CN106468751A - A kind of transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption - Google Patents

A kind of transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption Download PDF

Info

Publication number
CN106468751A
CN106468751A CN201610865601.8A CN201610865601A CN106468751A CN 106468751 A CN106468751 A CN 106468751A CN 201610865601 A CN201610865601 A CN 201610865601A CN 106468751 A CN106468751 A CN 106468751A
Authority
CN
China
Prior art keywords
adaption
neutral net
network
fuzzy self
winding state
Prior art date
Application number
CN201610865601.8A
Other languages
Chinese (zh)
Inventor
马宏忠
黄春梅
施恂山
付明星
张艳
刘宝稳
Original Assignee
河海大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 河海大学 filed Critical 河海大学
Priority to CN201610865601.8A priority Critical patent/CN106468751A/en
Publication of CN106468751A publication Critical patent/CN106468751A/en

Links

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01RMEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/50Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
    • G01R31/72Testing of electric windings
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computer systems based on biological models
    • G06N3/02Computer systems based on biological models using neural network models
    • G06N3/04Architectures, e.g. interconnection topology
    • G06N3/0409Adaptive resonance theory [ART] networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computer systems based on biological models
    • G06N3/02Computer systems based on biological models using neural network models
    • G06N3/04Architectures, e.g. interconnection topology
    • G06N3/0436Architectures, e.g. interconnection topology in combination with fuzzy logic

Abstract

The invention discloses a kind of transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption, including following step:(1) arrange the monitoring point of vibrating sensor, the sample frequency of setting data collecting instrument and sampling time, gather transformer vibration signal;(2) WAVELET PACKET DECOMPOSITION and reconstruct are carried out to transformer vibration signal;(3) extract vibration signal sub-band energy value;(4) the vibration signal sub-band energy value of each vibration measuring point is analyzed, selects the feature band energy structural features vector of effective measuring point, the input of the neutral net that resonates as fuzzy self-adaption;(5) build fuzzy self-adaption resonance neutral net, adjust network parameter, till reaching the precision of setting;(6) by fuzzy self-adaption resonance neutral net, transformer winding state is identified.The present invention can quickly and accurately judge Transformer Winding impaction state, can be used for on-line monitoring and the identification of Transformer Winding impaction state.

Description

A kind of transformer winding state identification of the neutral net that resonated based on fuzzy self-adaption Method

Technical field

The present invention relates to a kind of transformer winding state recognition methodss, more particularly, to a kind of refreshing based on fuzzy self-adaption resonance Transformer winding state recognition methodss through network, belong to the condition monitoring and fault diagnosis technical field of power transformer.

Background technology

Transformer fault produces a very large impact to the safety and stability economical operation of whole electrical network, and short circuit accident is transformator It is easiest to one of accident of generation, for general transformator, anti-short circuit capability is not by force its common fault in day-to-day operation.Transformation Device anti-short circuit capability is closely related with winding state, anti-short circuit capability deficiency will result directly in winding loosen, thrust decline.So And, also do not have can effectively monitor the effective ways of winding state both at home and abroad at present it is impossible to grasp inside transformer winding in time Running status.

Vibratory drilling method gathers vibration signal using the vibrating sensor being pasted onto transformer body surface, by entering to vibration signal Row analyzing and processing extracts vibration performance amount to carry out on-line monitoring.With short-circuit impedance, action of low-voltage pulse and Frequency Response Analysis etc. electrically Laboratory method passes through to detect change difference, the method and the whole power system electrical isolation of electric parameter, monitors conveniently, safely, Safe operation on transformator and whole power system does not affect, and can be greatly enhanced the accuracy of diagnosis.

Transformer device structure is complicated, and equipment itself exists non-linear.Again due to form and the reason of transformer fault generation Numerous, along with equipment operating condition change so that contain in vibration signal complexity dynamic, non-stationary with non-linear become Point.Therefore cannot directly by simple Fourier transformation, vibration signal be analyzed processing it is necessary to through signal processing Extract characteristic quantity to judge running state of transformer.

Wavelet package transforms are a kind of nonstationary random response methods growing up on the basis of wavelet transformation, can not only Signal low frequency part is decomposed, fine resolution capability can also be provided to HFS.WAVELET PACKET DECOMPOSITION can will be believed Number irredundant exhaustively Orthogonal Decomposition, in independent sub-band, can be carried out to the vibration signal comprising a large amount of high-frequency informations Preferably Time-Frequency Localization analysis.

Neutral net is widely used in pattern recognition, and fuzzy ART net is by fuzzy theory and adaptive resonance network phase In conjunction with being a kind of self-organizing simulative neural network.With general neutral net with the monotone decreasing of network error or energy function Different for algorithm criterion, it adopts competition learning mechanism, using between biological neural cell excited with suppression principle, allow input mould Formula is identified and is compared by being bi-directionally connected of network, finally reaches resonance to complete the memory to itself and network returned Recall, solve the problems, such as that Stability of Neural Networks is faced a difficult choice with plasticity.

Content of the invention

In view of the shortcomings of the prior art, it is an object of the present invention to provide a kind of be based on fuzzy self-adaption resonance neutral net Transformer winding state recognition methodss, transformer vibration signal is carried out with wavelet-packet energy extraction, structural features vector conduct Fuzzy self-adaption resonates the input of neutral net, by continuous study under resonance state for the network and adjustment, comes to transformator Winding impaction state is identified;The present invention can quickly and accurately judge Transformer Winding impaction state, can be used for transformation The on-line monitoring of device winding impaction state and identification.

To achieve these goals, the present invention is to realize by the following technical solutions:

The present invention a kind of based on fuzzy self-adaption resonate neutral net transformer winding state recognition methodss, including with Under several steps:

(1) monitoring point of multiple vibrating sensors, the sample frequency of setting data collecting instrument and sampling time, collection are arranged Transformer vibration signal X (t);

(2) WAVELET PACKET DECOMPOSITION and reconstruct are carried out to described transformer vibration signal X (t);

(3) extract the sub-band energy value of wavelet package reconstruction signal;

(4) the sub-band energy value of each vibration measuring point is analyzed, selects the energy value of the feature band of effective measuring point As characteristic vector T, the input of the neutral net that resonates as fuzzy self-adaption;

(5) set up fuzzy self-adaption resonance neutral net, and carry out the optimal value of network parameter and select;

(6) by described fuzzy self-adaption resonance neutral net, transformer winding state is identified.

In step (1), described vibrating sensing implement body uses piezoelectric type vibration acceleration transducer.

In step (1), a part of described vibrating sensor is fixed in the middle of the high-low pressure winding of transformer oil box top, in addition A part of vibrating sensor is fixed at the height of oil tank of transformer side 1/2.

In step (1), the vibration signal of continuous at least 3 times collection each measuring point of transformator of described data collecting instrument, by setting Fixed sample frequency and complete cycle in sampling time intercept transformer vibration signal X (t).

In step (4), the method that each sub-band energy value vibrating measuring point is analyzed is as follows:First, make all Measuring point sub-band energy scattergram, in the sub-band energy scattergram of all measuring points, first selects vibration signal strong and in difference Distinguishing measuring point is vibrated as effective measuring point under winding state;Then, in the sub-band energy scattergram of effective measuring point, choosing Select the sub-band that energy value changes under different winding states as feature band.

In step (5), set up fuzzy self-adaption resonance neutral net using MATLAB platform.

In step (5), described network parameter includes selection parameter, learning rate and alarm threshold;The optimum of network parameter Value system of selection is as follows:First, described learning rate and alarm threshold are set to default value, described selection parameter is set to difference Value carries out winding state identification to network, selects the best selection parameter of recognition effect as network optimum selection parameter;Then, Described selection parameter and alarm threshold are set to default value, described learning rate are set to different value winding state is carried out to network Identification, selects the best learning rate of recognition effect as the optimal learning rate of network;Finally, by described selection parameter and study Speed is set to default value, described alarm threshold is set to different value network is carried out with winding state identification, select recognition effect Good alarm threshold is as the optimal alarm threshold of network.

