CN109164362A - A kind of recognition methods and system of direct current cables shelf depreciation defect failure - Google Patents
A kind of recognition methods and system of direct current cables shelf depreciation defect failure Download PDFInfo
- Publication number
- CN109164362A CN109164362A CN201811113943.XA CN201811113943A CN109164362A CN 109164362 A CN109164362 A CN 109164362A CN 201811113943 A CN201811113943 A CN 201811113943A CN 109164362 A CN109164362 A CN 109164362A
- Authority
- CN
- China
- Prior art keywords
- partial discharge
- pulse waveform
- waveform signal
- point
- discharge pulse
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000007547 defect Effects 0.000 title claims abstract description 83
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000012549 training Methods 0.000 claims abstract description 59
- 238000009413 insulation Methods 0.000 claims abstract description 45
- 238000012545 processing Methods 0.000 claims abstract description 11
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 239000000284 extract Substances 0.000 claims description 10
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 9
- 238000001514 detection method Methods 0.000 claims description 6
- 230000001629 suppression Effects 0.000 claims description 5
- 230000000052 comparative effect Effects 0.000 description 26
- 238000007599 discharging Methods 0.000 description 9
- 229920003020 cross-linked polyethylene Polymers 0.000 description 8
- 239000004703 cross-linked polyethylene Substances 0.000 description 8
- 230000000694 effects Effects 0.000 description 8
- 239000011159 matrix material Substances 0.000 description 7
- 230000005611 electricity Effects 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 238000001914 filtration Methods 0.000 description 5
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- NIPNSKYNPDTRPC-UHFFFAOYSA-N N-[2-oxo-2-(2,4,6,7-tetrahydrotriazolo[4,5-c]pyridin-5-yl)ethyl]-2-[[3-(trifluoromethoxy)phenyl]methylamino]pyrimidine-5-carboxamide Chemical compound O=C(CNC(=O)C=1C=NC(=NC=1)NCC1=CC(=CC=C1)OC(F)(F)F)N1CC2=C(CC1)NN=N2 NIPNSKYNPDTRPC-UHFFFAOYSA-N 0.000 description 3
- 230000009977 dual effect Effects 0.000 description 3
- 230000005764 inhibitory process Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 210000001367 artery Anatomy 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000001151 other effect Effects 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 210000003462 vein Anatomy 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- AFCARXCZXQIEQB-UHFFFAOYSA-N N-[3-oxo-3-(2,4,6,7-tetrahydrotriazolo[4,5-c]pyridin-5-yl)propyl]-2-[[3-(trifluoromethoxy)phenyl]methylamino]pyrimidine-5-carboxamide Chemical compound O=C(CCNC(=O)C=1C=NC(=NC=1)NCC1=CC(=CC=C1)OC(F)(F)F)N1CC2=C(CC1)NN=N2 AFCARXCZXQIEQB-UHFFFAOYSA-N 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000003822 epoxy resin Substances 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000009432 framing Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 229920000647 polyepoxide Polymers 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000002834 transmittance Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—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
- G01R31/1227—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 of components, parts or materials
- G01R31/1263—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 of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation
- G01R31/1272—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 of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation of cable, line or wire insulation, e.g. using partial discharge measurements
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Testing Relating To Insulation (AREA)
Abstract
本发明公开了一种直流电缆局部放电缺陷故障的识别方法，其包括步骤：(1)采集直流电缆的若干种绝缘缺陷放电模型的局部放电脉冲波形信号；(2)提取局部放电脉冲波形信号的有效信息，以得到训练样本；(3)构建基于受限玻尔兹曼机的深度信念网络，采用训练样本对所述深度信念网络进行无监督训练，以得到网络参数；(4)对所述深度信念网络进行有监督训练，以优化网络参数；(5)将待识别局部放电脉冲波形信号输入经过训练的所述深度信念网络中，以从其输出获得识别结果。此外，本发明还公开了一种直流电缆局部放电缺陷故障的识别系统，包括：信号采集模块、预处理模块以及信号处理模块。直流电缆局部放电缺陷故障的识别方法及系统准确率高。
The invention discloses a method for identifying a partial discharge defect fault of a DC cable, which comprises the steps of: (1) collecting partial discharge pulse waveform signals of several insulation defect discharge models of the DC cable; (2) extracting the partial discharge pulse waveform signal effective information to obtain training samples; (3) construct a deep belief network based on restricted Boltzmann machine, and use training samples to perform unsupervised training on the deep belief network to obtain network parameters; The deep belief network performs supervised training to optimize network parameters; (5) input the PD pulse waveform signal to be identified into the trained deep belief network to obtain the identification result from its output. In addition, the invention also discloses a fault identification system for partial discharge defects of a DC cable, which includes a signal acquisition module, a preprocessing module and a signal processing module. The identification method and system of DC cable partial discharge defect fault have high accuracy.
Description
Technical field
The present invention relates to a kind of recognition methods and system more particularly to a kind of recognition methods for direct current cables failure and
System.
Background technique
With the fast development of flexible high pressure direct current transportation, direct current crosslinked polyethylene (Cross-linked
Polyethylene, abbreviation XLPE) cable more and more applied by its excellent insulation performance.With more and more
D.C. high voltage transmission project put into operation successively, need improve on-line monitoring and fault early warning system with guarantee power supply it is reliable
Property.Since shelf depreciation (Partial discharge, abbreviation PD) is as having judged an important indicator of status of electric power
Examination criteria is included in by International Electrotechnical Commission, but the pattern-recognition of direct current cables shelf depreciation and fault diagnosis research are still located
In the starting stage.
System not yet forms direct current cables to exchange XLPE cable about the research spininess of cable local discharge at this stage
One detection method and evaluation criteria, therefore, there are also biggish research skies for the pattern-recognition of the local discharge signal of direct current cables
Between.
Summary of the invention
One of the objects of the present invention is to provide a kind of recognition methods of direct current cables shelf depreciation defect failure, the identifications
The pulsed current signal that method issues when shelf depreciation occurs based on direct current cables, then pre-processes pulsed current signal
After extract effective information, then by construct neural network be trained, by the neural network after training to signal to be identified
It is identified, the final fault diagnosis realized to direct current cables shelf depreciation defect.The recognition methods is compared to the prior art, right
The accuracy rate of failure modes is higher.
Based on above-mentioned purpose, the invention proposes a kind of recognition methods of direct current cables shelf depreciation defect failure, packets
Include step:
(1) partial discharge pulse's waveform signal of the several insulation defect discharging model of direct current cables is acquired；
(2) effective information of partial discharge pulse's waveform signal is extracted, to obtain training sample；
(3) deepness belief network based on limited Boltzmann machine is constructed, using training sample to the depth conviction net
Network carries out unsupervised training, to obtain network parameter；
(4) Training is carried out to the deepness belief network, to optimize network parameter；
(5) partial discharge pulse's waveform signal to be identified is inputted in the trained deepness belief network, with from
It, which is exported, obtains recognition result.
In the recognition methods of direct current cables shelf depreciation defect failure of the present invention, when part occurs for direct current cables
When electric discharge, partial discharge pulse's waveform signal is acquired, then collected partial discharge pulse's waveform signal is pre-processed,
Extract its effective information.Building is based on limited Boltzmann machine (Restricted Boltzmannmachine, abbreviation RBM)
Deepness belief network carries out unsupervised training, to obtain network parameter, then, carries out Training to deepness belief network,
Optimize network parameter, finally obtain trained deepness belief network, partial discharge pulse's waveform signal to be identified is inputted and is passed through
In the trained deepness belief network, to obtain recognition result from its output.
The recognition methods recognition accuracy of direct current cables shelf depreciation defect failure of the present invention is high.
Further, in recognition methods of the present invention, the insulation defect discharging model is put including at least air gap
Electric model, corona discharge model, scratch discharging model and creeping discharge model.
Further, in recognition methods of the present invention, in step (2), part is extracted using Canny algorithm and is put
The effective information of electric pulse waveform signal, the effective information include the changed segment of partial discharge pulse's waveform signal.
Further, in recognition methods of the present invention, in step (2), part is extracted using Canny algorithm and is put
The effective information of electric pulse waveform signal comprising steps of
Local Discharge pulse waveform signal f (x) is smoothed using one-dimensional Gaussian function, after obtaining gaussian filtering
Waveform signal
It seeksDerivative g (x), carry out non-maxima suppression, only retain derivative maximum point；
Carry out dual threshold detection: setting Low threshold δlWith high threshold δhIf derivative g (xi) it is less than δlThen mark xiFor non-edge
Point, if more than δhThen mark point xiFor strong edge point, remaining point is labeled as weak marginal point；
Weak marginal point is isolated in inhibition: setting neighborhood ε, the point is by as non-if weak edge vertex neighborhood ε is interior without strong edge point
Marginal point；Weak marginal point then chooses the 1st weak marginal point as partial discharge pulse's waveform signal changed if it exists
Otherwise the starting point of section chooses starting of the 1st strong edge point as the changed segment of partial discharge pulse's waveform signal
Point.
In some embodiments, the filter window 20 when gaussian filtering, Gaussian Profile standard deviation take 1.5.
Further, it in recognition methods of the present invention, in step (3), is successively instructed using contrast divergence algorithm
Practice the deepness belief network and obtains the network parameter, so that data that deepness belief network reconstructs and training sample
Data are consistent as much as possible.
Further, in recognition methods of the present invention, in step (4), using ADAM algorithm to the depth
Belief network carries out Training.
It should be noted that the Training of deepness belief network can use gradient descent method and Conjugate gradient descent
Method preferably uses adaptivity moments estimation (Adaptive Moment Estimation, abbreviation ADAM), more suitable to choose
Learning rate, be easy to converge to local optimum.
Correspondingly, another object of the present invention is to provide a kind of identification systems of direct current cables shelf depreciation defect failure
System, the shelf depreciation defect which can rapidly and accurately occur direct current cables identify.
Based on above-mentioned purpose, the invention also provides a kind of identifying system of direct current cables shelf depreciation defect failure,
Including
Signal acquisition module acquires partial discharge pulse's waveform of the several insulation defect discharging model of direct current cables
Signal；
Preprocessing module extracts the effective information of partial discharge pulse's waveform signal, to obtain training sample；
Signal processing module, use training sample to the deepness belief network based on limited Boltzmann machine of building into
The unsupervised training of row, to obtain network parameter；And Training is carried out to the deepness belief network, with the net optimized
Network parameter；
Wherein, partial discharge pulse's waveform signal to be identified is inputted in the trained deepness belief network, energy
It is enough to obtain recognition result from its output.
Further, in identifying system of the present invention, the preprocessing module extracts part using Canny algorithm
The effective information of Discharge pulse waveform signal, wherein the effective information includes that partial discharge pulse's waveform signal is changed
Segment, wherein using Canny algorithm extract partial discharge pulse's waveform signal effective information comprising steps of
Local Discharge pulse waveform signal f (x) is smoothed using one-dimensional Gaussian function, after obtaining gaussian filtering
Waveform signal
It seeksDerivative g (x), carry out non-maxima suppression, only retain derivative maximum point；
Carry out dual threshold detection: setting Low threshold δlWith high threshold δhIf derivative g (xi) it is less than δlThen mark xiFor non-edge
Point, if more than δhThen mark point xiFor strong edge point, remaining point is labeled as weak marginal point；
Weak marginal point is isolated in inhibition: setting neighborhood ε, the point is by as non-if weak edge vertex neighborhood ε is interior without strong edge point
Marginal point；Weak marginal point then chooses the 1st weak marginal point as partial discharge pulse's waveform signal changed if it exists
Otherwise the starting point of section chooses starting of the 1st strong edge point as the changed segment of partial discharge pulse's waveform signal
Point.
Further, in identifying system of the present invention, the signal processing module using contrast divergence algorithm without
Deepness belief network described in supervised training obtains the network parameter.
Further, in identifying system of the present invention, the signal processing module is using ADAM algorithm to described
Deepness belief network carries out Training and obtains the network parameter of the optimization.
The recognition methods of direct current cables shelf depreciation defect failure of the present invention and system have the following advantages that and have
Beneficial effect:
The recognition methods of direct current cables shelf depreciation defect failure of the present invention and system can be directed to direct current cables
Shelf depreciation occurs and carries out fault identification, partial discharge pulse's waveform signal is acquired, then to collected shelf depreciation arteries and veins
It rushes waveform signal to be pre-processed, extracts its effective information.The deepness belief network based on limited Boltzmann machine is constructed to carry out
Then unsupervised training, carries out Training to deepness belief network, optimizes network parameter, finally to obtain network parameter
Trained deepness belief network is obtained, partial discharge pulse's waveform signal to be identified is inputted into the trained depth conviction
In network, to obtain recognition result from its output.
The recognition methods of the direct current cables shelf depreciation defect failure and system identification accuracy rate are high.
Detailed description of the invention
Fig. 1 be the identifying system of DC current shelf depreciation defect failure of the present invention in one embodiment
Structural framing figure.
Fig. 2 be DC current shelf depreciation defect failure of the present invention recognition methods all in one embodiment
Flow diagram.
Fig. 3 shows partial discharge pulse's waveform signal when shelf depreciation defect failure occurs for DC current.
Fig. 4 is shown by partial discharge pulse's waveform letter under pretreated corona discharge insulation defect fault model
Number.
Fig. 5 is shown by partial discharge pulse's waveform letter under pretreated bubble-discharge insulation defect fault model
Number.
Fig. 6 shows the partial discharge pulse's waveform letter to discharge under insulation defect fault model by pretreated scratch
Number.
Fig. 7 shows the partial discharge pulse's wave climbed by pretreated along face under discharge of electricity insulation defect fault model
Shape signal.
Fig. 8 schematically illustrates the depth in the recognition methods of direct current cables shelf depreciation defect failure of the present invention
Belief network structure.
Fig. 9 shows the confusion matrix of the insulation defect fault identification result identified using comparative example 1.
Figure 10 shows the confusion matrix of the insulation defect fault identification result identified using comparative example 2.
Figure 11 shows the confusion matrix of the insulation defect fault identification result identified using embodiment 1.
Specific embodiment
Below will according to specific embodiment and Figure of description to direct current cables shelf depreciation defect of the present invention therefore
The recognition methods of barrier and system are described further, but the explanation does not constitute the improper restriction to technical solution of the present invention.
As shown in Figure 1, in the present embodiment, the identifying system of direct current cables shelf depreciation defect failure includes: signal
Acquisition module, preprocessing module and signal processing module.
Wherein, partial discharge pulse's wave of the several insulation defect discharging model of signal acquisition module acquisition direct current cables
Shape signal.
Preprocessing module extracts the effective information of partial discharge pulse's waveform signal, to obtain training sample.In this reality
It applies in mode, preprocessing module extracts the effective information of partial discharge pulse's waveform signal, effective information packet using Canny algorithm
The changed segment of partial discharge pulse's waveform signal is included, wherein extracting partial discharge pulse's waveform letter using Canny algorithm
Number effective information comprising steps of
Local Discharge pulse waveform signal f (x) is smoothed using one-dimensional Gaussian function, after obtaining gaussian filtering
Waveform signalSuch as when smoothing processing, it is assumed that filter window central point moves at x=μ, is using one-dimensional Gaussian function
Other points in central point and window distribute weight, are shown below:
In formula, ωxFor the weight of point x, σ is Gaussian Profile standard deviation.After being filtered to the point weighted average in window
'sPoint-by-point mobile filter window, the waveform signal after obtaining gaussian filteringSetting filter window 20 in this case, σ=
1.5。
It seeksDerivative g (x), carry out non-maxima suppression, only retain derivative maximum point.
Carry out dual threshold detection: setting Low threshold δlWith high threshold δhIf derivative g (xi) it is less than δlThen mark xiFor non-edge
Point, if more than δhThen mark point xiFor strong edge point, remaining point is labeled as weak marginal point.
Weak marginal point is isolated in inhibition: setting neighborhood ε, the point is by as non-if weak edge vertex neighborhood ε is interior without strong edge point
Marginal point；Weak marginal point then chooses the 1st weak marginal point as partial discharge pulse's waveform signal changed if it exists
Otherwise the starting point of section chooses starting of the 1st strong edge point as the changed segment of partial discharge pulse's waveform signal
Point.In view of when acquiring partial discharge pulse's waveform signal, collection capacity is larger, it is thus preferable to which shelf depreciation arteries and veins can be intercepted
The waveform length for rushing waveform signal is 600 points.
Signal processing module, use training sample to the deepness belief network based on limited Boltzmann machine of building into
The unsupervised training of row, to obtain network parameter；And Training is carried out to deepness belief network, with the network ginseng optimized
Number.In the present embodiment, signal processing module uses the unsupervised trained deepness belief network of contrast divergence algorithm, and uses
ADAM algorithm carries out Training to deepness belief network and obtains optimization network parameter.
Partial discharge pulse's waveform signal to be identified is inputted in trained deepness belief network, it can be from its output
Obtain recognition result
In order to verify this case direct current cables shelf depreciation defect failure identifying system recognition effect, using Fig. 2 institute
The recognition methods shown is identified.As shown in Fig. 2, the recognition methods of direct current cables shelf depreciation defect failure comprising steps of
(1) partial discharge pulse's waveform signal of the several insulation defect discharging model of direct current cables is acquired；
(2) effective information of partial discharge pulse's waveform signal is extracted, to obtain training sample；
(3) deepness belief network based on RBM is constructed, the deepness belief network is carried out using training sample unsupervised
Training, to obtain network parameter；
(4) Training is carried out to the deepness belief network, to optimize network parameter；
(5) partial discharge pulse's waveform signal to be identified is inputted in the trained deepness belief network, with from
It, which is exported, obtains recognition result.
In order to obtain partial discharge pulse's waveform signal of insulation defect discharging model not of the same race, simulation constructs four kinds not
The defect fault model of same direct current cables, building process are as described below:
Corona discharge insulation defect fault model: the corona discharge insulation defect fault model of production XLPE dielectric tip
When, the metal needle of a 3cm long, and contact wire core are penetrated in XLPE insulation position, to simulate corona discharge defect failure.
Bubble-discharge insulation defect fault model: it when production bubble-discharge insulation defect fault model, is pricked on the surface XLPE
Several micropores out, then sealed with epoxy resin, so that analog insulation air entrapment remains bubble-discharge defect failure occurred.
Scratch electric discharge insulation defect fault model: it when making scratch electric discharge insulation defect fault model, insulate in XLPE
On mark a road width 2mm, long 10mm, the scratch of deep 1mm, to simulate scratch discharge defect failure.
Discharge of electricity insulation defect fault model is climbed along face: when production climbs discharge of electricity insulation defect fault model along face,
There are wide 3mm when one end strips outer semiconducting layer, the residual of long 10mm climbs discharge of electricity defect failure along face to simulate.
It should be pointed out that above-mentioned insulation defect discharging model is only to schematically illustrate, the direct current cables of this case is locally put
The recognition methods of electric defect failure is not limited only to identify the insulation defect discharging model of the above-mentioned type, can also be to this
The insulation defect failure for other direct current cables that technical staff knows in field identifies.
The defect fault model of above-mentioned four kinds of different direct current cables is connect with identifying system, utilizes signal acquisition module
Partial discharge pulse's waveform signal is acquired, in the present embodiment, acquires partial discharge pulse's wave using High Frequency Current Sensor
Shape signal, the partial discharge model pulse waveform signal of acquisition is referring to Fig. 3.As shown in figure 3, when shelf depreciation occurs, voltage
Big ups and downs variation can occur, thus, the starting point for fluctuating generation is that the starting point of local discharge signal occurs.Due to one
In a little embodiments, collected partial discharge pulse's waveform signal, which has more interference information, to be influenced, it is made to be not easy to be obtained
Starting point, thus, local Discharge pulse waveform signal is pre-processed by using Canny algorithm, to extract it effectively
Information.It should be noted that ordinate indicates this actual magnitude U in Fig. 3, abscissa indicate practical nth strong point divided by
102Numerical value afterwards, such as the abscissa of certain point is x, then it is xth × 10 that this is practical2A data point.
Fig. 4 to Fig. 7 respectively illustrates different insulation defect fault models using the shelf depreciation obtained after Canny algorithm
Pulse waveform signal effective information.Wherein, Fig. 4 is shown by under pretreated corona discharge insulation defect fault model
Partial discharge pulse's waveform signal.Fig. 5 is shown by the part under pretreated bubble-discharge insulation defect fault model
Discharge pulse waveform signal.Fig. 6 shows the shelf depreciation discharged under insulation defect fault model by pretreated scratch
Pulse waveform signal.Fig. 7 shows the shelf depreciation climbed by pretreated along face under discharge of electricity insulation defect fault model
Pulse waveform signal.
It should be noted that ordinate of the Fig. 4 into Fig. 7 is indicated this actual magnitude U divided by waveform maximum amplitude Um
Institute's value, abscissa indicate practical nth strong point divided by 102Numerical value afterwards, such as the abscissa of certain point are x, then the reality
Border is xth × 102A data point.
The mechanism of deepness belief network based on RBM is referring to Fig. 8.As shown in figure 8, RBM is by visual layers and hidden layer group
At undirected probability graph model, there is no connections between the unit in visual layers or hidden layer.For including visual layers v and implying
The RBM, energy function E of layer hθ(v, h) can be expressed using following formula:
In formula, θ is RBM model parameter, θ={ w, a, b }；aiAnd bjRespectively aobvious member viWith hidden member hjBiasing；wijIt is aobvious
First viWith hidden member hjBetween connection weight；nvAnd nhRespectively aobvious member viWith hidden member hjNumber.
It for the deepness belief network based on RBM of this case, is stacked by several RBM, the hidden layer of lower layer RBM
As the input layer of upper layer RBM, i.e. RBMkIndicate the RBM, hidden layer h of kth time iterationk, input layer hk-1, and
RBMkInput layer hk-1It is also the hidden layer of the RBM of -1 iteration of kth.
Due to the RBM that the RBM in this case is given aobvious member and hidden first unit number, using training sample to the depth based on RBM
Degree belief network is trained, so that it is determined that RBM model parameter θ, training objective is the RBM mould under controlling RBM model parameter θ
The data that type reconstructs are consistent as far as possible with given training sample data.Method is carried out to RBM by contrast divergence algorithm
Successively quick unsupervised training.
Training when, random initializtion RBM model parameter θ first, using training sample be used as using training sample as reconstruct before
Aobvious member vo, and the hidden member h before reconstruct is calculated according to the following formula0:
In above formula, Pθ(hj 0=1 | v0) indicate given aobvious member v0When j-th of hidden yuan of hj 0It is set to 1 probability, sigmoid is sharp
Function living, sigmoid=(1+exp (- x))-1, x is mapped in (0,1) section.
Aobvious member v after then reconstructing according to the following formula1:
In above formula, Pθ(vi 1=1 | h0) indicate to give hidden member h0When i-th of aobvious member vj 1It is set to 1 probability.
Again by the aobvious member v after reconstructing1Recalculate the hidden member h after being reconstructed1.And so on obtain the depth based on RBM
The parameters of belief network, more new formula are shown below:
In formula, ε is to sdpecific dispersion calligraphy learning rate；< > is mathematic expectaion.ΔwijFor weight wijKnots modification, Δ aiFor
Bias aijKnots modification, Δ bjTo bias bijKnots modification, vi 0For i-th of aobvious member before reconstruct, hj 0For j-th before reconstruct
Hidden member, vi 1For i-th of aobvious member after reconstruct, hj 1For reconstruct after j-th hidden yuan.
In order to choose suitable learning rate, it is easy to converge to local optimum, using ADAM algorithm for deepness belief network
Training is carried out, Optimal Parameters are obtained.
When optimization, the parameter θ of network is obtained in kth time iterationkOn the basis of, gradient is obtained by following formula
Update has inclined single order moments estimation mk+1And have inclined second order moments estimation vk+1:
In formula, β1And β2Respectively hyper parameter.
And then first moment deviation is obtained by following formulaWith second moment deviation
The parameter θ of+1 iteration of kth after finally being updatedk+1:
In formula, α is step-length, and τ is stability constant.
In the present embodiment, step-length α is 0.001, and stability constant τ is 10-8, hyper parameter β1And β2Respectively 0.9 with
And 0.999.
In order to verify the recognition effect of this case, using this case training based on the deepness belief network of RBM as embodiment 1 and
Comparative example 1 using support vector machines and the comparative example 2 using reverse transmittance nerve network are drawn respectively to be obtained such as Fig. 9 to figure
Confusion matrix shown in 11, wherein Fig. 9 shows the mixed of the insulation defect fault identification result identified using comparative example 1
Confuse matrix.Figure 10 shows the confusion matrix of the insulation defect fault identification result identified using comparative example 2.Figure 11 is shown
Using the confusion matrix for the insulation defect fault identification result that embodiment 1 is identified.
In addition, it should be noted that, Fig. 9, into Figure 11, I indicates that corona discharge insulation defect fault model, II indicate edge
Discharge of electricity insulation defect fault model is climbed in face, and III indicates that bubble-discharge insulation defect fault model, IV indicate scratch electric discharge insulation
Defect fault model.
It can be seen that comparative example 1-2 and embodiment 1 in conjunction with Fig. 9 to Figure 11 and all have preferable knowledge for the defect of I and II
Other effect, recall rate are more than 97.50%.However for the defect of III and IV, the recall rate decline of comparative example 1 and comparative example 2 compared with
Greatly, comparative example 1 is respectively 92.03% and 92.73% in the recall rate of two kinds of defects, and comparative example 2 is called together two kinds of defects
The rate of returning is respectively 90.32% and 91.82%.And the effect of embodiment of this case 1 is better than comparative example 1 and comparative example 2, at this two kinds
The recall rate of defect is respectively 95.16% and 95.39%.Thus illustrate that the recognition effect of embodiment of this case 1 is best, both to I and
The defect of II recognition capability with higher, while good recognition effect is also possessed to the defect of III and IV.
Compared with traditional deepness belief network, network weight of the ADAM to the deepness belief network based on RBM of this case
Have supervision optimization efficiency it is higher, make the feature of the extraction of this case to partial discharge defect have stronger separating capacity, therefore identify
Effect has further promotion.And comparative example 1, comparative example 2 are poor to the recognition capability of III and IV, this is because III and IV
In defect forming process, air gap and insulation scuffing are that SI semi-insulation medium is filled with air, and there is centainly similar in mechanism
Property, therefore, the two misidentifies main Types each other, so that comparative example 1 and comparative example 2 need to mention by artificially carrying out feature
It takes, is otherwise not enough to fully express the partial discharge impulse waveform difference of defect in III and IV.And embodiment of this case 1 is logical
The pre-training unsupervised to RBM is crossed, partial discharge wave character is mapped to the top layer of deepness belief network.When these features include
Frequency and distribution characteristics, are difficult to the further feature being observed containing other simultaneously, have feature to initial data more accurate
Description, therefore, so that this case has more preferably recognition effect.
In addition, the partial discharge pulse's wave obtained after pretreatment will be acquired in order to further verify the recognition effect of this case
The effective information of shape signal is used as test sample for 4000 therein as 6400 samples, and remaining sample is with 400
It is a, 800,1200,1600,2000 and 2400 scale respectively as training set to the depth based on RBM of this case
It spends belief network and is trained using the comparative example 3 of traditional deepness belief network.Characteristic quantity sample is extracted according to same
Ratio is also trained comparative example 1, comparative example 2, and the recognition result of distinct methods is listed in table 1.
Table 1.
As shown in Table 1, embodiment 1 and average recognition accuracy of the comparative example 1-3 under set training set scale exist
90% or more, with the increase of training set, comparative example 1, comparative example 2, the recognition accuracy of comparative example 3 and embodiment 1 are higher.
In the case where small sample (such as training set sample size is at 400 and 800), embodiment of this case 1 and comparative example 1-3 know
Other effect is suitable, but with the increase of training set scale, the partial discharge impulse waveform feature that this case deepness belief network is extracted is more
Comprehensively, the average recognition accuracy of the embodiment 1 of this case is substantially better than comparative example 1-3.Simultaneously as embodiment of this case 1 uses
The convergence rate for the improvement deepness belief network that ADAM algorithm has supervision to finely tune, makes it faster, and calculates under same training scale
Method recognition accuracy is higher.
It should be noted that prior art part is not limited to given by present specification in protection scope of the present invention
Embodiment, all prior arts not contradicted with the solution of the present invention, including but not limited to first patent document, formerly
Public publication, formerly openly use etc., it can all be included in protection scope of the present invention.
In addition, it should also be noted that, institute in the combination of each technical characteristic and unlimited this case claim in this case
Combination documented by the combination or specific embodiment of record, all technical characteristics documented by this case can be to appoint
Where formula is freely combined or is combined, unless generating contradiction between each other.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811113943.XA CN109164362B (en) | 2018-09-25 | 2018-09-25 | Method and system for identifying partial discharge defect fault of direct current cable |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811113943.XA CN109164362B (en) | 2018-09-25 | 2018-09-25 | Method and system for identifying partial discharge defect fault of direct current cable |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109164362A true CN109164362A (en) | 2019-01-08 |
| CN109164362B CN109164362B (en) | 2021-01-08 |
Family
ID=64880196
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811113943.XA Active CN109164362B (en) | 2018-09-25 | 2018-09-25 | Method and system for identifying partial discharge defect fault of direct current cable |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109164362B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109799423A (en) * | 2019-01-09 | 2019-05-24 | 西安科技大学 | A kind of method of cable fault inline diagnosis |
| CN112305381A (en) * | 2020-09-21 | 2021-02-02 | 国网山东省电力公司临沂供电公司 | Method and system for online partial discharge monitoring and positioning of distribution cable |
| CN112307851A (en) * | 2019-08-02 | 2021-02-02 | 上海交通大学烟台信息技术研究院 | A method and system for identifying a bird's nest on a power tower |
| CN112363099A (en) * | 2020-10-30 | 2021-02-12 | 天津大学 | TMR current sensor temperature drift and geomagnetic field correction device and method |
| CN113376478A (en) * | 2021-06-22 | 2021-09-10 | 清华大学 | Power transmission line lightning stroke or short circuit fault positioning method based on edge detection |
| CN113848435A (en) * | 2021-08-31 | 2021-12-28 | 四川大学 | Direct current partial discharge signal classification and identification method based on frequency spectrum energy distribution probability |
| CN114034990A (en) * | 2021-10-22 | 2022-02-11 | 中核核电运行管理有限公司 | High-voltage electrical equipment online diagnosis system and method |
| CN114252737A (en) * | 2021-12-23 | 2022-03-29 | 武汉大学 | A state identification and evaluation method and system for the development process of power cable partial discharge |
| CN116736051A (en) * | 2023-05-22 | 2023-09-12 | 内蒙古工业大学 | GIS insulation defect intelligent diagnosis method, device, equipment and storage medium |
| CN118503620A (en) * | 2024-04-25 | 2024-08-16 | 山东亿元电力工程有限公司 | A partial discharge pattern recognition system based on power big data |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103954897A (en) * | 2014-05-20 | 2014-07-30 | 电子科技大学 | Intelligent power grid high-voltage insulation damage monitoring system and method based on ultraviolet imaging |
| CN104700421A (en) * | 2015-03-27 | 2015-06-10 | 中国科学院光电技术研究所 | Edge detection algorithm based on canny self-adaptive threshold value |
| CN107255779A (en) * | 2017-06-15 | 2017-10-17 | 国网河南省电力公司济源供电公司 | A kind of high-tension switch cabinet Analysis of Partial Discharge method |
| CN107449994A (en) * | 2017-07-04 | 2017-12-08 | 国网江苏省电力公司电力科学研究院 | Partial discharge method for diagnosing faults based on CNN DBN networks |
| CN107831409A (en) * | 2017-09-22 | 2018-03-23 | 国网山东省电力公司电力科学研究院 | The method and method for detecting abnormality of superfrequency partial discharge detection TuPu method parameter extraction |
-
2018
- 2018-09-25 CN CN201811113943.XA patent/CN109164362B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103954897A (en) * | 2014-05-20 | 2014-07-30 | 电子科技大学 | Intelligent power grid high-voltage insulation damage monitoring system and method based on ultraviolet imaging |
| CN104700421A (en) * | 2015-03-27 | 2015-06-10 | 中国科学院光电技术研究所 | Edge detection algorithm based on canny self-adaptive threshold value |
| CN107255779A (en) * | 2017-06-15 | 2017-10-17 | 国网河南省电力公司济源供电公司 | A kind of high-tension switch cabinet Analysis of Partial Discharge method |
| CN107449994A (en) * | 2017-07-04 | 2017-12-08 | 国网江苏省电力公司电力科学研究院 | Partial discharge method for diagnosing faults based on CNN DBN networks |
| CN107831409A (en) * | 2017-09-22 | 2018-03-23 | 国网山东省电力公司电力科学研究院 | The method and method for detecting abnormality of superfrequency partial discharge detection TuPu method parameter extraction |
Non-Patent Citations (3)
| Title |
|---|
| DIEDERIK P. KINGMA 等: "ADAM : A METHOD FOR STOCHASTIC OPTIMIZATION", 《INTERNATIONAL CONFERENCE ON LEARNING REPRESENTATIONS2015》 * |
| JOHN CANNY: "A Computational Approach to Edge Detection", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 龙红星: "基于振荡波局部放电检测的电缆缺陷类型识别研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109799423B (en) * | 2019-01-09 | 2021-05-07 | 西安科技大学 | Cable fault on-line diagnosis method |
| CN109799423A (en) * | 2019-01-09 | 2019-05-24 | 西安科技大学 | A kind of method of cable fault inline diagnosis |
| CN112307851A (en) * | 2019-08-02 | 2021-02-02 | 上海交通大学烟台信息技术研究院 | A method and system for identifying a bird's nest on a power tower |
| CN112305381A (en) * | 2020-09-21 | 2021-02-02 | 国网山东省电力公司临沂供电公司 | Method and system for online partial discharge monitoring and positioning of distribution cable |
| CN112363099B (en) * | 2020-10-30 | 2023-05-09 | 天津大学 | A TMR current sensor temperature drift and geomagnetic field correction device and method |
| CN112363099A (en) * | 2020-10-30 | 2021-02-12 | 天津大学 | TMR current sensor temperature drift and geomagnetic field correction device and method |
| CN113376478A (en) * | 2021-06-22 | 2021-09-10 | 清华大学 | Power transmission line lightning stroke or short circuit fault positioning method based on edge detection |
| CN113848435A (en) * | 2021-08-31 | 2021-12-28 | 四川大学 | Direct current partial discharge signal classification and identification method based on frequency spectrum energy distribution probability |
| CN113848435B (en) * | 2021-08-31 | 2023-12-15 | 四川大学 | Direct-current partial discharge signal classification and identification method based on frequency spectrum energy distribution probability |
| CN114034990A (en) * | 2021-10-22 | 2022-02-11 | 中核核电运行管理有限公司 | High-voltage electrical equipment online diagnosis system and method |
| CN114252737A (en) * | 2021-12-23 | 2022-03-29 | 武汉大学 | A state identification and evaluation method and system for the development process of power cable partial discharge |
| CN114252737B (en) * | 2021-12-23 | 2022-12-02 | 武汉大学 | State identification and evaluation method and system for power cable partial discharge development process |
| CN116736051A (en) * | 2023-05-22 | 2023-09-12 | 内蒙古工业大学 | GIS insulation defect intelligent diagnosis method, device, equipment and storage medium |
| CN118503620A (en) * | 2024-04-25 | 2024-08-16 | 山东亿元电力工程有限公司 | A partial discharge pattern recognition system based on power big data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109164362B (en) | 2021-01-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109164362A (en) | A kind of recognition methods and system of direct current cables shelf depreciation defect failure | |
| CN109685138B (en) | XLPE power cable partial discharge type identification method | |
| CN109635928B (en) | Identification method of voltage sag causes based on deep learning model fusion | |
| CN113486865B (en) | Power transmission line suspended foreign object target detection method based on deep learning | |
| CN110069886A (en) | The identification of cable initial failure and classification method based on VMD and CNN | |
| CN109212392A (en) | Recognition methods, system and the Caffe convolutional neural networks of direct current cables shelf depreciation defect failure | |
| CN109884419B (en) | An online fault diagnosis method for power quality of smart grid | |
| CN110161343A (en) | A kind of non-intrusion type real-time dynamic monitoring method of intelligence train exterior power receiving device | |
| CN103728535A (en) | Extra-high-voltage direct-current transmission line fault location method based on wavelet transformation transient state energy spectrum | |
| CN104809722A (en) | Electrical device fault diagnosis method based on infrared thermography | |
| CN107449994A (en) | Partial discharge method for diagnosing faults based on CNN DBN networks | |
| CN113514743B (en) | Method for constructing GIS partial discharge pattern recognition system based on multidimensional features | |
| CN109102508B (en) | Identification method of AC cable insulation defect based on partial discharge image features | |
| CN112787591B (en) | A fault diagnosis method for photovoltaic arrays based on fine-tuning densely connected convolutional neural networks | |
| CN109165604A (en) | The recognition methods of non-intrusion type load and its test macro based on coorinated training | |
| CN113589100A (en) | Abnormal discharge identification method based on bidirectional long-short term memory and attention mechanism | |
| CN109188211A (en) | A kind of high voltage equipment insulation method for diagnosing faults and system | |
| CN114545292A (en) | Cable fault type identification method based on deep learning | |
| CN108154223A (en) | Power distribution network operating mode recording sorting technique based on network topology and long timing information | |
| Xu et al. | DC cable feature extraction based on the PD image in the non-subsampled contourlet transform domain | |
| CN108509701A (en) | A kind of direct intelligent diagnosing method of rotating machinery fault based on vibration signal | |
| CN105046322A (en) | Method for diagnosing leading screw faults | |
| CN114169249A (en) | An artificial intelligence identification method for high-resistance grounding faults in distribution networks | |
| CN107832729A (en) | A kind of bearing rust intelligent diagnosing method | |
| Uckol et al. | Partial Discharge pattern classification based on deep learning for defect identification in MV cable terminations |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |