CN109472268A - A kind of protection pressing plate state identification method based on convolutional neural networks - Google Patents
A kind of protection pressing plate state identification method based on convolutional neural networks Download PDFInfo
- Publication number
- CN109472268A CN109472268A CN201811046237.8A CN201811046237A CN109472268A CN 109472268 A CN109472268 A CN 109472268A CN 201811046237 A CN201811046237 A CN 201811046237A CN 109472268 A CN109472268 A CN 109472268A
- Authority
- CN
- China
- Prior art keywords
- pressing plate
- convolutional neural
- neural networks
- protection pressing
- matrix
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of protection pressing plate state identification method based on convolutional neural networks.The core of this method is to complete identification protecting pressing plate using convolutional neural networks to put into, exit, the Computer Vision Task of stand-by state.Protection pressing plate state identification method provided by the invention based on convolutional neural networks is to carry out image recognition and analysis to protection pressing plate state using convolutional neural networks, to judge to protect the investment of pressing plate, exit and stand-by state.The recognition methods patrols dimension for substation secondary and provides effective technological means, effectively increases the quality and efficiency, reduction field labor intensity of power transformation O&M inspection work, reduces security risk.
Description
Technical field
The present invention relates to substation secondaries to patrol dimension, computer vision field, is based on convolutional Neural more particularly, to one kind
The protection pressing plate state identification method of network.
Background technique
With the continuous expansion of power grid scale, the higher transmission line of electricity main protection of voltage class a set of is developed to by past
Several sets, protection pressing plate (protection is in flakes) have developed to tens by more than ten, have protected the throwing of pressing plate to move back mode also by individual
It throws 1 end and develops to 2 ~ 3 end of throwing, the arrangement of second protection pressing plate is intensive, quantity is more.However, current transformer substation second protection pressing plate master
It to be constrained and be managed by management system, intelligence degree is low.Operation maintenance personnel lacks effective technology during tour
Means, accuracy cannot be guaranteed.Further, since can not make an inspection tour with history, information foundation is contacted and comparative analysis, operation maintenance personnel need
It is verified repeatedly protection pressing plate is maked an inspection tour, certain puzzlement is caused to maintenance work.
Summary of the invention
In order to overcome above-mentioned protection pressing plate to arrange, intensive, quantity is more, makes an inspection tour difficult problem, and the present invention proposes a kind of base
In the protection pressing plate state identification method of convolutional neural networks, the technical solution adopted by the present invention is that:
A kind of protection pressing plate state identification method based on convolutional neural networks, comprising the following steps:
S10. the protection pressing plate image for inputting certain location type of substation field carries out RGB color feature decomposition, in 0-255 model
Enclose each pixel of interior assessment, and by the feux rouges of the whole each pixel of image, green light, blue light pixel number with 28 × 28 matrixes
Form storage, using 3 28 × 28 numerical matrixs as the input of training neural network；
S20. feature identification is carried out, convolution algorithm is carried out to pixel group by convolutional neural networks, setting step-length is come to entire defeated
Enter matrix to be scanned, and then completes to detect the feature of input picture；
S30. amendment linear unit is introduced in convolutional neural networks to increase nonlinear pattern recognition；It will be carried out for the first time in S20
Convolution algorithm show that each value after Feature Mapping passes through this function, to be activated or be lighted.
S40. two dimensional character matrix is obtained to adjacent feature detection matrix progress dimensionality reduction is obtained by flattening: in adjacent spy
Only retain maximum value in sign detection matrix, to obtain the key message of image；
S50. feature flattening: being that a column vector pixel exports by the two dimensional character matrix conversion of above-mentioned acquisition；
S60. one group of weight of Initialize installation and biasing between flattening feature vector and protection pressing plate state output value；
S70. the confidence level probability-weighted of the state judgement output of three kinds of protection pressing plates is obtained by operation, probability is highest to be
This method judges the protection pressing plate state obtained;
S80. cross entropy is used to be used to the gap of comparison prediction and actual value as the loss function of convolutional neural networks, in convolution
During neural metwork training, with network weight/biasing levelling, carry out the accuracy of judgement degree of improved model;
S90. the normal condition of pressing plate will be protected in the state of each protection pressing plate exported in convolutional neural networks and database
Data compare, and identify the protection pressing plate changed.
A kind of protection pressing plate state identification method based on convolutional neural networks provided by the invention carries out input picture
RGB color feature decomposition, and the input as convolutional neural networks；It completes to detect the feature of input picture；Increase non-linear
Pattern-recognition；Pass through the key feature in the detection of maximum pond keeping characteristics；Dimensionality reduction is carried out to eigenmatrix, full connection obtains three
The feature vector of kind state；Protection pressing plate state is obtained by operation；Sentencing for this method is measured by cross entropy loss function
Gap between disconnected output and true pressing plate state, adjusts the ginseng of links by rewards and punishments with certain step-length and direction
Number, to promote the condition discrimination ability of this method.
Preferably, the step of S20 specifically:
S201. avoiding the loss of pixel in learning process by way of inputting the zero padding of picture element matrix boundary；
S202. setting weighted convolution matrix carries out convolution algorithm to input picture element matrix, completes to carry out two to input picture element matrix
A Feature Mapping, the weight for initializing convolution algorithm are arranged between 0.01 to 0.1；
S203. it is that 2 pairs of input picture element matrix features are checked comprehensively that setting step, which is moved,.
Preferably, the step of S70 is specially to be divided into two parts: Logit score and Softmax operation.
Preferably, the Logit score is one basic linear the specific steps are, the Logit score of each output
Function: Logit score=(characteristic quantity × weight)+biasing, highest scoring is the conjecture of judgment models；
Preferably, the Softmax calculation step is specifically, the confidence level that model is guessed in order to obtain, is with natural logrithm e
The truth of a matter is scored at index with Logit, to obtain the confidence level of protection pressing plate state output；By each state output confidence level
Divided by the sum of all confidence scores, confidence level probability-weighted, the highest protection pressure for as judging to obtain of probability have just been obtained
Board status.
Preferably, the identification state of the protection pressing plate include investment, exit with it is spare.
Compared with prior art, the beneficial effect of technical solution of the present invention is:
Protection pressing plate state identification method provided by the invention based on convolutional neural networks is using convolutional neural networks to guarantor
It protects pressing plate state and carries out image recognition and analysis, to judge to protect the investment of pressing plate, exit and stand-by state.The recognition methods
Patrol dimension for substation secondary and provide effective technological means, effectively increase power transformation O&M inspection work quality and efficiency,
Field labor intensity is reduced, security risk is reduced.
Detailed description of the invention
Fig. 1 is the method flow diagram of the protection pressing plate state identification method provided by the invention based on convolutional neural networks.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, only for illustration, Bu Nengli
Solution is the limitation to this patent.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative labor
Every other embodiment obtained under the premise of dynamic, shall fall within the protection scope of the present invention.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
Embodiment 1
A kind of protection pressing plate state identification method based on convolutional neural networks, comprising the following steps:
S10. the protection pressing plate image for inputting certain location type of substation field carries out RGB color feature decomposition, and by image
The feux rouges of each pixel, green light, blue light pixel number with a matrix type, the input as convolutional neural networks；In 0-255
The each pixel of assessment in range, and by the feux rouges of the whole each pixel of image, green light, blue light pixel number with 28 × 28 matrixes
Form storage, using 3 28 × 28 numerical matrixs as the input of trained neural network；
S20. feature identification is carried out, convolution algorithm is carried out to pixel group by convolutional neural networks, setting step-length is come to entire defeated
Enter matrix to be scanned, and then complete to detect the feature of input picture: inputting the zero padding of picture element matrix boundary at 3 28 × 28
Mode avoids the loss of pixel in learning process；Secondly, the weighted convolution matrix of setting 3 × 3 carries out input picture element matrix
Convolution algorithm is completed to carry out two Feature Mappings to input picture element matrix, and the weight setting for initializing convolution algorithm is arrived 0.01
Between 0.1；It is that 2 pairs of input picture element matrix features are checked comprehensively that setting step, which is moved,；
S30. amendment linear unit is introduced in convolutional neural networks to increase nonlinear pattern recognition: linear by introducing amendment
The activation primitive of unit (Rectified Linear Unit), abbreviation ReLU.It is introduced in convolutional neural networks non-linear.S20
In carried out after convolution algorithm obtains Feature Mapping for the first time, each value is by this function, to be activated or be lighted.If defeated
Entering value is negative, then output will be zero.If input value is positive number, output will be as input, and each value passes through
After ReLU, it is created that nonlinear pattern recognition；
S40. two dimensional character matrix is obtained to adjacent feature detection matrix progress dimensionality reduction is obtained by flattening: is examined in adjacent feature
It surveys in matrix and only retains maximum value, to obtain the key message of image；
S50. feature flattening: being that a column vector pixel exports by the two dimensional character matrix conversion of above-mentioned acquisition；
S60. the identification state of the protection pressing plate identified include investment, exit with it is spare.In flattening feature vector and protection pressure
One group of weight of Initialize installation and biasing between board status output valve；These values can be adjusted during " study ", thus
Obtain more accurate state judgement；
S70. the confidence level probability-weighted of the state judgement output of three kinds of protection pressing plates is obtained by Logit and Softmax operation,
Highest probability is that this method judges the protection pressing plate state obtained;
The step of embodiment as a further preference, S70 is specially to be divided into two parts: Logit score and Softmax fortune
It calculates.
Wherein, the specific steps are the Logit score of each output is a basic linear letter to the Logit score
Number: Logit score=(characteristic quantity × weight)+biasing, highest scoring is the conjecture of judgment models；
Wherein, the Softmax calculation step specifically, in order to obtain model conjecture confidence level, using natural logrithm e the bottom of as
Number, is scored at index with Logit, to obtain the confidence level of protection pressing plate state output；Each state output confidence level is removed
With the sum of all confidence scores, confidence level probability-weighted, the highest protection pressing plate for as judging to obtain of probability have just been obtained
State；
S80. cross entropy is used to be used to the gap of comparison prediction and actual value as the loss function of convolutional neural networks, in convolution
During neural metwork training, with network weight/biasing levelling, carry out the accuracy of judgement degree of improved model;
S90. the normal condition of pressing plate will be protected in the state of each protection pressing plate exported in convolutional neural networks and database
Data compare, and identify the protection pressing plate changed.
A kind of protection pressing plate state identification method based on convolutional neural networks provided by the invention carries out input picture
RGB color feature decomposition, and the input as convolutional neural networks；It completes to detect the feature of input picture；Increase non-linear
Pattern-recognition；Pass through the key feature in the detection of maximum pond keeping characteristics；Dimensionality reduction is carried out to eigenmatrix, full connection obtains three
The feature vector of kind state；Protection pressing plate state is obtained by operation；Sentencing for this method is measured by cross entropy loss function
Gap between disconnected output and true pressing plate state, adjusts the ginseng of links by rewards and punishments with certain step-length and direction
Number, to promote the condition discrimination ability of this method.
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description
To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this
Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention
Protection scope within.
Claims (6)
1. a kind of protection pressing plate state identification method based on convolutional neural networks, which comprises the following steps:
S10. the protection pressing plate image for inputting certain location type of substation field carries out RGB color feature decomposition, and by image
The feux rouges of each pixel, green light, blue light pixel number with a matrix type, the input as convolutional neural networks；
S20. feature identification is carried out, convolution algorithm is carried out to pixel group by convolutional neural networks, setting step-length is come to entire defeated
Enter matrix to be scanned, and then completes to detect the feature of input picture；
S30. amendment linear unit is introduced in convolutional neural networks to increase nonlinear pattern recognition；
S40. two dimensional character matrix is obtained to adjacent feature detection matrix progress dimensionality reduction is obtained by flattening: is examined in adjacent feature
It surveys in matrix and only retains maximum value, to obtain the key message of image；
S50. feature flattening: being that a column vector pixel exports by the two dimensional character matrix conversion of above-mentioned acquisition；
S60. one group of weight of Initialize installation and biasing between flattening feature vector and protection pressing plate state output value；
S70. the confidence level probability-weighted of the state judgement output of three kinds of protection pressing plates is obtained by operation, probability is highest to be
This method judges the protection pressing plate state obtained;
S80. cross entropy is used to be used to the gap of comparison prediction and actual value as the loss function of convolutional neural networks, in convolution
During neural metwork training, with network weight/biasing levelling, carry out the accuracy of judgement degree of improved model;
S90. the normal condition of pressing plate will be protected in the state of each protection pressing plate exported in convolutional neural networks and database
Data compare, and identify the protection pressing plate changed.
2. the protection pressing plate state identification method according to claim 1 based on convolutional neural networks, which is characterized in that
The step of S20 specifically:
S201. avoiding the loss of pixel in learning process by way of inputting the zero padding of picture element matrix boundary；
S202. setting weighted convolution matrix carries out convolution algorithm to input picture element matrix, completes to carry out two to input picture element matrix
A Feature Mapping, the weight for initializing convolution algorithm are arranged between 0.01 to 0.1；
S203. it is that 2 pairs of input picture element matrix features are checked comprehensively that setting step, which is moved,.
3. the protection pressing plate state identification method according to claim 1 based on convolutional neural networks, which is characterized in that
The step of S70 specially two parts: Logit score and Softmax operation.
4. the protection pressing plate state identification method according to claim 3 based on convolutional neural networks, which is characterized in that institute
The specific steps are the Logit score of each output is a basic linear function to the Logit score stated: Logit score=spy
Sign amount × weight+biasing, highest scoring are the conjecture of judgment models.
5. the protection pressing plate state identification method according to claim 3 or 4 based on convolutional neural networks, feature exist
In, the Softmax calculation step specifically, the confidence level of model conjecture in order to obtain, using natural logrithm e as the truth of a matter, with
Logit is scored at index, to obtain the confidence level of protection pressing plate state output；By each state output confidence level divided by all
The sum of confidence score has just obtained confidence level probability-weighted, the highest protection pressing plate state for as judging to obtain of probability.
6. described in any item protection pressing plate state identification methods based on convolutional neural networks according to claim 1 ~ 5, special
Sign is, the identification state of the protection pressing plate include investment, exit with it is spare.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811046237.8A CN109472268A (en) | 2018-09-07 | 2018-09-07 | A kind of protection pressing plate state identification method based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811046237.8A CN109472268A (en) | 2018-09-07 | 2018-09-07 | A kind of protection pressing plate state identification method based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109472268A true CN109472268A (en) | 2019-03-15 |
Family
ID=65663030
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811046237.8A Pending CN109472268A (en) | 2018-09-07 | 2018-09-07 | A kind of protection pressing plate state identification method based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109472268A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110675397A (en) * | 2019-10-14 | 2020-01-10 | 国网山东省电力公司泰安供电公司 | Transformer substation protection pressing plate state checking method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105488783A (en) * | 2015-09-25 | 2016-04-13 | 许世昌 | Mobile computing-based system for checking status of electric power secondary pressure plate and using method thereof |
| CN106971182A (en) * | 2017-02-06 | 2017-07-21 | 王兴照 | Embedded electric power relay pressing plate is thrown and moves back state Intelligent Identify device and implementation method |
| CN107808162A (en) * | 2017-09-27 | 2018-03-16 | 广东电网有限责任公司江门供电局 | A kind of transformer substation protection pressure plate state comparison method and device |
| CN107944396A (en) * | 2017-11-27 | 2018-04-20 | 国网安徽省电力有限公司经济技术研究院 | A kind of disconnecting link state identification method based on improvement deep learning |
-
2018
- 2018-09-07 CN CN201811046237.8A patent/CN109472268A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105488783A (en) * | 2015-09-25 | 2016-04-13 | 许世昌 | Mobile computing-based system for checking status of electric power secondary pressure plate and using method thereof |
| CN106971182A (en) * | 2017-02-06 | 2017-07-21 | 王兴照 | Embedded electric power relay pressing plate is thrown and moves back state Intelligent Identify device and implementation method |
| CN107808162A (en) * | 2017-09-27 | 2018-03-16 | 广东电网有限责任公司江门供电局 | A kind of transformer substation protection pressure plate state comparison method and device |
| CN107944396A (en) * | 2017-11-27 | 2018-04-20 | 国网安徽省电力有限公司经济技术研究院 | A kind of disconnecting link state identification method based on improvement deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 陈敏: "《认知计算导论》", 30 April 2017 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110675397A (en) * | 2019-10-14 | 2020-01-10 | 国网山东省电力公司泰安供电公司 | Transformer substation protection pressing plate state checking method |
| CN110675397B (en) * | 2019-10-14 | 2023-03-28 | 国网山东省电力公司泰安供电公司 | Transformer substation protection pressing plate state checking method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107180070B (en) | Automatic risk information classification, identification and early warning method and system | |
| CN109242306A (en) | Safety production risk assessment method and system based on multilevel gray correlation analysis | |
| CN104992223A (en) | Intensive population estimation method based on deep learning | |
| CN107256449A (en) | A kind of relay protection device of intelligent substation state evaluation and appraisal procedure | |
| CN104318198A (en) | Identification method and device suitable for substation robot patrolling | |
| CN107766823A (en) | Anomaly detection method in video based on key area feature learning | |
| CN114913403B (en) | Visual question-answering method based on metric learning | |
| CN112200263B (en) | Self-organizing federal clustering method applied to power distribution internet of things | |
| CN110889587A (en) | Power distribution network line risk assessment method | |
| CN111797892B (en) | Electric power market generator market force monitoring method based on random forest regression | |
| Mongwe et al. | The efficacy of financial ratios for fraud detection using self organising maps | |
| CN109472268A (en) | A kind of protection pressing plate state identification method based on convolutional neural networks | |
| CN114049589A (en) | Transformer substation monitoring system based on artificial intelligence | |
| CN116740654A (en) | Substation operation prevention and control method based on image recognition technology | |
| CN110197475A (en) | Insulator automatic recognition system, method and application in a kind of transmission line of electricity | |
| CN115018291A (en) | Large bridge engineering risk assessment method | |
| CN113822533A (en) | Real-time event-driven risk assessment quantitative model construction method and system | |
| CN112257335B (en) | Oil-immersed transformer fault diagnosis method combining PNN and SVM | |
| CN112734637B (en) | Thermal infrared image processing method and system for monitoring temperature of lead | |
| CN108510002B (en) | Method for detecting short circuit impact resistance of wound core traction transformer winding | |
| CN109460761A (en) | Bank card number detection and recognition methods based on dimension cluster and multi-scale prediction | |
| CN116070790B (en) | Prediction method and system for reducing major safety risk hidden danger of construction site | |
| CN106655181A (en) | Priority setting method and system for power grid nodes | |
| CN116884192A (en) | Power production operation risk early warning method, system and equipment | |
| CN110378407A (en) | Power equipment image-recognizing method and device based on parametric texture and neural network |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190315 |