In step (5), described fuzzy self-adaption resonance neutral net is constituted by 3 layers, if being originally inputted vectorial I=(I1, I2…,Im), wherein Ii∈ [0,1] (i=1,2 ..., m);

1. pretreatment layer F0, it is made up of 2m node, carry out complement code process to being originally inputted vectorial I, be originally inputted vectorial I F can directly be passed through0Layer is input in network, F0Layer is output as X, X=Wherein

2. compare a layer F1, it is made up of 2m node, M is F1The nodes of layer, then M=2m, F1Layer accepts to be derived from F0Layer is the bottom of from Input X and F upwards2The top-down input W of layer, wherein W are network weight, whereinI.e.:Work as i=1, when 2 ..., m, xi=Ii;Work as i =m+1, m+2 ..., during 2m,

3. identification layer F2, by N number of node yj(j=1,2 ..., N) is constituted, and N is input vector generic sum, each section Point yj(j=1,2 ..., N) represents input vector generic numbering, Y=(y1,y2,…,yN), Y represents that network is stored Input vector classification.

In step (6), resonated neutral net side that transformer winding state is identified by described fuzzy self-adaption Method is as follows:

(1-1) pretreatment:Input external vector I, in pretreatment layer F0Form complement code input vector X,

(1-2) netinit:Carry out initialization and selection parameter α, learning rate β and the Police sports of network weight W The optimal value of value ρ selects;

(1-3) model selection:To input vector X and identification layer F2Node j (j=1,2 ..., N) calculate and select function Tj,By TJ=max (T1,T2,…,TN), selection node J is winning neuron;Wherein, | | X ∧ wj| |=min (X,wj),wjFor F2Node layer j points to the network weight of F1 layer, is the component of network weight W;wijFor F2Layer section Point j points to F1The network weight of node layer i, is network weight wjComponent;

(1-4) pattern match:Calculate matching degree M of input vector X and node JJ,By MJWith warning Threshold value ρ is compared, wherein,

If (1-5) MJ>=ρ, then resonate, and this input pattern is classified as node J affiliated pattern class, and carries out weights tune Whole:Node J weighed value adjusting is:wij=β (I ∧ wij)+(1-β)wij, other node weights are constant;Otherwise return to step (1-3) weight Next winning neuron is newly selected to carry out pattern match;

If (1-6) identification layer F2All nodes all mismatch, then a newly-built network node stores this pattern.

The present invention can be with the wavelet-packet energy value construction vibration performance vector of transformer vibration signal, as fuzzy adaptive The input of the neutral net that should resonate, carries out automatic identification to transformer winding state exactly.Not only can be in neutral net The winding state of storage is identified, and can also accurately identify unknown winding new state, be particularly well-suited to Transformer Winding On-line monitoring and state recognition.

Brief description

Fig. 1 is the transformer winding state recognition methodss workflow of the neutral net that resonated based on fuzzy self-adaption of the present invention Cheng Tu;

Fig. 2 is vibration monitoring point layout drawing;

Fig. 3 (a) is the relation of No. 1 measuring point the 4th floor wavelet packet sub-band energy and winding state;

Fig. 3 (b) is the relation of No. 1 measuring point the 4th floor wavelet packet sub-band energy and winding state;

Fig. 3 (c) is the relation of No. 1 measuring point the 4th floor wavelet packet sub-band energy and winding state;

Fig. 3 (d) is the relation of No. 1 measuring point the 4th floor wavelet packet sub-band energy and winding state;

Fig. 3 (e) is the relation of No. 1 measuring point the 4th floor wavelet packet sub-band energy and winding state;

Fig. 3 (f) is the relation of No. 1 measuring point the 4th floor wavelet packet sub-band energy and winding state;

Fig. 4 is the structure chart of fuzzy self-adaption of the present invention resonance neutral net;

Fig. 5 is fuzzy self-adaption of the present invention resonance neural network recognization process workflow journey figure.

Specific embodiment

Technological means, creation characteristic, reached purpose and effect for making the present invention realize are easy to understand, with reference to Specific embodiment, is expanded on further the present invention.

Referring to Fig. 1, a kind of transformer winding state identification side of neutral net that resonated based on fuzzy self-adaption of the present invention Method, including following step:

(1) monitoring point of multiple vibrating sensors, the sample frequency of setting data collecting instrument and sampling time, collection are arranged Transformer vibration signal X (t);

(2) WAVELET PACKET DECOMPOSITION and reconstruct are carried out to described transformer vibration signal X (t);

(3) extract the sub-band energy value of wavelet package reconstruction signal;

(4) all measuring point sub-band energy scattergrams are made.Because vibration signal is in communication process, different vibrations can phase Mutually affect, have different degrees of decay through different routes of transmission.Therefore, in the surveyed vibration signal of tank surface diverse location Power is different.In the sub-band energy scattergram of all measuring points, first select vibration signal strong and in different winding states Lower vibration has the measuring point being clearly distinguished from as effective measuring point.Then, in the sub-band energy scattergram of effective measuring point, select Under different winding states, energy value has the sub-band of significant change as feature band.Finally, select the feature frequency of effective measuring point The energy value of band is as characteristic vector T, the input of the neutral net that resonates as fuzzy self-adaption;

(5) set up fuzzy self-adaption resonance neutral net using MATLAB or other platforms, and carry out network parameter The figure of merit selects.Different choice parameter, learning rate and alarm threshold can produce a very large impact to Network Recognition effect, therefore, need Optimal value selection is carried out to these three network parameters.First learning rate and alarm threshold are set to default value, ginseng will be selected Number is set to different value and network is carried out with winding state identification, selects the best selection parameter of recognition effect as network optimum selection Parameter.Then selection parameter and alarm threshold are set to default value, learning rate are set to different value winding shape is carried out to network State identifies, selects the best learning rate of recognition effect as the optimal learning rate of network.Finally by selection parameter and study speed Rate is set to default value, alarm threshold is set to different value network is carried out with winding state identification, selects the best police of recognition effect Guard against threshold value as the optimal alarm threshold of network;

(6) by described fuzzy self-adaption resonance neutral net, transformer winding state is identified.

The transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption, by changing Transformer Winding Pretightning force and structural member between is carried out analogue transformer winding and is loosened state, and 9 kinds of windings of Setup Experiments loosen state, as table 1 institute Show.Before and after winding, the specified thrust of screw rod normal condition is 28MPa, and 0MPa represents the state that loosens completely, and 0.5 is (corresponding 14MPa) represent and not exclusively loosen state.Transformator is carried out pumping, hangs cover, compressed come precise control winding by hydraulic system Power, to realize the different setting loosening state of winding.Then refill cover, oiling, reinstall transformator, standing.

Table 1 winding impaction state

Using the vibration acceleration sensor of model JF2020, using magnetic support, sensor is absorbed and fixed at transformer oil Case surface.Sensor is arranged at the measuring point 1~6 shown in Fig. 2, and 1,2,3 are located in the middle of top of oil-tank high-low pressure winding, and 4,5,6 At the height of fuel tank side 1/2, corresponding with A, B, C respectively.Data collecting instrument selects Nicolet Acquisition Instrument, sample frequency It is set to 10kHz.

Winding state is set by table 1 winding state order, often after setting first winding state, by step down side three-phase Short circuit, the outfan of high pressure side joint pressure regulator, adjusting pressure regulator and making low-pressure side electric current is rated current, collection transformator vibration letter Number.Experiment amounts to 9 vibration signals of collection.

Using db10 small echo, 4 layers of WAVELET PACKET DECOMPOSITION and reconstruct (each sub-band frequency scope are carried out to the vibration signal of collection As shown in table 2), extract the 4th straton frequency band energy value.6 measuring points the 4th layer of wavelet-packet energy value under 9 kinds of winding states is such as Shown in Fig. 3 (a), 3 (b), 3 (c), 3 (d), 3 (e), 3 (f).Can be seen by Fig. 3 (a), 3 (b), 3 (c), 3 (d), 3 (e), 3 (f) Go out, the wavelet-packet energy under No. 1 and 9 kinds of winding states of No. 4 measuring points does not have too big difference substantially, its sub-band energy is not composed and make It is characterized the subvector of vector.4th floor Wavelet Packet Energy Spectrum of No. 2, No. 3, No. 5 and No. 6 measuring points first (S (4,0), 0~ 312.5Hz) He seven (S (4,6), 1250~1562.5Hz) sub-band has larger difference under 9 kinds of winding states.Therefore 2,3, 5th, 6 is effective measuring point, and meanwhile, S (4,0) and S (4,6) are vibration performance frequency bands.

Using feature band S (4,0) and S (4,6) totally 8 sub-band energy of 2,3,5,6 this 4 effective measuring points, as Eigenvalue under Transformer Winding different conditions, to judge winding state.I.e. characteristic vector T=[E240,E246,E340,E346, E540,E546,E640,E646] (E represents that sub-band energy is composed, and subscript first digit represents measuring point, and second and third numeral represents the 4 layers of wavelet packet sub-band).

24 layers of WAVELET PACKET DECOMPOSITION sub-band frequency scope (Hz) of table

Build fuzzy self-adaption resonance neutral net, its network structure is as shown in figure 4, constituted by 3 layers.If input vector I =(I1,I2…,Im), wherein Ii∈ [0,1] (i=1,2 ..., m).

1. pretreatment layer F0:It is made up of 2m node, carry out complement code process to being originally inputted vector, be originally inputted vectorial I F can directly be passed through0Layer is input in network,Wherein

2. compare a layer F1:It is made up of the individual node of 2m (making M=2m), accept to be derived from F0Bottom-up input X and F of layer2Layer is certainly Push up downward input W, wherein X=(x1,x2,…,xM)=A.

3. identification layer F2:It is made up of N number of node, Y=(y1,y2,…,yN), each node represents input pattern generic Numbering.

Take each 20 groups of characteristic quantity under front 8 kinds of different conditions for the Transformer Winding, totally 160 groups of data are as sample data, Network is carried out with the identification of the network parameters such as selection parameter α, learning rate β and alarm threshold ρ, makes up to satisfied precision.

Network Recognition process is as shown in figure 5, concrete grammar is as follows:

(1-1) pretreatment:Input external vector I, in F0Layer forms complement code input vector X,

(1-2) netinit:Carry out weights W, the initialization of selection parameter α, learning rate β and alarm threshold ρ;

(1-3) model selection:To input pattern X and F2Node j (j=1,2 ..., N) calculate and select function Tj,By TJ=max (T1,T2,…,TN), selection node J is winning neuron.Wherein, (x ∧ wj)i=min (xi, wij),

(1-4) pattern match:Calculate matching degree M of input vector X and node JJ,By MJWith warning Threshold value ρ is compared;

If (1-5) MJ>=ρ, then resonate, and this input pattern is classified as node J affiliated pattern class, and carries out weights tune Whole:Node J weighed value adjusting is:wij=β (I ∧ wij)+(1-β)wij, other node weights are constant.Otherwise return to (1-3) again to select Select next winning neuron and carry out pattern match.

If (1-6) F2The all node of layer all mismatches, then a newly-built network node stores this pattern.

When selecting parameter alpha=0.05, learning rate β=1 (Fast Learning), during alarm threshold ρ=0.96, network performance reaches To most preferably.Network parameter is set to optimum, network is tested.Separately take characteristic quantity under front 8 kinds of states for the winding each 20 groups, then take 20 groups of characteristic quantity under state 9 for the winding, totally 180 groups of test datas are tested to network, test result such as table Shown in 3.

Table 3 fuzzy ART net recognition result

From table 3 it can be seen that

(1) for the winding state of storage in network, fuzzy self-adaption resonance neural network recognization rate is maintained at 95% More than, can rapidly and accurately carry out winding state identification.

(2) for the new model state 9 not stored in network although being trained to it not over training sample, but Network still can identify that this state is the new classification of network, without judging by accident as existing classification.This It is that other neutral nets do not have the advantage that.

(3) state 9 is classified as -1 and is because that the new model not stored class is defined as -1 by network by network, using as new The pattern class of classification and network storage makes a distinction.So when next input pattern enters network and carries out knowing classification, network Weighed value adjusting can be re-started, the pattern as storage carries out network memory.

Ultimate principle and principal character and the advantages of the present invention of the present invention have been shown and described above.The technology of the industry , it should be appreciated that the present invention is not restricted to the described embodiments, the simply explanation described in above-described embodiment and description is originally for personnel The principle of invention, without departing from the spirit and scope of the present invention, the present invention also has various changes and modifications, these changes Change and improvement both falls within scope of the claimed invention.Claimed scope by appending claims and its Equivalent thereof.

Claims (9)

1. a kind of transformer winding state recognition methodss based on fuzzy self-adaption resonance neutral net are it is characterised in that include Following step:
(1) arrange the monitoring point of multiple vibrating sensors, the sample frequency of setting data collecting instrument and sampling time, gather transformation Device vibration signal X (t);
(2) WAVELET PACKET DECOMPOSITION and reconstruct are carried out to described transformer vibration signal X (t);
(3) extract the sub-band energy value of wavelet package reconstruction signal;
(4) the sub-band energy value of each vibration measuring point is analyzed, selects the energy value conduct of the feature band of effective measuring point Characteristic vector T, the input of the neutral net that resonates as fuzzy self-adaption;
(5) set up fuzzy self-adaption resonance neutral net, and carry out the optimal value of network parameter and select;
(6) by described fuzzy self-adaption resonance neutral net, transformer winding state is identified.
2. the transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption according to claim 1, It is characterized in that, in step (1), described vibrating sensing implement body uses piezoelectric type vibration acceleration transducer.
3. the transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption according to claim 1, It is characterized in that, in step (1), a part of described vibrating sensor is fixed in the middle of the high-low pressure winding of transformer oil box top, Another part vibrating sensor is fixed at the height of oil tank of transformer side 1/2.
4. the transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption according to claim 1, It is characterized in that, in step (1), the vibration signal of continuous at least 3 times collection each measuring point of transformator of described data collecting instrument, press Set sample frequency and intercept transformer vibration signal X (t) complete cycle in sampling time.
5. the transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption according to claim 1, It is characterized in that, in step (4), the method that each sub-band energy value vibrating measuring point is analyzed is as follows:First, make institute There is measuring point sub-band energy scattergram, in the sub-band energy scattergram of all measuring points, first select vibration signal strong and not With vibrating distinguishing measuring point under winding state as effective measuring point;Then, in the sub-band energy scattergram of effective measuring point, Select the sub-band that energy value changes under different winding states as feature band.
6. the transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption according to claim 1, It is characterized in that, in step (5), set up fuzzy self-adaption resonance neutral net using MATLAB platform.
7. the transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption according to claim 1, It is characterized in that, in step (5), described network parameter includes selection parameter, learning rate and alarm threshold;Network parameter is Figure of merit system of selection is as follows:First, described learning rate and alarm threshold are set to default value, described selection parameter is set to not With value, network is carried out with winding state identification, select the best selection parameter of recognition effect as network optimum selection parameter;So Afterwards, described selection parameter and alarm threshold are set to default value, described learning rate are set to different value winding is carried out to network State recognition, selects the best learning rate of recognition effect as the optimal learning rate of network;Finally, by described selection parameter and Learning rate is set to default value, described alarm threshold is set to different value network is carried out with winding state identification, selects identification effect Really best alarm threshold is as the optimal alarm threshold of network.
8. the transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption according to claim 1, It is characterized in that, in step (5), described fuzzy self-adaption resonance neutral net is constituted by 3 layers, if being originally inputted vectorial I= (I1,I2…,Im), wherein Ii∈ [0,1] (i=1,2 ..., m);
1. pretreatment layer F0, it is made up of 2m node, carry out complement code process to being originally inputted vectorial I, be originally inputted vectorial I permissible Directly pass through F0Layer is input in network, F0Layer is output as X, Wherein
2. compare a layer F1, it is made up of 2m node, M is F1The nodes of layer, then M=2m, F1Layer accepts to be derived from F0Layer is bottom-up Input X and F2The top-down input W of layer, wherein W are network weight, whereinI.e.:Work as i=1, when 2 ..., m, xi=Ii;Work as i =m+1, m+2 ..., during 2m,
3. identification layer F2, by N number of node yj(j=1,2 ..., N) is constituted, and N is input vector generic sum, each node yj (j=1,2 ..., N) represents input vector generic numbering, Y=(y1,y2,…,yN), Y represents the stored input of network Vectorial classification.
9. the transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption according to claim 8, It is characterized in that, in step (6), resonate what neutral net was identified to transformer winding state by described fuzzy self-adaption Method is as follows:
(1-1) pretreatment:Input external vector I, in pretreatment layer F0Form complement code input vector X,
(1-2) netinit:Carry out the initialization of network weight W and selection parameter α, learning rate β and alarm threshold ρ Optimal value selects;
(1-3) model selection:To input vector X and identification layer F2Node j (j=1,2 ..., N) calculate and select function Tj,By TJ=max (T1,T2,…,TN), selection node J is winning neuron;Wherein, | | X ∧ wj| |=min (X,wj),wjFor F2Node layer j points to F1The network weight of layer, is the component of network weight W;wijFor F2Layer section Point j points to F1The network weight of node layer i, is network weight wjComponent;
(1-4) pattern match:Calculate matching degree M of input vector X and node JJ,By MJWith alarm threshold ρ It is compared, wherein,
If (1-5) MJ>=ρ, then resonate, and this input pattern is classified as node J affiliated pattern class, and carries out weighed value adjusting:Section Point J weighed value adjusting is:wij=β (I ∧ wij)+(1-β)wij, other node weights are constant;Otherwise return to step (1-3) again to select Select next winning neuron and carry out pattern match;
If (1-6) identification layer F2All nodes all mismatch, then a newly-built network node stores this pattern.
CN201610865601.8A 2016-09-29 2016-09-29 A kind of transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption CN106468751A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610865601.8A CN106468751A (en) 2016-09-29 2016-09-29 A kind of transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610865601.8A CN106468751A (en) 2016-09-29 2016-09-29 A kind of transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption

Publications (1)

Publication Number Publication Date
CN106468751A true CN106468751A (en) 2017-03-01

Family

ID=58230782

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610865601.8A CN106468751A (en) 2016-09-29 2016-09-29 A kind of transformer winding state recognition methodss of the neutral net that resonated based on fuzzy self-adaption

Country Status (1)

Country Link
CN (1) CN106468751A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106990303A (en) * 2017-03-15 2017-07-28 国家电网公司 A kind of Diagnosis Method of Transformer Faults
CN107290041A (en) * 2017-07-28 2017-10-24 河海大学 It is a kind of that state monitoring method is loosened based on phase space reconfiguration and the KPCM Transformer Winding clustered
CN107831415A (en) * 2017-10-20 2018-03-23 广东电网有限责任公司河源供电局 The Interval Valued Fuzzy diversity method that a kind of transformer insulating paper ageing state is assessed
CN108414610A (en) * 2018-05-09 2018-08-17 南开大学 It is a kind of to compose construction method based on the comprehensive pollution derived components of individual particle aerosol mass spectrometer and ART-2a neural networks
CN108921124A (en) * 2018-07-17 2018-11-30 河海大学 A kind of on-load tap changers of transformers mechanical breakdown on-line monitoring method

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2003072431A (en) * 2001-08-30 2003-03-12 Central Japan Railway Co Feeder circuit failure spotting device
CN102721465A (en) * 2012-06-13 2012-10-10 江苏省电力公司南京供电公司 System and method for diagnosing and preliminarily positioning loosening faults of iron core of power transformer
CN103135035A (en) * 2011-11-25 2013-06-05 江西省电力科学研究院 Transformer winding state diagnosis method
CN103163420A (en) * 2011-12-08 2013-06-19 沈阳工业大学 Intelligent power transformer on-line state judgment method
CN104102838A (en) * 2014-07-14 2014-10-15 河海大学 Transformer noise prediction method based on wavelet neural network and wavelet technology
CN105181120A (en) * 2015-09-02 2015-12-23 江苏省电力公司南京供电公司 High-sensitivity transformer winding loosening determination method

