CN109683066A - Power cable typical defect local discharge signal recognition methods - Google Patents
Power cable typical defect local discharge signal recognition methods Download PDFInfo
- Publication number
- CN109683066A CN109683066A CN201811326013.2A CN201811326013A CN109683066A CN 109683066 A CN109683066 A CN 109683066A CN 201811326013 A CN201811326013 A CN 201811326013A CN 109683066 A CN109683066 A CN 109683066A
- Authority
- CN
- China
- Prior art keywords
- discharge
- parameter
- threshold value
- electric discharge
- weight
- 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
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (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)
- Testing Relating To Insulation (AREA)
Abstract
The invention discloses power cable typical defect local discharge signal recognition methods, it include: the Partial Discharge Data for acquiring known type, characteristic parameter is extracted as input parameter, the information bank of shelf depreciation type identification is stored in for every kind of electric discharge type feature-set electric discharge type label;The neural network model of electric discharge type for identification is built, neural network model is modified using weight and threshold value of the ant colony algorithm to network and obtains optimum model parameter input weight, hidden layer threshold value and output weight, and saves optimum model parameter;Based on optimum model parameter, power cable partial discharge signal to be identified is acquired, extracts discharge pulse characteristic parameter, discharge characteristic parameter to be identified is inputted to the neural network model put up, identification is carried out and obtains electric discharge type.Using the power threshold value of artificial bee colony algorithm optimization ExtremeLearningMachine, weight is exported using obtained optimal power threshold calculations, improves the generalization ability and accuracy of identification of ExtremeLearningMachine.
Description
Technical field
This disclosure relates to which signal processing technology field, identifies more particularly to power cable typical defect local discharge signal
Method and system.
Background technique
As sustained economic growth, urban distribution network are grown rapidly, the quantity that cable run puts into operation is being quicklyd increase.Electricity
The operation conditions of cable directly influences the safety of electric system.In numerous cable monitoring means, shelf depreciation test can
Relatively intuitive effective reflection influences the defect of cable life and safe operation.Shelf depreciation is as Voltage Cable Lines Construction insulation fault
The main forms of early stage, be both the main reason for causing insulation ag(e)ing and characterization insulation status main characteristic parameters,
It is significant to Fault Diagnosis for Electrical Equipment.Shelf depreciation is a kind of common electrical discharge phenomenon, it is that electrical equipment exists
Insulate a kind of sign being damaged when longtime running.
Since cable laying is embedded in underground in cable duct or directly, laying environment and use state will greatly affect cable
Service life.Long-term same soil, moisture, moisture contact, the office insulated vulnerable to corrosion penetration, when along with cables manufacturing or installation
Portion's defect, all may cause failure.Buried cable once breaks down.Find get up it is very difficult, not only to waste a large amount of people
Power material resources, but also the loss of outage for being difficult to estimate will be brought.If failure cannot exclude in time, it will cause serious warp
Ji loss and social influence.
Inventor has found that shelf depreciation type is closely related with insulation defect under study for action, on-line checking High-Voltage Electrical Appliances fortune
Row state acquires insulated local discharge signal in real time and carries out mathematical analysis processing and attributive classification to it, infers, prediction insulation
Rejected region, partial discharge type machine electric discharge development degree, can forecast generation of preventing accident.Since there are many classes for shelf depreciation
Type, for example, it is internal, along face, corona etc., in addition the interference of noise, respectively corresponds different diagnosis and Strategies of Maintenance.Electric discharge letter
Number identification purpose seek to accurately find out signal type, judge its harmfulness, carry out successfully managing measure.
By the retrieval to Patents documents, inventor is as follows for pertinent literature brief analysis:
Existing patent document, number of patent application are that " 201810088179.9 " patent name is that " one kind is for cable part
The method and system of discharge mode identification ", in this document, mainly highlights at the dimensionality reduction to the local discharge characteristic of extraction
Reason.In addition, the document uses BP neural network and support vector machines forming types recognition classifier in terms of classifier.And support to
Amount machine is a kind of two classifiers, to solve shelf depreciation type identification, need to introduce kernel function, once kernel function is decided,
Mapping mode just uniquely determines, and removes non-selection other kernel function, the solution found in more classification problems not necessarily optimal solution.
It is slow compared to ExtremeLearningMachine training speed and in more classification problems.
Existing patent document, number of patent application are that " 2015104825900 " patent name is a kind of " electric cable fitting event
Hinder recognition methods and system ", which extracts 12 characteristic quantities as fisrt feature parameter in local discharge characteristic extraction,
6 characteristic quantities are extracted after dimensionality reduction as second feature parameter, and then the first, second characteristic parameter is identified respectively.This article
In offering, in 12 characteristic quantities of fisrt feature parameter, some characteristic quantities are to be mutually related, these features that are mutually related
Amount only needs one of them that can represent other characteristic quantity.And the second feature amount after dimensionality reduction is only really identification electric discharge type
Effectively not repeated characteristic.In addition, the document is in terms of constructing sorter model, with BP neural network, ExtremeLearningMachine and
Support vector machines respectively identifies characteristic quantity, and in practical application, and impracticable, can only be as verifying in previous work
Means, in practical applications, these three classifiers that the document is used should select the most efficiently highest classification of accuracy of identification
Device, rather than whether the result of more each classifier is identical in actual test, and is only modeled with sorter model,
On the generalization ability and accuracy of identification the problem of, effect is extremely difficult to optimal.
Summary of the invention
In order to solve the deficiencies in the prior art, there is provided power cable typical defects locally to put for an aspect of this disclosure
Electric signal recognition methods can be realized and effectively be identified to local discharge signal.
To achieve the goals above, the application uses following technical scheme:
Power cable typical defect local discharge signal recognition methods, comprising:
The Partial Discharge Data of known type is acquired, extracts characteristic parameter as input parameter, for every kind of electric discharge type
Feature-set electric discharge type label is stored in the information bank of shelf depreciation type identification;
The neural network model for building electric discharge type for identification is carried out using weight and threshold value of the ant colony algorithm to network
It corrects and obtains optimum model parameter input weight, hidden layer threshold value and output weight, and save optimum model parameter;
Based on optimum model parameter, power cable partial discharge signal to be identified is acquired, extracts discharge pulse feature ginseng
Discharge characteristic parameter to be identified is inputted the neural network model put up by number, is carried out identification and is obtained electric discharge type.
Further technical solution, electric discharge type include that the electric discharge of electricity tree, suspended discharge, bubble electric discharge and humidified insulation are put
Electricity.
When building the neural network model of electric discharge type for identification, for every kind of electric discharge type, several electric discharge letters are acquired
Number, a part is used to train neural network, and a part is used to test neural network.
For each electric discharge type, partial discharge pulse's mean value, partial discharge pulse's variance, shelf depreciation are extracted respectively
The cross correlation of pulse standout or dispersion degree, discharge pulse degree of skewness, that is, left-right asymmetry degree, discharge pulse profile difference
Several and discharge pulse amplitude distribution characteristic 6 characteristic attributes of Weibull form parameter are as characteristic quantity;
Neural network model uses ExtremeLearningMachine, according to the number of the discharge characteristic measure feature attribute of extraction, electric discharge class
The number of type determines ExtremeLearningMachine input number and output number.
Further technical solution, ant colony algorithm optimize the input weight of ExtremeLearningMachine and the process of threshold value, calculate
Process description is as follows:
(1) dimension that each individual to be optimized is determined according to input node number and hidden node number, should be (n+1)
× m, wherein n is input node number, and m is hidden node number;
(2) parameter initialization determines the parameter of ant colony algorithm, including population scale, maximum number of iterations and termination condition;
(3) fitness value is calculated, the fitness value of each individual is calculated, then sorts, finds out optimal to fitness value
Individual;
(4) optimization algorithm iteration optimizing acquires learning process, Population Regeneration by employing bee and looking around the food source of bee
Body position;If not reaching termination condition, returns to (3) and continue optimizing;Otherwise, global optimum's individual is found;
(5) optimizing terminates, and saves optimum individual value, i.e., optimal input weight and hidden layer threshold value;
(6) optimal power threshold value is brought into ExtremeLearningMachine, acquires optimal output weight matrix, it is defeated by what is be calculated
Weight and input weight and hidden layer threshold value are saved as optimum model parameter out.
Further technical solution, ExtremeLearningMachine are calculated output by each layer, are optimized using artificial bee colony algorithm
The power threshold value of ExtremeLearningMachine obtains most preferably inputting weight and hidden layer threshold value, output weight is calculated, saves as best model
Parameter, if input node number is n, hidden node number is m, then the dimension of individual to be optimized is (n+1) × m, and
Individual to be optimized can be expressed as αi=[ω11,ω12,…,ω1n,…,ωmn,b1,…,bm], αiFor i-th of to be optimized
Body, wherein ωmnFor the weight of input layer and hidden layer, biFor the threshold value of hidden layer, ωmn∈ [- 1,1], bi∈[0,1]。
Using the root-mean-square error function between the output valve and target value of ExtremeLearningMachine as fitness function, such as following formula
It is shown:
G () is hidden layer activation primitive, t in formulajFor j-th of target sample value, m is the node number of hidden layer, NtrainFor
The number of training sample, ωiFor the weight of input layer and hidden layer, biFor hidden layer threshold value, βiFor the weight of hidden layer and output layer.
Further technical solution, hidden layer activation primitive use Sigmoid function
Wherein, z be the weighted signal that receives of neuron and;B is the threshold value of neuron.
Another aspect of the present disclosure is to disclose power cable typical defect local discharge signal identifying system, comprising:
Signal pickup assembly acquires the Partial Discharge Data of known type, extracts characteristic parameter and is used as input parameter, for
Every kind of electric discharge type feature-set electric discharge type label is stored in the information bank of shelf depreciation type identification;
Neural network model builds device, builds the neural network model of electric discharge type for identification, uses ant colony algorithm
The weight and threshold value of network are modified and obtain optimum model parameter input weight, hidden layer threshold value and output weight, and is protected
Deposit optimum model parameter;
Electric discharge type identification device is based on optimum model parameter, acquires power cable partial discharge signal to be identified, mention
Electric pulse characteristic parameter is picked and placed, discharge characteristic parameter to be identified is inputted to the neural network model put up, identification is carried out and obtains
Obtain electric discharge type.
Compared with prior art, the beneficial effect of the disclosure is:
After the disclosure uses artificial bee colony algorithm, using artificial bee colony algorithm to the input weight of ExtremeLearningMachine model
It is optimized with hidden layer threshold value, obtains optimal power threshold value, and obtain the output of ExtremeLearningMachine by optimal power threshold calculations
Weight obtains optimal model parameter with this.This method improves ExtremeLearningMachine model accuracy of identification, and generalization ability is more
By force.
Detailed description of the invention
The accompanying drawings constituting a part of this application is used to provide further understanding of the present application, and the application's shows
Meaning property embodiment and its explanation are not constituted an undue limitation on the present application for explaining the application.
Fig. 1 is that the neural network model of some examples of implementation of the disclosure is the structure chart of ExtremeLearningMachine;
Fig. 2 is the time domain waveform of the tree electric discharge of electricity caused by the power cable defect of some examples of implementation of the disclosure;
Fig. 3 is the time domain waveform of suspended discharge caused by the power cable defect of some examples of implementation of the disclosure;
Fig. 4 is the time domain waveform of bubble electric discharge caused by the power cable defect of some examples of implementation of the disclosure
Fig. 5 is the time domain impulse wave that humidified insulation caused by the power cable defect of some examples of implementation of the disclosure discharges
Shape;
Fig. 6 is the flow chart that the ant colony algorithm of some examples of implementation of the disclosure optimizes the input power threshold value of ExtremeLearningMachine;
Fig. 7 is the flow chart of the entire shelf depreciation type identification of some examples of implementation of the disclosure;
Fig. 8 is the shelf depreciation type identification precision of some examples of implementation of the disclosure;
Fig. 9 is the shelf depreciation type identification result of some examples of implementation of the disclosure.
Specific embodiment
It is noted that following detailed description is all illustrative, it is intended to provide further instruction to the application.Unless another
It indicates, all technical and scientific terms used herein has usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.
In a kind of typical embodiment of the application, as shown in fig. 7, being the process of entire shelf depreciation type identification
Figure, provides power cable typical defect local discharge signal recognition methods, comprising:
Step 1: the Partial Discharge Data of acquisition known type, extracts characteristic parameter as input parameter, every kind is put
Electric type feature sets electric discharge type label, is stored in the information bank of shelf depreciation type identification.
Step 2: building the neural network model of electric discharge type for identification, joined using artificial bee colony algorithm Optimized model
Number.Firstly, the parameter of initialization ant colony algorithm, such as population scale, maximum number of iterations, " Limit " value.Then, it initializes
Individual to be optimized, i.e. the input weight and threshold value of ExtremeLearningMachine.Finally, carrying out optimizing according to preset fitness function, obtain
Optimal model parameter is obtained, and saves optimum model parameter.
Step 3: being based on optimum model parameter, discharge data to be identified is acquired, extracts characteristic parameter, it will be to be identified
Discharge characteristic parameter inputs the neural network model put up, and is identified.
The disclosure believes shelf depreciation as classifier in specific embodiment, using ant colony algorithm optimization neural network
It number is identified.Local discharge signal is acquired by partial discharge detection equipment first, collected discharge signal is carried out at denoising
Reason extracts discharge pulse characteristic quantity in time domain waveform, obtains eigenmatrix, the input as neural network.Due to nerve net
The setting of the hidden layer weight and threshold value of network model needs to update, and is modified using weight and threshold value of the ant colony algorithm to network.
Wherein neural network model is ExtremeLearningMachine, and structure chart is as shown in Figure 1.
Fig. 2-Fig. 5 is shelf depreciation caused by four kinds of different power cable defects.Fig. 2 is the time domain wave of electricity tree electric discharge
Shape figure, as seen from Figure 2, the pulse amplitude of electricity tree electric discharge are 0.5-0.6V, and the bandwidth of single discharge pulse is 18-20us.
Fig. 3 is the time domain waveform of suspended discharge, and as seen from Figure 3, the pulse amplitude of suspended discharge is 0.3-0.4V, single to discharge
The bandwidth of pulse is 10-15us.Fig. 4 is the time domain waveform of bubble electric discharge, as seen from Figure 4, the pulse width of bubble electric discharge
Value is 0.3V or so, and the bandwidth of single discharge pulse is 20us or so.Fig. 5 is the time domain impulse waveform of humidified insulation electric discharge, by
The pulse amplitude that Fig. 5 can be seen that discharge in insulation is 150mv, and the bandwidth of single discharge pulse is 15us.
In above-described embodiment, when building the neural network model of electric discharge type for identification, for each electric discharge
Type extracts partial discharge pulse's rise time, partial discharge pulse's fall time, partial discharge pulse 50% most substantially respectively
It is worth pulse duration, 6 discharge pulse total duration, discharge pulse mean value and discharge pulse variance characteristic attribute conducts
Characteristic quantity.The identification of electric discharge type is modeled using ExtremeLearningMachine, in order to illustrate the effect of this method, is discharged for every kind
Type collects and records 80 groups, extracts characteristic quantity according to feature extracting method, wherein 60 groups are used to train neural network, in addition
20 groups are used to test the accuracy of identification of neural network.Wherein, table 1 is the output of four kinds of electric discharge types.
Table 1
In above-described embodiment, the model that the modeling of shelf depreciation type identification uses is ExtremeLearningMachine, compared to tradition
The method of gradient decline weight that uses of neural network, least square method is utilized in ExtremeLearningMachine, and this new method makes
With enhancing the pace of learning of neural network to a certain extent.Compared to traditional BP neural network, the ginseng of ExtremeLearningMachine
Number only needs setting hidden layer number, and when algorithm operation, ExtremeLearningMachine can be randomly provided algorithm hidden layer weight and hidden layer biases, tool
There is faster speed.After the threshold value of given one determining input weight and hidden layer, there is no need to other for ExtremeLearningMachine
Given value, output weight are obtained with least-squares calculation.Herein since the discharge characteristic amount of extraction is 6 feature categories
Property, so input number is 6, hidden layer number is set as 10, and output number is 4, respectively represents 4 kinds of electric discharge types, and hidden layer activates letter
Number is selected as " sigmoid " function.
Specifically, output is calculated by each layer in ExtremeLearningMachine, artificial bee colony algorithm optimization ExtremeLearningMachine is used
Power threshold value, export weight using obtained optimal power threshold calculations, improve the generalization ability and accuracy of identification of ExtremeLearningMachine.
In the specific implementation, using the power threshold value of ant colony algorithm optimization ExtremeLearningMachine.If input node number is n,
Hidden node number is m, then the dimension of individual to be optimized is (n+1) × m, and individual to be optimized can be expressed as αi=
[ω11,ω12,…,ω1n,…,ωmn,b1,…,bm], αiFor i-th of individual to be optimized, wherein ωmnFor input layer and hidden layer
Weight, biFor the threshold value of hidden layer, ωmn∈ [- 1,1], bi∈[0,1]。
Using the root-mean-square error function between the output valve and target value of model as fitness function, it is shown below:
G () is hidden layer activation primitive, t in formulajFor j-th of target sample value, m is the node number of hidden layer, NtrainFor
The number of training sample, ωiFor the weight of input layer and hidden layer, biFor hidden layer threshold value, βiFor the weight of hidden layer and output layer.It is hidden
Layer activation primitive uses Sigmoid function
Wherein, z be the weighted signal that receives of hidden neuron and;B is the threshold value of hidden neuron.
In a particular embodiment, ant colony algorithm optimization ExtremeLearningMachine input power threshold value flow chart as shown in fig. 6,
Its calculating process is described as follows:
(1) dimension that each individual to be optimized is determined according to input node number and hidden node number, should be (n+1)
× m, wherein n is input node number, and m is hidden node number.
(2) parameter initialization.Determine the parameter of ant colony algorithm, such as population scale, maximum number of iterations and termination condition.
(3) fitness value is calculated, the fitness value of each individual is calculated, then sorts, finds out optimal to fitness value
Individual.
(4) optimization algorithm iteration optimizing.Learning process, Population Regeneration are acquired by employing bee and looking around the food source of bee
Body position.If not reaching termination condition, returns to (3) and continue optimizing;Otherwise, global optimum's individual is found.
(5) optimizing terminates, and saves optimum individual value, i.e., optimal input weight and hidden layer threshold value.
(6) optimal power threshold value is brought into ExtremeLearningMachine, acquires optimal output weight matrix.It is defeated by what is be calculated
Weight and input weight and hidden layer threshold value are saved as optimum model parameter out.(power threshold value refers to the weight of hidden layer and input layer
With hidden layer threshold value, export weight be by optimization after input weight and threshold calculations obtain, finally all save as model parameter)
Due to the discharge signal that practical power cable collects, about 18 kinds of characteristic quantities can be extracted, and numerous
Person's research is it has been proved that have several characteristic parameters of optimal classification ability, so the disclosure, which is directly selected, effectively to be identified
The parameter of electric discharge type reduces calculation amount as characteristic quantity, increases calculating speed.
The technical solution of the disclosure uses ExtremeLearningMachine, and the connection weight of input layer and implicit interlayer and hidden is randomly generated
The threshold value of the neuron containing layer, and the number of hidden layer neuron need to be only set, can be obtained only without adjustment in the training process
One optimal solution, compared with traditional BP neural network algorithm, ExtremeLearningMachine method pace of learning is fast, Generalization Capability is good.
