%.
It after A.C. contactor is fixed, is tested according to predetermined operation process, the data acquired each time can be all automatically saved to
Under specified file.
While acquiring experimental data, the mass loss data of contact are also obtained, here after every 600 experiments, are stopped
Experiment, is weighed, it is therefore desirable to 11 groups of matter of weighing with quality of the electronic scale of thousand quartiles to 6 static contacts and 3 moving contacts
Amount, wherein first group is the preceding contact quality weighed of experiment, because experiment number is more, contact wear is more serious, then quality
Loss is just with the difference of the qualitative data before experiment and after experiment.In order to more intuitively show the variation tendency of mass loss, indicate
At the form of Fig. 2, the point where the distinct symbols such as "+", " o " in figure is exactly the mass loss data of contact.
After data acquisition finishes, breaking arc data are extracted with waveform analysis software shown in Fig. 3, and save as mat
Format can import voltage (blue line) and electric current (red line) waveform on the left of interface after initial data is written on the right side of software interface,
And waveform is amplified to breaking arc position, then calculate the characteristic in the middle part of interface, finally save.
Fig. 4 is the breaking arc waveform diagram in a cycle, and each wave period has 600 sampled points.In order to adapt to
The tupe of CNN model needs sampling number being extended for 2n×2n, number of sampling points is increased using linear interpolation method herein
To 1024 (25×25), and normalize and be used to the sample point data scope limitation after linear interpolation in [0,1].
Meanwhile 1024 × 1 one-dimensional matrix can reassemble into 32 automatically in input AC contactor Endurance Prediction model ×
32 two-dimensional matrix.The calculation formula of normalization and recombination is as follows:
Wherein, x (i) (i=1 ..., m2) represents the value of electric arc sampled point waveform, y (j, k) (j=1 ..., m;K=
1 ..., m) indicate that the two-dimensional matrix after recombination, max (x) and min (x) respectively refer to the maximum value and minimum value of sample point data.
Fig. 5 is the arc waveform after linear interpolation, and Fig. 6 is the arc waveform after normalization.
For mass loss data, linear interpolation method is equally used, the data amount check that gross mass is lost is expanded to
6000, the contact mass loss variation tendency after interpolation is shown in Fig. 7.And breaking arc sample and contact mass loss data are pressed
Ratio is randomly divided into training set and test set.
Through deep learning frame TensorFlow and convolutional neural networks (CNN) model, in Ubuntu system
Realize deep learning modeling, in figure altogether there are two convolution pond alternating layer and two full articulamentums, wherein model layer
All parameters are shown in Fig. 8.
The convolution kernel of 3 × 3 sizes is all used in Life Prediction Model, the size of pond window is 2 × 2, non-linear to swash
Function Relu living is just used after each convolution operation, and from inputting in terms of output, the port number of CNN model is become from 6
32,32, which become 64,64, becomes 1024, and final output only has 1 channel.The eigenmatrix elongation that pond layer 2 exports is become one
Matrix is tieed up, then is input in full articulamentum 1, has used Dropout method in the output end of full articulamentum 1, the method is in training
Stage is given up the member of the partial nerve in network model by probability at random, in test phase, is needed to retain whole neuron and is commented
The index of valence model realizes that formula is as follows by mean square error function:
Wherein f (xt) indicate true value, herein refer to mass loss label data, ytIt indicates estimated value, herein refers to CNN model
Output, then the error amount obtained is sent in the Adam optimizer with adaptive learning ability, calculate global minima
Value, the composition of whole Life Prediction Model are shown in Fig. 9.
Here it is tested using the data of 5:1 ratio, that is to say, that training sample number is 5000, test sample
Number is 1000, and trained and test phase Dropout ratio is respectively 0.5 and 1.0, the number of iterations 106, and batch size is
256, every 250 iteration export a square mean error amount (MSE), and training result and test result are expressed as to the form of Figure 10,
Horizontal axis in figure is the number of iterations, and the longitudinal axis is square mean error amount, and black and red curve respectively indicate trained and test error,
The MSE value of the training and test that obtain after trained and test phase is finished is labelled in the lower right corner of figure.
By the way that processed breaking arc experimental data is imported trained A.C. contactor Endurance Prediction model
In, then execute prediction model and can obtain prediction result, the complexity of experimentation is greatly simplified, Efficiency is improved.
The implementation of the embodiments of the present invention has the following beneficial effects:
The present invention passes through convolutional neural networks model prediction using the current-voltage waveform of breaking arc as input feature vector
The mass loss of contact not only overcomes the variation at starting the arc phase angle to the tremendous influence of arc waveform, can also be for each time
Breaking operation predicts electric life in real time, without the data of breaking operation before, greatly simplifies the complexity of experimentation,
Improve Efficiency.
Those of ordinary skill in the art will appreciate that implement the method for the above embodiments be can be with
Relevant hardware is instructed to complete by program, the program can be stored in a computer readable storage medium,
The storage medium, such as ROM/RAM, disk, CD.
The above disclosure is only the preferred embodiments of the present invention, cannot limit the right of the present invention with this certainly
Range, therefore equivalent changes made in accordance with the claims of the present invention, are still within the scope of the present invention.
Claims (4)
1. a kind of method for obtaining A.C. contactor electric life based on convolutional neural networks, which is characterized in that the method packet
Include following steps:
Obtain the breaking arc experimental data of A.C. contactor and the contact qualitative data of the front and back of breaking arc experiment each time;
According to the contact qualitative data of accessed A.C. contactor breaking arc experiment each time front and back, calculate each time
The contact mass loss data of breaking arc experiment, and by the breaking arc Data Processing in Experiment of accessed A.C. contactor
At breaking arc discrete sample, and further by calculated contact mass loss data and it is handled obtain disjunction electricity
Arc discrete sample is randomly divided into training set and test set by a certain percentage;
Using breaking arc discrete point sample as the aspect of model, contact mass loss data are model label, and building is based on convolution mind
A.C. contactor Endurance Prediction model through net regression;
According to the training set and the test set, the A.C. contactor Endurance Prediction model is trained and is surveyed respectively
Examination, and label is lost in contact gross mass in training and test respectively by comparing the A.C. contactor Endurance Prediction model
With the square mean error amount of convolutional neural networks output, trained A.C. contactor Endurance Prediction model is obtained;
The current breaking arc data of A.C. contactor are obtained, and by the current breaking arc number of accessed A.C. contactor
According to importing in obtained trained A.C. contactor Endurance Prediction model, the trained A.C. contactor electric life
The result of prediction model output is A.C. contactor electric life.
2. the method for being obtained A.C. contactor electric life based on convolutional neural networks as described in claim 1, feature are existed
In the contact mass loss data of the experiment of breaking arc each time are the contacts according to the experiment of breaking arc each time front and back
Qualitative data is obtained the contact mass loss of the front and back of breaking arc experiment each time, and is further handled using linear interpolation method
Obtained from the contact mass loss of the front and back of breaking arc experiment each time.
3. the method for being obtained A.C. contactor electric life based on convolutional neural networks as described in claim 1, feature are existed
In, the breaking arc Data Processing in Experiment by accessed A.C. contactor at the specific step of breaking arc discrete sample
Suddenly include:
In the breaking arc experimental data of accessed A.C. contactor, sampled point is determined, and expand using linear interpolation method
Increase the number of the sampled point, and further normalizes the sample point data scope limitation after the linear interpolation in [0,1]
It is interior;
According to after the linear interpolation sampled point and its corresponding data area, extracted by preset data waveform software
One-dimensional discrete data, and the one-dimensional discrete data group synthesizing one-dimensional matrix that will be extracted;Wherein, one be combined into
Dimension matrix is reassembled as two-dimensional matrix in convolutional neural networks in the A.C. contactor Endurance Prediction model automatically.
4. the method for being obtained A.C. contactor electric life based on convolutional neural networks as described in claim 1, feature are existed
In described using breaking arc discrete point sample as the aspect of model, contact mass loss data are model label, are constructed based on volume
The specific steps of A.C. contactor Endurance Prediction model of product neural net regression include:
It realizes that convolutional neural networks return using deep learning frame TensorFlow, it is pre- to establish the A.C. contactor electric life
Model is surveyed, using breaking arc discrete point sample as the aspect of model, contact mass loss data are read in pre- respectively as model label
It surveys in model, defines batch function module, data read module, CNN construction module, model-evaluation index module, training and survey
Die trial block;Wherein, there are two convolution pond alternating layer and two full articulamentums for the convolutional neural networks.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910414443.8A CN110174610B (en) | 2019-05-17 | 2019-05-17 | Method for obtaining electric service life of alternating current contactor based on convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910414443.8A CN110174610B (en) | 2019-05-17 | 2019-05-17 | Method for obtaining electric service life of alternating current contactor based on convolutional neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110174610A true CN110174610A (en) | 2019-08-27 |
CN110174610B CN110174610B (en) | 2021-10-22 |
Family
ID=67691497
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910414443.8A Active CN110174610B (en) | 2019-05-17 | 2019-05-17 | Method for obtaining electric service life of alternating current contactor based on convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110174610B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111505490A (en) * | 2020-03-23 | 2020-08-07 | 温州大学乐清工业研究院 | AC contactor ablation condition evaluation method based on convolutional neural network regression |
CN112098833A (en) * | 2020-09-18 | 2020-12-18 | 株洲国创轨道科技有限公司 | Relay service life prediction method, system, medium and equipment |
CN112231910A (en) * | 2020-10-15 | 2021-01-15 | 华人运通(江苏)技术有限公司 | System and method for determining service life of high-voltage direct-current contactor |
CN112464556A (en) * | 2020-11-17 | 2021-03-09 | 沈阳工业大学 | AC contactor electric service life prediction method based on long-short term memory neural network |
CN113255579A (en) * | 2021-06-18 | 2021-08-13 | 上海建工集团股份有限公司 | Method for automatically identifying and processing construction monitoring abnormal acquisition data |
CN113420063A (en) * | 2021-06-18 | 2021-09-21 | 上海建工集团股份有限公司 | System for construction monitoring abnormal acquisition data automatic identification and processing |
CN113466681A (en) * | 2021-05-31 | 2021-10-01 | 国网浙江省电力有限公司营销服务中心 | Circuit breaker service life prediction method based on small sample learning |
CN114137403A (en) * | 2021-11-22 | 2022-03-04 | 重庆大学 | On-load tap-changer electrical life evaluation system and method based on radiation electromagnetic waves |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106291351A (en) * | 2016-09-20 | 2017-01-04 | 西安工程大学 | Primary cut-out fault detection method based on convolutional neural networks algorithm |
CN106291339A (en) * | 2015-05-20 | 2017-01-04 | 国网上海市电力公司 | A kind of circuit breaker failure diagnostic expert system based on artificial neural network |
US20180172767A1 (en) * | 2016-12-21 | 2018-06-21 | Eaton Corporation | System and method for monitoring contact life of a circuit interrupter |
CN108734332A (en) * | 2018-03-29 | 2018-11-02 | 浙江长兴笛卡尔科技有限公司 | The method and device of electric parameter is predicted using machine learning |
-
2019
- 2019-05-17 CN CN201910414443.8A patent/CN110174610B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106291339A (en) * | 2015-05-20 | 2017-01-04 | 国网上海市电力公司 | A kind of circuit breaker failure diagnostic expert system based on artificial neural network |
CN106291351A (en) * | 2016-09-20 | 2017-01-04 | 西安工程大学 | Primary cut-out fault detection method based on convolutional neural networks algorithm |
US20180172767A1 (en) * | 2016-12-21 | 2018-06-21 | Eaton Corporation | System and method for monitoring contact life of a circuit interrupter |
CN108226769A (en) * | 2016-12-21 | 2018-06-29 | 伊顿公司 | For the system and method in the contact service life of observation circuit cutout |
CN108734332A (en) * | 2018-03-29 | 2018-11-02 | 浙江长兴笛卡尔科技有限公司 | The method and device of electric parameter is predicted using machine learning |
Non-Patent Citations (1)
Title |
---|
YOUNGJI YOO等: "A Novel Image Feature for the Remaining Useful Lifetime Prediction of Bearings Based on Continuous Wavelet Transform and Convolutional Neural Network", 《APPLIED SCIENCES》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111505490A (en) * | 2020-03-23 | 2020-08-07 | 温州大学乐清工业研究院 | AC contactor ablation condition evaluation method based on convolutional neural network regression |
CN112098833A (en) * | 2020-09-18 | 2020-12-18 | 株洲国创轨道科技有限公司 | Relay service life prediction method, system, medium and equipment |
CN112098833B (en) * | 2020-09-18 | 2023-10-13 | 株洲国创轨道科技有限公司 | Relay service life prediction method, system, medium and equipment |
CN112231910A (en) * | 2020-10-15 | 2021-01-15 | 华人运通(江苏)技术有限公司 | System and method for determining service life of high-voltage direct-current contactor |
CN112464556A (en) * | 2020-11-17 | 2021-03-09 | 沈阳工业大学 | AC contactor electric service life prediction method based on long-short term memory neural network |
CN113466681A (en) * | 2021-05-31 | 2021-10-01 | 国网浙江省电力有限公司营销服务中心 | Circuit breaker service life prediction method based on small sample learning |
CN113255579A (en) * | 2021-06-18 | 2021-08-13 | 上海建工集团股份有限公司 | Method for automatically identifying and processing construction monitoring abnormal acquisition data |
CN113420063A (en) * | 2021-06-18 | 2021-09-21 | 上海建工集团股份有限公司 | System for construction monitoring abnormal acquisition data automatic identification and processing |
CN114137403A (en) * | 2021-11-22 | 2022-03-04 | 重庆大学 | On-load tap-changer electrical life evaluation system and method based on radiation electromagnetic waves |
Also Published As
Publication number | Publication date |
---|---|
CN110174610B (en) | 2021-10-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110174610A (en) | A method of obtaining A.C. contactor electric life based on convolutional neural networks | |
US20190243735A1 (en) | Deep belief network feature extraction-based analogue circuit fault diagnosis method | |
CN107423839A (en) | A kind of method of the intelligent building microgrid load prediction based on deep learning | |
CN107065843B (en) | Multi-direction KICA batch process fault monitoring method based on Independent subspace | |
CN112152201B (en) | Electricity load prediction method and system based on convolution length time memory neural network | |
CN109871858A (en) | Prediction model foundation, object recommendation method and system, equipment and storage medium | |
CN106203531A (en) | A kind of transmission line fault sorting technique based on the sparse autocoder of convolution | |
CN112149873B (en) | Low-voltage station line loss reasonable interval prediction method based on deep learning | |
CN110503004A (en) | A kind of online method of discrimination of Switching Power Supply operating status | |
CN112241605B (en) | Breaker energy storage process state identification method for constructing CNN feature matrix by using acoustic vibration signals | |
CN116776273B (en) | Running state monitoring method and system of Internet of things equipment | |
CN110633870A (en) | Harmonic early warning method, harmonic early warning device and terminal equipment | |
CN110672905A (en) | CNN-based self-supervision voltage sag source identification method | |
CN112085111B (en) | Load identification method and device | |
CN105626502A (en) | Plunger pump health assessment method based on wavelet packet and Laplacian Eigenmap | |
CN111476299A (en) | Improved convolutional neural network and power grid intelligent alarm system based on same | |
Wu et al. | Multi-sensor signal fusion for a compound fault diagnosis method with strong generalization and noise-tolerant performance | |
Rahimpour et al. | Applying artificial optimization methods for transformer model reduction of lumped parameter models | |
CN113158781B (en) | Lightning trip type identification method | |
CN114004162A (en) | Modeling method for smelting load harmonic emission level under multi-working-condition scene | |
CN106295691A (en) | The feature selection of single classification SVM and parameter synchronization optimization method | |
CN102890718B (en) | Electronic nose data mining method based on supervised explicit manifold learning algorithm | |
CN109470954B (en) | Power grid running state monitoring system based on big data and monitoring method thereof | |
CN115685780A (en) | Method for constructing digital twin body of circuit breaker | |
CN113671287B (en) | Intelligent detection method, system and readable storage medium for power grid automation terminal |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |