CN109188198A - A kind of fault data matching process based on convolutional neural networks - Google Patents
A kind of fault data matching process based on convolutional neural networks Download PDFInfo
- Publication number
- CN109188198A CN109188198A CN201811002835.5A CN201811002835A CN109188198A CN 109188198 A CN109188198 A CN 109188198A CN 201811002835 A CN201811002835 A CN 201811002835A CN 109188198 A CN109188198 A CN 109188198A
- Authority
- CN
- China
- Prior art keywords
- neural networks
- convolutional neural
- fault
- data matching
- data
- 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
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/08—Locating faults in cables, transmission lines, or networks
- G01R31/081—Locating faults in cables, transmission lines, or networks according to type of conductors
- G01R31/085—Locating faults in cables, transmission lines, or networks according to type of conductors in power transmission or distribution lines, e.g. overhead
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/08—Locating faults in cables, transmission lines, or networks
- G01R31/088—Aspects of digital computing
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of two ends of electric transmission line fault recorder data matching process.More particularly to a kind of fault data matching process based on convolutional neural networks.Model is built by power system simulation software, by the combination to factors such as system frequency, abort situation, system voltage, fault type, transition resistances, obtains a large amount of training sample and test sample.The preliminary structure of convolutional neural networks is primarily determined according to sample data feature, adjusts batch processing quantity, is increased frequency of training, so that test sample error rate is reached minimum, save optimal convolutional neural networks.When carrying out fault data matching, trained convolutional neural networks are inputted after two sides three-phase current is carried out format analysis processing, matching result can be obtained.
Description
Technical field
The present invention relates to a kind of two ends of electric transmission line fault recorder data matching process.More particularly to it is a kind of refreshing using convolution
Two ends of electric transmission line fault recorder data matching process through network.
Background technique
Containing the fault recorder data of magnanimity in failure information system, often there is timing, clock asynchronism, no
The features such as consistency, imperfection, redundancy, the transmission line of electricity both ends recorder data under same markers not necessarily match
's.Transmission line of electricity both ends fault recorder data is matched, and is applied to Two-terminal Fault Location, protects behavioural analysis, failure
Equivalent verification etc., can preferably play the value of data under playback, accident condition, to accident analysis with fault recovery with important
Meaning.In terms of Data Matching, more advanced at present is to guarantee route both ends number of faults using GPS precision time service system
According to synchronism, ideally the identical both ends fault data of markers is matched data.But in practice due to external interference
It may make that GPS precision time service system generates error and relay protection and what fault oscillograph equipment was equipped with is frequently not high-precision
Clock (crystal oscillator), if for a long time do not calibrate, the time between each equipment might have bigger error, in some instances it may even be possible to meeting
Accumulation has the error of hour grade, and the markers that will lead to homogeneous fault data is not identical.Currently, under markers misalignments, also
Imperfect data matching method.Especially under thunderstorm, strong wind, the extreme externals environment such as congeal, failure is multiple, can experience
Multiple groups failure file can all be will record in a short time to failure and the equipment of disturbance, and upload to failure information system, these
Failure possible breakdown is separate consistent, and the time is close, the multiple groups fault data of these multiple and different equipment of accurate match, for failure
Analysis is very necessary.And multi-group data will appear the match condition of 1:n, if markers is inaccurate, be difficult accurate match correlation
Failure file.With the development of artificial intelligence, deep learning is concerned, handwritten word identification, recognition of face, speech recognition,
Image recognition etc. shows powerful advantage；For electric system, in the side such as wind power prediction, equipment fault diagnosis
Face also obtains preferable effect, but its application in terms of two ends of electric transmission line fault data matching is still blank.And it is saving
Thousands of relay protections and oscillograph equipment can be generally accessed in the failure information system of grade and the above control centre, is store big
The fault data of amount provides good data basis for the use of deep learning.
Summary of the invention
Present invention is generally directed in the case where markers misalignment, imperfect data matching method introduces convolution at present
Neural network carries out fault data matching.Convolutional neural networks (abbreviation CNN) are made of input layer, hidden layer, output layer, hidden
Generally is made of convolution (C) layer, down-sampled (S) layer and full connection (F) layer containing layer, C layers with S layers of alternate combinations, upper one layer defeated
Out as next layer of input, classification expression and feature extraction are carried out to data.Pass through the extraction realization pair to feature in data
The classification of Trouble Match data is few with required electrical quantity and do not need to carry out the advantage of adjusting etc., this method to various parameters
Very little is influenced by factors such as system frequency, transition resistance, abort situation, fault types.
Above-mentioned technical problem of the invention is mainly to be addressed by following technical proposals:
Line fault current data is extracted, after pressing unified format analysis processing to current data, CNN passes through to mass data
It practises, the fast and accurately matching to line fault data may be implemented.More particularly to a kind of failure based on convolutional neural networks
Data matching method, it is characterised in that: steps are as follows:
Step 1, build fault model, simplified simulation model built according to practical transmission system, configured power supply voltage,
The relevant parameter of the parameters such as phase angle, equivalent impedance and power transmission line.Fault type and transition resistance are set, transmission line of electricity two is obtained
Hold three-phase sampled current value.Training sample and test sample are formed after treatment.
Step 2, the structure of convolutional neural networks is primarily determined according to data characteristics.
Step 3, constantly adjustment batch processing number, increase frequency of training make the error rate of test sample tend to 0, save training
Good convolutional neural networks relevant parameter, including network architecture parameters, weight bias matrix etc..
Step 4, after matched fault current data prediction will be needed, it is input to trained network, that is, can determine whether to count
According to whether matching.
It is specific to walk in the step 2 in a kind of above-mentioned fault data matching process based on convolutional neural networks
It is rapid as follows:
Step 2.1, input matrix dimension is determined.Fault oscillograph every cycle current sample points are N, when Data Matching only
The three-phase current data at both ends are needed, therefore the input matrix dimension formed is 2N × 6.
Step 2.2, the number of plies of hidden layer is determined according to matrix feature.Since convolutional layer and down-sampled layer all have dimensionality reduction work
With generally determining the number of plies of hidden layer by minimum dimension.
Step 2.3, the size of convolution kernel is determined according to the hidden layer number of plies.If convolution kernel size is k × k, can make after convolution
Inputting dimension reduces k-1 dimension；It is down-sampled that dimension can be made to reduce (dimension is necessary for positive integer) by multiple.It is possible thereby to which determination is several
The structure of the convolutional neural networks of different size convolution kernel combination.
It is specific to walk in the step 3 in a kind of above-mentioned fault data matching process based on convolutional neural networks
It is rapid as follows:
Step 3.1, in the case where identical batch processing quantity and frequency of training, several convolution determining to step 2.3
Neural network is trained, and the structure for selecting test sample error rate minimum is as optimal convolutional neural networks structure.
Step 3.2, change batch processing quantity, increase frequency of training, further decrease test sample error rate, save instruction
The relevant parameters such as the network structure perfected and weight bias matrix.
The invention has the following advantages that the data matching method based on convolutional neural networks is not substantially by system frequency, mistake
Cross the influence of the factors such as resistance, abort situation, fault type, reliability with higher.And the magnitude of current is only needed, do not need electricity
Pressure amount, required electrical quantity is few, does not need the adjusting for carrying out parameter threshold, has better practicability, robustness and generalization ability.
Detailed description of the invention
Fig. 1 is the simplification Double-End Source transmission system that power system simulation software is built.
Fig. 2 is implementation flow chart of the invention.
Specific embodiment
Implementation process of the invention is illustrated below in conjunction with attached drawing.
Step 1, the transmission system of Double-End Source as shown in Figure 1 is established using electric analog software, it is used herein to make
Simulink simulation software.
Step 2, program is write by MATLAB, to system frequency, abort situation, system voltage, fault type, transition electricity
The factor that resistance etc. influences Data Matching is successively traversed by way of permutation and combination, generates parameter matrix, parameter matrix
Every a line represents the parameter setting under a kind of fault condition.Training sample and the parameter traversals table of test sample are as follows.
1 training sample of table traverses parameter list
2 test sample of table traverses parameter list
Parameter list is traversed by 1 training sample of table it is found that emulation generation data areGroup has matched data
9000 groups, after increasing nonmatched data by random number, the sum of training sample is 18000.Similarly, test sample traversal ginseng
Number table is as shown in table 2, and emulation generates data and isGroup, that is, have 3750 groups of matched data, and the sum of training sample is
7000.The sample rate of emulation is 1.2kHz, i.e., every cycle sampling number is 24, takes each cycle data before and after failure.
In conclusion training sample input matrix is 48 × 6 × 18000 dimensional matrix；Test sample input matrix is 48 × 6 × 7500
Dimensional matrix.
Step 4, the dimension 48 × 6 of this paper input sample matrix, dimension is not high, so using 4 layers of CNN structure, i.e., 2
Convolutional layer and 2 down-sampled layers.Input is unfolded after multiple convolution and down-sampled extraction feature, is connected to a label entirely, will
It is connected with output, if label A is greater than 0.5, exports A and sets 1, indicate Data Matching；Otherwise output sets 0, and data mismatch.
Step 5, setting batch processing number is 50, frequency of training 10, under batch processing number and the identical situation of frequency of training,
Test is trained by the network structure and convolution kernel that are arranged different.Wherein, C indicates that convolutional layer, S indicate down-sampled layer, volume
Two numbers are respectively the convolution kernel dimension of two convolutional layers in product one column of core, and batch processing number indicates every batch of input in training process
Training sample number, error rate is the sample of output result mistake obtained after trained network tests test sample
Originally the percentage of test sample sum is accounted for, test result is as shown in table 3.As shown in Table 3, the corresponding error rate of serial number 4 is lower, therefore
Network structure is 6C-1S-12C-2S, and when convolution kernel is followed successively by 2 and 4, CNN Data Matching has preferable effect.
Data Matching result under 3 difference CNN structure of table
Step 6, under the network structure that step 5 determines, different batch processing number and frequency of training is set, obtained corresponding
Error rate, as shown in table 4.
4 CNN Data Matching result of table
As shown in Table 4, error rate can be reduced to 0.987%, save the structure and weight bias square of the CNN that training is completed
Battle array etc..
It can be seen from the experiment that test sample with the system frequency of training data, transition resistance, abort situation, fault type
In the case where difference, still there is very low matching error rate.This illustrate CNN have very strong generalization ability, substantially not by
The influence of the factors such as system frequency, transition resistance, abort situation, fault type, reliability with higher.
Specific embodiment described herein is only an example for the spirit of the invention.The neck of technology belonging to the present invention
The technical staff in domain can make various modifications or additions to the described embodiments or replace by a similar method
In generation, however, it does not deviate from the spirit of the invention or beyond the scope of the appended claims.
Claims (3)
1. a kind of fault data matching process based on convolutional neural networks, it is characterised in that: including
Step 1, fault model is built, simplified simulation model is built according to practical transmission system, has configured voltage, the phase of power supply
The parameter at angle, equivalent impedance parameter and power transmission line；Fault type and transition resistance are set, obtains transmission line of electricity both ends three-phase and adopts
Sample current value forms training sample and test sample；
Step 2, the structure of convolutional neural networks is primarily determined according to data characteristics；
Step 3, constantly adjustment batch processing number, increase frequency of training make the error rate of test sample tend to 0, save trained
Convolutional neural networks relevant parameter, including network architecture parameters, weight bias matrix, obtain trained network；
Step 4, after matched fault current data prediction will be needed, it is input to trained network, that is, can determine whether that data are
No matching.
2. a kind of fault data matching process based on convolutional neural networks according to claim 1, it is characterised in that:
In the step 2, the specific steps are as follows:
Step 2.1, input matrix dimension is determined；Fault oscillograph every cycle current sample points are N, and when Data Matching only needs
The three-phase current data at both ends, therefore the input matrix dimension formed is 2N × 6；
Step 2.2, the number of plies of hidden layer is determined according to matrix feature；Since convolutional layer and down-sampled layer all have dimensionality reduction effect,
The number of plies of hidden layer is generally determined by minimum dimension；
Step 2.3, the size of convolution kernel is determined according to the hidden layer number of plies；If convolution kernel size is k × k, can make to input after convolution
Dimension reduces k-1 dimension；It is down-sampled dimension to be made to reduce by multiple；It is possible thereby to determine the volume of several different size convolution kernel combinations
The structure of product neural network.
3. a kind of fault data matching process based on convolutional neural networks according to claim 1, it is characterised in that:
In the step 3, the specific steps are as follows:
Step 3.1, in the case where identical batch processing quantity and frequency of training, several convolutional Neurals determining to step 2.3
Network is trained, and the structure for selecting test sample error rate minimum is as optimal convolutional neural networks structure；
Step 3.2, change batch processing quantity, increase frequency of training, further decrease test sample error rate, preservation trains
Network structure and weight bias matrix correlation parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811002835.5A CN109188198A (en) | 2018-08-30 | 2018-08-30 | A kind of fault data matching process based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811002835.5A CN109188198A (en) | 2018-08-30 | 2018-08-30 | A kind of fault data matching process based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109188198A true CN109188198A (en) | 2019-01-11 |
Family
ID=64917388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811002835.5A Pending CN109188198A (en) | 2018-08-30 | 2018-08-30 | A kind of fault data matching process based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109188198A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109917223A (en) * | 2019-03-08 | 2019-06-21 | 广西电网有限责任公司电力科学研究院 | A kind of transmission line malfunction current traveling wave feature extracting method |
| CN110031227A (en) * | 2019-05-23 | 2019-07-19 | 桂林电子科技大学 | A kind of Rolling Bearing Status diagnostic method based on binary channels convolutional neural networks |
| CN110095689A (en) * | 2019-05-10 | 2019-08-06 | 广东工业大学 | A kind of method of discrimination of fault direction, system and equipment |
| CN110533331A (en) * | 2019-08-30 | 2019-12-03 | 广东电网有限责任公司江门供电局 | A kind of fault early warning method and system based on transmission line of electricity data mining |
| CN112381141A (en) * | 2020-11-13 | 2021-02-19 | 西安建筑科技大学 | Air conditioner sensor fault detection method and system |
| CN113392589A (en) * | 2021-06-30 | 2021-09-14 | 云南电网有限责任公司电力科学研究院 | High-voltage direct-current converter station fault analysis method and system based on convolutional neural network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022317A (en) * | 2016-06-27 | 2016-10-12 | 北京小米移动软件有限公司 | Face identification method and apparatus |
| CN106291233A (en) * | 2016-07-29 | 2017-01-04 | 武汉大学 | A kind of fault phase-selecting method based on convolutional neural networks |
| CN106980073A (en) * | 2017-05-23 | 2017-07-25 | 武汉大学 | A kind of two ends of electric transmission line fault recorder data matching process based on convolutional neural networks |
| CN107341611A (en) * | 2017-07-06 | 2017-11-10 | 浙江大学 | A kind of operation flow based on convolutional neural networks recommends method |
| CN107607992A (en) * | 2017-08-24 | 2018-01-19 | 电子科技大学 | More ripple matching process based on convolutional neural networks |
| CN107992844A (en) * | 2017-12-14 | 2018-05-04 | 合肥寰景信息技术有限公司 | Face identification system and method based on deep learning |
| CN108133222A (en) * | 2016-12-01 | 2018-06-08 | 富士通株式会社 | The device and method that convolutional neural networks CNN models are determined for database |
-
2018
- 2018-08-30 CN CN201811002835.5A patent/CN109188198A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022317A (en) * | 2016-06-27 | 2016-10-12 | 北京小米移动软件有限公司 | Face identification method and apparatus |
| CN106291233A (en) * | 2016-07-29 | 2017-01-04 | 武汉大学 | A kind of fault phase-selecting method based on convolutional neural networks |
| CN108133222A (en) * | 2016-12-01 | 2018-06-08 | 富士通株式会社 | The device and method that convolutional neural networks CNN models are determined for database |
| CN106980073A (en) * | 2017-05-23 | 2017-07-25 | 武汉大学 | A kind of two ends of electric transmission line fault recorder data matching process based on convolutional neural networks |
| CN107341611A (en) * | 2017-07-06 | 2017-11-10 | 浙江大学 | A kind of operation flow based on convolutional neural networks recommends method |
| CN107607992A (en) * | 2017-08-24 | 2018-01-19 | 电子科技大学 | More ripple matching process based on convolutional neural networks |
| CN107992844A (en) * | 2017-12-14 | 2018-05-04 | 合肥寰景信息技术有限公司 | Face identification system and method based on deep learning |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109917223A (en) * | 2019-03-08 | 2019-06-21 | 广西电网有限责任公司电力科学研究院 | A kind of transmission line malfunction current traveling wave feature extracting method |
| CN110095689A (en) * | 2019-05-10 | 2019-08-06 | 广东工业大学 | A kind of method of discrimination of fault direction, system and equipment |
| CN110031227A (en) * | 2019-05-23 | 2019-07-19 | 桂林电子科技大学 | A kind of Rolling Bearing Status diagnostic method based on binary channels convolutional neural networks |
| CN110533331A (en) * | 2019-08-30 | 2019-12-03 | 广东电网有限责任公司江门供电局 | A kind of fault early warning method and system based on transmission line of electricity data mining |
| CN110533331B (en) * | 2019-08-30 | 2024-03-15 | 广东电网有限责任公司江门供电局 | Fault early warning method and system based on transmission line data mining |
| CN112381141A (en) * | 2020-11-13 | 2021-02-19 | 西安建筑科技大学 | Air conditioner sensor fault detection method and system |
| CN113392589A (en) * | 2021-06-30 | 2021-09-14 | 云南电网有限责任公司电力科学研究院 | High-voltage direct-current converter station fault analysis method and system based on convolutional neural network |
| CN113392589B (en) * | 2021-06-30 | 2022-09-27 | 云南电网有限责任公司电力科学研究院 | High-voltage direct-current converter station fault analysis method and system based on convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109188198A (en) | A kind of fault data matching process based on convolutional neural networks | |
| US11544426B2 (en) | Systems and methods for enhanced sequential power system model parameter estimation | |
| US10706332B2 (en) | Analog circuit fault mode classification method | |
| CN106980073A (en) | A kind of two ends of electric transmission line fault recorder data matching process based on convolutional neural networks | |
| CN112509564A (en) | End-to-end voice recognition method based on connection time sequence classification and self-attention mechanism | |
| CN108107324A (en) | A kind of electrical power distribution network fault location method based on depth convolutional neural networks | |
| CN106646140B (en) | HVDC transmission line internal fault external fault recognition methods based on measurement wave impedance | |
| CN102636742B (en) | Large-scale analogue circuit fault diagnosis method based on wavelet neural network | |
| CN108520301A (en) | A kind of circuit intermittent fault diagnostic method based on depth confidence network | |
| CN106470122A (en) | A kind of network failure locating method and device | |
| CN110838075A (en) | Training and predicting method and device for prediction model of transient stability of power grid system | |
| CN108154223A (en) | Power distribution network operating mode recording sorting technique based on network topology and long timing information | |
| CN106779066A (en) | A kind of radar circuit plate method for diagnosing faults | |
| CN108563634A (en) | Recognition methods, system, computer equipment and the storage medium of word misspelling | |
| CN115587329A (en) | Power distribution network fault classification method and system based on convolutional neural network extraction features | |
| CN111949459B (en) | Hard disk failure prediction method and system based on transfer learning and active learning | |
| Todeschini et al. | An image-based deep transfer learning approach to classify power quality disturbances | |
| CN113488997A (en) | Micro-grid critical energy evaluation method based on disturbance response multi-scale features | |
| CN109145764A (en) | The unjustified section recognition methods of the multiple groups detection waveform of synthetic detection vehicle and device | |
| CN115494349B (en) | Active power distribution network single-phase earth fault section positioning method | |
| CN115795360A (en) | Cable fault detection method based on artificial neural network | |
| US20240045752A1 (en) | Methods and apparatuses for troubleshooting a computer system | |
| CN115563464A (en) | Nuclear valve abnormal state identification method based on quantum convolution neural network | |
| CN113379279A (en) | Deep reinforcement learning short-term voltage stability evaluation method based on incomplete data | |
| Liu et al. | Practical Methods of Defective Input Feature Correction to Enable Machine Learning in Power Systems |
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 |
Application publication date: 20190111 |
|
| RJ01 | Rejection of invention patent application after publication |