Characteristic quantity selected by this patent is exactly to remove those characteristic quantities that are mutually related, and only use can characterize electric discharge type most
Effective characteristic quantity reduces calculation amount as characteristic parameter.This patent is using artificial bee colony algorithm to ExtremeLearningMachine classifier
The power threshold value of model optimizes, and can solve this problem.
Another embodiment of the present disclosure, discloses power cable typical defect local discharge signal identifying system, comprising:
Signal pickup assembly acquires the Partial Discharge Data of known type, extracts characteristic parameter and is used as input parameter, for
Every kind of electric discharge type feature-set electric discharge type label is stored in the information bank of shelf depreciation type identification;
Neural network model builds device, builds the neural network model of electric discharge type for identification, uses ant colony algorithm
Optimum model parameter input weight and hidden layer threshold value, and output weight square are modified and obtained to the weight and threshold value of network
Battle array, and save optimum model parameter;
Electric discharge type identification device is based on optimum model parameter, acquires power cable partial discharge signal to be identified, mention
Electric pulse characteristic parameter is picked and placed, discharge characteristic parameter to be identified is inputted to the neural network model put up, identification is carried out and obtains
Obtain electric discharge type.
In order to prove the validity of this method, every kind of electric discharge type is chosen wherein 80 groups of data and is emulated, wherein at random
60 groups of every kind of electric discharge type are chosen as training data, remaining 20 groups respectively as test data.Fig. 8 is 240 groups of training datas,
The optimal classification precision reached by training.Fig. 9 is the recognition result of 80 groups of test datas.
The foregoing is merely preferred embodiment of the present application, are not intended to limit this application, for the skill of this field
For art personnel, various changes and changes are possible in this application.Within the spirit and principles of this application, made any to repair
Change, equivalent replacement, improvement etc., should be included within the scope of protection of this application.
Claims (10)
1. power cable typical defect local discharge signal recognition methods, characterized in that include:
The Partial Discharge Data of known type is acquired, extracts characteristic parameter as input parameter, for every kind of electric discharge type feature
Electric discharge type label is set, the information bank of shelf depreciation type identification is stored in;
The neural network model for building electric discharge type for identification is modified using weight and threshold value of the ant colony algorithm to network
And optimum model parameter input weight, hidden layer threshold value and output weight are obtained, and save optimum model parameter;
Based on optimum model parameter, power cable partial discharge signal to be identified is acquired, extracts discharge pulse characteristic parameter, it will
Discharge characteristic parameter to be identified inputs the neural network model put up, and carries out identification and obtains electric discharge type.
2. power cable typical defect local discharge signal recognition methods as described in claim 1, characterized in that electric discharge type
Including the electric discharge of electricity tree, suspended discharge, bubble electric discharge and humidified insulation electric discharge.
3. power cable typical defect local discharge signal recognition methods as described in claim 1, characterized in that build and be used for
When identifying the neural network model of electric discharge type, for every kind of electric discharge type, several discharge signals are acquired, a part is used to train
Neural network, a part are used to test neural network.
4. power cable typical defect local discharge signal recognition methods as claimed in claim 3, characterized in that for each
Kind electric discharge type, extracts partial discharge pulse's mean value, partial discharge pulse's variance, partial discharge pulse's standout or discrete respectively
Degree, discharge pulse degree of skewness, that is, left-right asymmetry degree, the cross-correlation coefficient of discharge pulse profile difference and discharge pulse width
6 characteristic attributes of Weibull form parameter of Distribution value characteristic are as characteristic quantity.
5. power cable typical defect local discharge signal recognition methods as described in claim 1, characterized in that neural network
Model uses ExtremeLearningMachine, is determined according to the number of the discharge characteristic measure feature attribute of extraction, the number of electric discharge type extreme
Learning machine inputs number and output number.
6. power cable typical defect local discharge signal recognition methods as described in claim 1, characterized in that ant colony algorithm
Optimize the input weight of ExtremeLearningMachine and the process of threshold value, calculating process is described as follows:
(1) dimension that each individual to be optimized is determined according to input node number and hidden node number, should be (n+1) × m,
Wherein n is input node number, and m is hidden node number;
(2) parameter initialization determines the parameter of ant colony algorithm, including population scale, maximum number of iterations and termination condition;
(3) fitness value is calculated, the fitness value of each individual is calculated, then sorts to fitness value, finds out optimum individual;
(4) optimization algorithm iteration optimizing acquires learning process, Population Regeneration position by employing bee and looking around the food source of bee
It sets;If not reaching termination condition, returns to (3) and continue optimizing;Otherwise, global optimum's individual is found;
(5) optimizing terminates, and saves optimum individual value, i.e., optimal input weight and hidden layer threshold value;
(6) optimal power threshold value is brought into ExtremeLearningMachine, acquires optimal output weight matrix, the output being calculated is weighed
Value and input weight and hidden layer threshold value are saved as optimum model parameter.
7. power cable typical defect local discharge signal recognition methods as described in claim 1, characterized in that extreme study
Output is calculated by each layer in machine, using the power threshold value of artificial bee colony algorithm optimization ExtremeLearningMachine, obtains most preferably inputting power
Value and hidden layer threshold value, are calculated output weight, save as optimum model parameter, if input node number is n, hidden node
Number is m, then the dimension of individual to be optimized is (n+1) × m, and individual to be optimized can be expressed as αi=[ω11,
ω12,…,ω1n,…,ωmn,b1,…,bm], αiFor i-th of individual to be optimized, wherein ωmnFor the power of input layer and hidden layer
Value, biFor the threshold value of hidden layer, ωmn∈ [- 1,1], bi∈[0,1]。
8. power cable typical defect local discharge signal recognition methods as claimed in claim 7, characterized in that extremely to learn
Root-mean-square error function between the output valve and target value of habit machine is shown below as fitness function:
G () is hidden layer activation primitive, t in formulajFor j-th of target sample value, m is the node number of hidden layer, NtrainFor training
The number of sample, ωiFor the weight of input layer and hidden layer, biFor hidden layer threshold value, βiFor the weight of hidden layer and output layer.
9. power cable typical defect local discharge signal recognition methods as claimed in claim 8, characterized in that hidden layer activation
Function uses Sigmoid function
Wherein, z be the weighted signal that receives of neuron and;B is the threshold value of neuron.
10. power cable typical defect local discharge signal identifying system, characterized in that include:
Signal pickup assembly acquires the Partial Discharge Data of known type, extracts characteristic parameter as input parameter, for every kind
Electric discharge type feature-set electric discharge type label is stored in the information bank of shelf depreciation type identification;
Neural network model builds device, the neural network model of electric discharge type for identification is built, using ant colony algorithm to net
The weight and threshold value of network are modified and obtain optimum model parameter input weight, hidden layer threshold value and output weight, and save most
Good model parameter;
Electric discharge type identification device is based on optimum model parameter, acquires power cable partial discharge signal to be identified, extraction is put
Discharge characteristic parameter to be identified is inputted the neural network model put up by electric pulse characteristic parameter, is carried out identification and is put
Electric type.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811326013.2A CN109683066A (en) | 2018-11-08 | 2018-11-08 | Power cable typical defect local discharge signal recognition methods |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811326013.2A CN109683066A (en) | 2018-11-08 | 2018-11-08 | Power cable typical defect local discharge signal recognition methods |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109683066A true CN109683066A (en) | 2019-04-26 |
Family
ID=66185846
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811326013.2A Pending CN109683066A (en) | 2018-11-08 | 2018-11-08 | Power cable typical defect local discharge signal recognition methods |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109683066A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111308280A (en) * | 2019-12-11 | 2020-06-19 | 云南电网有限责任公司临沧供电局 | Non-contact ultrasonic detection method for partial discharge noise and discharge |
CN111999382A (en) * | 2020-09-17 | 2020-11-27 | 海南电网有限责任公司电力科学研究院 | Cable partial discharge characteristic parameter extraction method considering insulation aging |
CN112036451A (en) * | 2020-08-13 | 2020-12-04 | 广东电网有限责任公司 | Novel insulation defect identification method based on transfer learning, computer equipment and storage medium |
CN112036450A (en) * | 2020-08-12 | 2020-12-04 | 国家电网有限公司 | High-voltage cable partial discharge mode identification method and system based on transfer learning |
CN112147474A (en) * | 2020-10-30 | 2020-12-29 | 广东电网有限责任公司阳江供电局 | XLPE power cable typical defect partial discharge type identification system and method |
CN112329936A (en) * | 2020-11-05 | 2021-02-05 | 西北工业大学 | Signal continuation method based on multi-population genetic algorithm and similarity calculation |
CN112485610A (en) * | 2020-11-05 | 2021-03-12 | 国网电力科学研究院有限公司 | GIS partial discharge characteristic parameter extraction method considering insulation degradation |
CN112485622A (en) * | 2020-12-09 | 2021-03-12 | 西安华谱电力设备制造有限公司 | Partial discharge pattern recognition method based on GA-BPNN |
CN113780308A (en) * | 2021-08-27 | 2021-12-10 | 吉林省电力科学研究院有限公司 | GIS partial discharge mode identification method and system based on kernel principal component analysis and neural network |
CN116415186A (en) * | 2023-06-07 | 2023-07-11 | 广东电网有限责任公司珠海供电局 | Method and device for identifying local defects of high-voltage power cable |
CN116821591A (en) * | 2023-04-04 | 2023-09-29 | 浙江万能弹簧机械有限公司 | Real-time monitoring method and system for discharge condition of high-frequency power supply |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102156245A (en) * | 2011-03-11 | 2011-08-17 | 太原理工大学 | On-line fault diagnosis and prewarning method of mine high-voltage cable |
CN102809718A (en) * | 2012-07-25 | 2012-12-05 | 华南理工大学 | Ultra-high-frequency partial discharge signal identification method for gas insulated switchgear (GIS) |
CN102944769A (en) * | 2012-10-18 | 2013-02-27 | 上海市电力公司 | Fault diagnosis method of power transformer based on extreme learning machine |
CN103323755A (en) * | 2013-06-17 | 2013-09-25 | 广东电网公司电力科学研究院 | Method and system for recognition of GIS ultrahigh frequency partial discharge signal |
CN104299052A (en) * | 2014-10-12 | 2015-01-21 | 刘岩 | Building energy consumption prediction method based on artificial bee colony algorithm and neural network |
CN104777410A (en) * | 2015-04-22 | 2015-07-15 | 东北电力大学 | Partial discharge pattern identification method for crosslinked polyethylene cable |
CN105548832A (en) * | 2015-12-10 | 2016-05-04 | 国网四川省电力公司电力科学研究院 | High-voltage power cable fault recognition method |
CN106022465A (en) * | 2016-05-19 | 2016-10-12 | 江南大学 | Extreme learning machine method for improving artificial bee colony optimization |
CN106503689A (en) * | 2016-11-14 | 2017-03-15 | 哈尔滨理工大学 | Neutral net local discharge signal mode identification method based on particle cluster algorithm |
CN107728014A (en) * | 2017-08-23 | 2018-02-23 | 国网山东省电力公司检修公司 | Method and system based on multiple sensor signals characteristic identificating equipment insulation defect |
-
2018
- 2018-11-08 CN CN201811326013.2A patent/CN109683066A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102156245A (en) * | 2011-03-11 | 2011-08-17 | 太原理工大学 | On-line fault diagnosis and prewarning method of mine high-voltage cable |
CN102809718A (en) * | 2012-07-25 | 2012-12-05 | 华南理工大学 | Ultra-high-frequency partial discharge signal identification method for gas insulated switchgear (GIS) |
CN102944769A (en) * | 2012-10-18 | 2013-02-27 | 上海市电力公司 | Fault diagnosis method of power transformer based on extreme learning machine |
CN103323755A (en) * | 2013-06-17 | 2013-09-25 | 广东电网公司电力科学研究院 | Method and system for recognition of GIS ultrahigh frequency partial discharge signal |
CN104299052A (en) * | 2014-10-12 | 2015-01-21 | 刘岩 | Building energy consumption prediction method based on artificial bee colony algorithm and neural network |
CN104777410A (en) * | 2015-04-22 | 2015-07-15 | 东北电力大学 | Partial discharge pattern identification method for crosslinked polyethylene cable |
CN105548832A (en) * | 2015-12-10 | 2016-05-04 | 国网四川省电力公司电力科学研究院 | High-voltage power cable fault recognition method |
CN106022465A (en) * | 2016-05-19 | 2016-10-12 | 江南大学 | Extreme learning machine method for improving artificial bee colony optimization |
CN106503689A (en) * | 2016-11-14 | 2017-03-15 | 哈尔滨理工大学 | Neutral net local discharge signal mode identification method based on particle cluster algorithm |
CN107728014A (en) * | 2017-08-23 | 2018-02-23 | 国网山东省电力公司检修公司 | Method and system based on multiple sensor signals characteristic identificating equipment insulation defect |
Non-Patent Citations (10)
Title |
---|
吴德烽: "《计算智能及其在三维表面扫描机器人系统中的应用》", 30 November 2012, 大连海事大学出版社 * |
张卫等: "人工蜂群算法用于优化KELM参数的研究及应用", 《NCMMSC 2015》 * |
张秦梫等: "基于OS-ELM的变压器局部放电模式识别", 《高电压技术》 * |
毛羽等: "改进人工蜂群优化的极限学习机", 《传感器与微系统》 * |
牛培峰等: "极端学习机算法的改进及应用研究", 《燕山大学学报》 * |
王文锦等: "基于人工蜂群优化极限学习机的短期负荷预测", 《电测与仪表》 * |
苑津莎等: "基于相关系数矩阵和概率神经网络的局部放电模式识别", 《电力系统保护与控制》 * |
许翀寰等: "《复杂社会情境下的个性化推荐方法与应用》", 31 August 2018, 浙江工商大学出版社 * |
赵虎等: "改进人工蜂群算法优化ELM分类模型", 《计算机测量与控制》 * |
马云鹏等: "基于混沌分组教与学优化算法锅炉NOx模型优化研究", 《计量学报》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111308280A (en) * | 2019-12-11 | 2020-06-19 | 云南电网有限责任公司临沧供电局 | Non-contact ultrasonic detection method for partial discharge noise and discharge |
CN111308280B (en) * | 2019-12-11 | 2022-02-01 | 云南电网有限责任公司临沧供电局 | Non-contact ultrasonic detection method for partial discharge noise and discharge |
CN112036450A (en) * | 2020-08-12 | 2020-12-04 | 国家电网有限公司 | High-voltage cable partial discharge mode identification method and system based on transfer learning |
CN112036450B (en) * | 2020-08-12 | 2024-02-23 | 国家电网有限公司 | High-voltage cable partial discharge mode identification method and system based on transfer learning |
CN112036451A (en) * | 2020-08-13 | 2020-12-04 | 广东电网有限责任公司 | Novel insulation defect identification method based on transfer learning, computer equipment and storage medium |
CN111999382A (en) * | 2020-09-17 | 2020-11-27 | 海南电网有限责任公司电力科学研究院 | Cable partial discharge characteristic parameter extraction method considering insulation aging |
CN112147474A (en) * | 2020-10-30 | 2020-12-29 | 广东电网有限责任公司阳江供电局 | XLPE power cable typical defect partial discharge type identification system and method |
CN112485610B (en) * | 2020-11-05 | 2024-02-02 | 国网电力科学研究院有限公司 | GIS partial discharge characteristic parameter extraction method considering insulation degradation |
CN112329936A (en) * | 2020-11-05 | 2021-02-05 | 西北工业大学 | Signal continuation method based on multi-population genetic algorithm and similarity calculation |
CN112485610A (en) * | 2020-11-05 | 2021-03-12 | 国网电力科学研究院有限公司 | GIS partial discharge characteristic parameter extraction method considering insulation degradation |
CN112329936B (en) * | 2020-11-05 | 2023-03-24 | 西北工业大学 | Signal continuation method based on multi-population genetic algorithm and similarity calculation |
CN112485622A (en) * | 2020-12-09 | 2021-03-12 | 西安华谱电力设备制造有限公司 | Partial discharge pattern recognition method based on GA-BPNN |
CN113780308A (en) * | 2021-08-27 | 2021-12-10 | 吉林省电力科学研究院有限公司 | GIS partial discharge mode identification method and system based on kernel principal component analysis and neural network |
CN116821591A (en) * | 2023-04-04 | 2023-09-29 | 浙江万能弹簧机械有限公司 | Real-time monitoring method and system for discharge condition of high-frequency power supply |
CN116821591B (en) * | 2023-04-04 | 2024-03-08 | 浙江万能弹簧机械有限公司 | Real-time monitoring method and system for discharge condition of high-frequency power supply |
CN116415186A (en) * | 2023-06-07 | 2023-07-11 | 广东电网有限责任公司珠海供电局 | Method and device for identifying local defects of high-voltage power cable |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109683066A (en) | Power cable typical defect local discharge signal recognition methods | |
CN103914735B (en) | A kind of fault recognition method and system based on Neural Network Self-learning | |
CN105973594B (en) | A kind of rolling bearing fault Forecasting Methodology based on continuous depth confidence network | |
CN112098714B (en) | Electricity stealing detection method and system based on ResNet-LSTM | |
CN108038300B (en) | Optical fiber state evaluation method based on improved membership function combined with neural network | |
CN110263866A (en) | A kind of power consumer load setting prediction technique based on deep learning | |
CN109659933A (en) | A kind of prediction technique of power quality containing distributed power distribution network based on deep learning model | |
CN107843718A (en) | A kind of method that transformer insulation oil ageing state is assessed | |
CN102609612B (en) | Data fusion method for calibration of multi-parameter instruments | |
CN106777984A (en) | A kind of method that photovoltaic array Working state analysis and fault diagnosis are realized based on density clustering algorithm | |
CN107274067B (en) | Distribution transformer overload risk assessment method | |
CN105095963A (en) | Method for accurately diagnosing and predicting fault of wind tunnel equipment | |
Cui et al. | Statistical scenarios forecasting method for wind power ramp events using modified neural networks | |
Jiao | Application and prospect of artificial intelligence in smart grid | |
CN109973331A (en) | A kind of fan blade of wind generating set fault diagnosis algorithm based on bp neural network | |
Li et al. | Analysis and modelling of flood risk assessment using information diffusion and artificial neural network | |
CN110766313A (en) | Cable tunnel comprehensive state evaluation method based on operation and maintenance system | |
Ma et al. | An estimating combination method for interval forecasting of electrical load time series | |
CN110321555A (en) | A kind of power network signal classification method based on Recognition with Recurrent Neural Network model | |
CN105488598A (en) | Medium-and-long time electric power load prediction method based on fuzzy clustering | |
Zhang et al. | Prediction of daily water demands by neural networks | |
CN109242166A (en) | A kind of environmental forecasting prevention and control system based on multiple dimensioned deep neural network | |
CN106682651A (en) | Monitoring method for cable insulation leather damage based on neural network | |
Thilakarathne et al. | Applying artificial neural networks for predicting beach vulnerability to storm-induced erosion | |
Momani et al. | Short-term load forecasting based on NARX and radial basis neural networks approaches for the Jordanian power grid |
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 | ||
CB02 | Change of applicant information |
Address after: 250101 A203, block F, entrepreneurship square, Qilu Software Park, No.1 Shunhua Road, high tech Development Zone, Jinan City, Shandong Province Applicant after: Conway Communication Technology Co., Ltd Address before: 250101 A203, block F, entrepreneurship square, Qilu Software Park, No.1 Shunhua Road, high tech Development Zone, Jinan City, Shandong Province Applicant before: SHANDONG KANGWEI COMMUNICATION TECHNOLOGY Co.,Ltd. |
|
CB02 | Change of applicant information |