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2003072431A (en) * 2001-08-30 2003-03-12 Central Japan Railway Co Feeder circuit failure spotting device
CN103135035A (en) * 2011-11-25 2013-06-05 江西省电力科学研究院 Transformer winding state diagnosis method
CN103163420A (en) * 2011-12-08 2013-06-19 沈阳工业大学 Intelligent power transformer on-line state judgment method
CN102721465A (en) * 2012-06-13 2012-10-10 江苏省电力公司南京供电公司 System and method for diagnosing and preliminarily positioning loosening faults of iron core of power transformer
CN104102838A (en) * 2014-07-14 2014-10-15 河海大学 Transformer noise prediction method based on wavelet neural network and wavelet technology
CN105181120A (en) * 2015-09-02 2015-12-23 江苏省电力公司南京供电公司 High-sensitivity transformer winding loosening determination method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
王家炜: "基于模糊ART神经网络的变压器局部放电模式识别研究", 《中国优秀硕士学位论文全文数据库 工程科技‖辑》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106990303A (en) * 2017-03-15 2017-07-28 国家电网公司 A kind of Diagnosis Method of Transformer Faults
CN107290041A (en) * 2017-07-28 2017-10-24 河海大学 It is a kind of that state monitoring method is loosened based on phase space reconfiguration and the KPCM Transformer Winding clustered
CN107290041B (en) * 2017-07-28 2019-05-28 河海大学 A kind of transformer winding loosening state monitoring method clustered based on phase space reconfiguration and KPCM
CN107831415A (en) * 2017-10-20 2018-03-23 广东电网有限责任公司河源供电局 The Interval Valued Fuzzy diversity method that a kind of transformer insulating paper ageing state is assessed
CN107831415B (en) * 2017-10-20 2020-02-04 广东电网有限责任公司河源供电局 Interval value fuzzy set method for transformer insulation paper aging state evaluation
CN108414610A (en) * 2018-05-09 2018-08-17 南开大学 It is a kind of to compose construction method based on the comprehensive pollution derived components of individual particle aerosol mass spectrometer and ART-2a neural networks
CN108921124A (en) * 2018-07-17 2018-11-30 河海大学 A kind of on-load tap changers of transformers mechanical breakdown on-line monitoring method

Similar Documents

Publication Publication Date Title
CN104969438B (en) Nonlinear Systems Identification for the detection object in wireless power transmission system
CN103955750B (en) Rolling bearing remaining life prediction method based on feature fusion and particle filtering
CN102607845B (en) Bearing fault characteristic extracting method for redundantly lifting wavelet transform based on self-adaptive fitting
CN104655425B (en) Bearing fault classification diagnosis method based on sparse representation and LDM (large margin distribution machine)
Cui et al. Double-dictionary matching pursuit for fault extent evaluation of rolling bearing based on the Lempel-Ziv complexity
Sun et al. Statistical wavelet-based method for structural health monitoring
CN103544392B (en) Medical science Gas Distinguishing Method based on degree of depth study
CN100567978C (en) The supersonic phased array for detecting oil gas pipeline girth weld defect type automatic identifying method
CN103776480B (en) Small fault detection method based on repeatedly rolling average and device
CN103983757B (en) Based on the transformer insulated heat ageing state reliability estimation method of mixture Weibull distribution
US20160145994A1 (en) Evaluation Method and Evaluation Device for Water Breakthrough Risk of Production Wells in Aquifer Drive Gas Reservoirs
CN103617684B (en) Interference-type optical fiber circumference vibrating intruding recognizer
CN100520425C (en) Post-wavelet analysis treating method and device for electric power transient signal
CN102944418B (en) Wind turbine generator group blade fault diagnosis method
CN107480341B (en) A kind of dam safety comprehensive method based on deep learning
CN1331099C (en) Content based image recognition method
CN104598734B (en) Life prediction method of rolling bearing integrated expectation maximization and particle filter
CN105354587A (en) Fault diagnosis method for gearbox of wind generation unit
US20110066391A1 (en) Methods and systems for energy prognosis
CN101799367B (en) Electromechanical device neural network failure trend prediction method
CN107220618A (en) Method for detecting human face and device, computer-readable recording medium, equipment
CN107144428B (en) A kind of rail traffic vehicles bearing residual life prediction technique based on fault diagnosis
CN103472008B (en) Embryo Gallus domesticus gender identification method in hatching early stage near-infrared hatching egg
CN106556781A (en) Shelf depreciation defect image diagnostic method and system based on deep learning
CN102053016A (en) System for monitoring vibration of rotating machinery rolling bearing in wireless mode

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20170301