CN103558519A - GIS partial discharge ultrasonic signal identification method - Google Patents

GIS partial discharge ultrasonic signal identification method Download PDF

Info

Publication number
CN103558519A
CN103558519A CN201310531833.6A CN201310531833A CN103558519A CN 103558519 A CN103558519 A CN 103558519A CN 201310531833 A CN201310531833 A CN 201310531833A CN 103558519 A CN103558519 A CN 103558519A
Authority
CN
China
Prior art keywords
sample
partial discharge
ultrasonic signals
index
follows
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201310531833.6A
Other languages
Chinese (zh)
Inventor
闫杰
王天正
芦山
刘晓飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Corp of China SGCC
Electric Power Research Institute of State Grid Shanxi Electric Power Co Ltd
Original Assignee
State Grid Corp of China SGCC
Electric Power Research Institute of State Grid Shanxi Electric Power Co Ltd
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 State Grid Corp of China SGCC, Electric Power Research Institute of State Grid Shanxi Electric Power Co Ltd filed Critical State Grid Corp of China SGCC
Priority to CN201310531833.6A priority Critical patent/CN103558519A/en
Publication of CN103558519A publication Critical patent/CN103558519A/en
Pending legal-status Critical Current

Links

Images

Abstract

The invention discloses a GIS partial discharge ultrasonic signal identification method which solves the problem that the accuracy and the reliability of GIS partial discharge ultrasonic detecting and diagnosing are not high. The method comprise a network learning process and a defect identification process and specifically comprises the following steps that first, a known sample of a GIS partial discharge ultrasonic signal is subjected to preprocessing, then an average amplitude value, a root-mean-square, a peak value index, kurtosis, a waveform index, a pulse index, a margin index and other discharge characteristic parameters are extracted, finally a fuzzy logic cluster neuron network is established, the GIS partial discharge ultrasonic signal to be identified is subjected to preprocessing, then the corresponding characteristic parameters are extracted, finally an established model is used for carrying out classification on all samples including samples to be identified, the fuzzy nearness of the samples to be identified and other known samples in the same type is computed, and the defect type is judged according to the magnitude of the nearness. The method has significance in GIS insulation condition assessment and reasonable overhaul strategy generating.

Description

The recognition methods of a kind of GIS Processing of Partial Discharge Ultrasonic Signals
Technical field
The present invention relates to insulation of electrical installation detection technique field, particularly a kind of GIS Processing of Partial Discharge Ultrasonic Signals recognition methods based on fuzzy logic clustering neuroid.
Background technology
Gas insulated combined electrical equipment (GIS) has the advantages such as floor area is little, reliability is high, high safety, operation maintenance convenience, therefore in electric system, is widely applied.There is in succession a lot of GIS faults or accident in electrical network in recent years, having a strong impact on the safe and stable operation of system, and the partial discharges fault diagnostic techniques of therefore studying GIS equipment is significant.
Shelf depreciation ultrasound examination is at present GIS equipment to be carried out to a kind of important means of fault diagnosis and insulation status assessment.Shelf depreciation can cause insulation system aging, causes insulation fault, shortens the serviceable life of equipment.The reason of GIS device interior generation shelf depreciation is varied, corresponding electric discharge type and also different to apparatus insulated influence degree size, therefore in equipment running process, not only to detect the size of shelf depreciation, also will further judge the type of insulation defect.Existing GIS Recognition of Partial Discharge utilizes high-frequency local discharging signal to identify more, the three-dimensional spectrum of extraction local discharge superhigh frequency signal, statistical nature parameter, fractal parameter, image moment characteristic parameter etc., reuse mode recognizer is identified, and Processing of Partial Discharge Ultrasonic Signals is owing to cannot extracting the phase information of electric discharge generation, therefore when concrete identification, there is limitation.Currently used algorithm for pattern recognition is main mainly with BP neural network, but BP neural network is owing to adopting gradient descent method, unavoidably can exist speed of convergence slowly, to be easily absorbed in local minimum point, be difficult to determine the problems such as hidden layer node number.Therefore find the recognition methods of effective GIS Processing of Partial Discharge Ultrasonic Signals, realize the accurate division of GIS shelf depreciation defect type, the accuracy and the reliability that improve GIS shelf depreciation ultrasound examination and diagnosis are problem demanding prompt solutions in current GIS Partial Discharge Detection.
Summary of the invention
The object of the invention is to overcome the shortcoming of prior art, the recognition methods of a kind of GIS Processing of Partial Discharge Ultrasonic Signals is provided, extract the time domain charactreristic parameter of GIS Processing of Partial Discharge Ultrasonic Signals, utilize fuzzy logic clustering neuroid to carry out discriminator to it, effectively improved accuracy and the reliability of GIS partial discharges fault diagnosis.
Object of the present invention is achieved through the following technical solutions:
A Processing of Partial Discharge Ultrasonic Signals recognition methods, comprises network learning procedure and defect recognition process,
Described network learning procedure comprises the following steps:
(1-1) input known GIS Processing of Partial Discharge Ultrasonic Signals as learning sample;
(1-2) the GIS Processing of Partial Discharge Ultrasonic Signals of step (1-1) input is carried out to pre-service;
(1-3) pretreated GIS Processing of Partial Discharge Ultrasonic Signals is extracted to following discharge characteristic parameter: average amplitude, r.m.s., peak index, kurtosis, waveform index, pulse index, nargin index;
(1-4) the discharge characteristic parameter of extracting with step (1-3) is carried out modeling, specifically comprises the following steps:
(1-4-1) the discharge characteristic parameter of all samples is carried out to regularization trans formation, using the discharge characteristic parameter after conversion as learning sample, form study sample set;
(1-4-2) set up fuzzy logic clustering neuroid preference pattern parameter;
(1-4-3) output to each learning sample computational grid, and adjust network parameter according to learning algorithm, when meeting the condition of convergence, learning process finishes, and obtains final network parameter, has set up GIS Processing of Partial Discharge Ultrasonic Signals model of cognition;
Described defect recognition process comprises:
(2-1) input GIS Processing of Partial Discharge Ultrasonic Signals to be identified;
(2-2) the GIS Processing of Partial Discharge Ultrasonic Signals to be identified of step (2-1) input is carried out to pre-service;
(2-3) characteristic parameter of the GIS Processing of Partial Discharge Ultrasonic Signals to be identified that extraction step (2-2) obtains: average amplitude, r.m.s., peak index, kurtosis, waveform index, pulse index, nargin index;
(2-4) to comprising that the discharge characteristic parameter of all samples of sample to be identified carries out regularization trans formation, using the discharge characteristic parameter after conversion as sample set, the GIS Processing of Partial Discharge Ultrasonic Signals model of cognition obtaining by step (1-4-3) calculates sample set, obtains its correspondence
Figure 389624DEST_PATH_IMAGE002
thereby, obtain treating the classification of diagnostic sample;
(2-5), according to the classification results of step (2-4), calculate the fuzzy nearness of other known sample in sample to be identified and same class;
(2-6) fuzzy nearness step (2-5) being calculated sorts according to size, and exchange premium degree is larger, and defect classification is more similar, thereby determines the defect classification of sample to be identified.
The described model parameter of step (1-4-2) comprises cluster centre vector v, network parameter w, cluster numbers c and study end condition
Figure 110193DEST_PATH_IMAGE004
.
The described output to each learning sample computational grid of step (1-4-3), and adjust network parameter according to learning algorithm, when meeting the condition of convergence, learning process finishes, obtain final network parameter, thereby set up GIS Processing of Partial Discharge Ultrasonic Signals model of cognition, be specially:
(1-4-3-1) initialization, initialization network parameter
Figure 460403DEST_PATH_IMAGE006
with
Figure 488402DEST_PATH_IMAGE008
, and hard clustering is counted c and study end condition , for
Figure 797340DEST_PATH_IMAGE010
can choose arbitrarily, but for
Figure 697163DEST_PATH_IMAGE012
must be initialized as very little value, this is to be avoided dead unit problem necessary, sets study algebraically
Figure 200957DEST_PATH_IMAGE014
;
(1-4-3-2) to each learning sample
Figure 502625DEST_PATH_IMAGE016
, calculate
Figure 302347DEST_PATH_IMAGE018
, then according to learning algorithm, adjust network parameter with
Figure 996951DEST_PATH_IMAGE022
;
(1-4-3-3) judgement
Figure 825230DEST_PATH_IMAGE024
whether set up, if set up, learning process finishes,
Figure 622284DEST_PATH_IMAGE026
be exactly cluster centre,
Figure 168803DEST_PATH_IMAGE028
which kind of the sample data that shows input belongs to, and obtains GIS Processing of Partial Discharge Ultrasonic Signals model of cognition; If be false,
Figure 778514DEST_PATH_IMAGE030
, turn to step (1-4-3-2).
Described network parameter
Figure 789195DEST_PATH_IMAGE020
with adjustment algorithm is as follows:
In network
Figure 790966DEST_PATH_IMAGE032
with
Figure 440253DEST_PATH_IMAGE034
be respectively the output of hidden node and output layer node,
Figure 305441DEST_PATH_IMAGE036
for the final output of neural network, its computing formula is as follows respectively:
Figure 381982DEST_PATH_IMAGE038
Figure 965410DEST_PATH_IMAGE040
Wherein, be i sample and the similarity of k cluster centre on j dimensional feature, it is defined as follows:
Figure 875270DEST_PATH_IMAGE046
Wherein, ,
Figure 874767DEST_PATH_IMAGE050
,
Figure 714547DEST_PATH_IMAGE052
,
Figure 631425DEST_PATH_IMAGE054
,
Figure 923866DEST_PATH_IMAGE056
If , so
Figure 412934DEST_PATH_IMAGE060
, the triumph of k class is described,
Figure 2178DEST_PATH_IMAGE062
with
Figure 47494DEST_PATH_IMAGE064
to adjust, now target output will be defined as
Figure 442703DEST_PATH_IMAGE066
if k class is won, so
Figure 929180DEST_PATH_IMAGE068
target output should be 1, so error may be defined as:
Figure 456369DEST_PATH_IMAGE070
If
Figure 723403DEST_PATH_IMAGE072
, judgement
Figure 859986DEST_PATH_IMAGE074
whether set up, if be false,
Figure 263285DEST_PATH_IMAGE076
, wherein for custom parameter; If set up, continue judgement whether set up, if set up,
Figure 888936DEST_PATH_IMAGE082
if, be false, , wherein
Figure 278384DEST_PATH_IMAGE086
for custom parameter.
Described data pre-service comprises the following steps:
(a) sample quantization: gather the GIS Processing of Partial Discharge Ultrasonic Signals in 1 cycle as an electric discharge sample;
(b) maximum value normalization: the maximal value by each data point in 1 electric discharge sample divided by data point in this electric discharge sample, computing formula is:
Figure 254430DEST_PATH_IMAGE088
,, wherein
Figure 732816DEST_PATH_IMAGE090
for original sample point, be the maximal value of data acquisition sampling point in an electric discharge sample, nfor sampled point number.
Described average amplitude is defined as follows:
Figure 711453DEST_PATH_IMAGE094
Wherein, nfor the sampling number of one-period, i.e. the number of data point in an electric discharge sample; u i for sampling point value;
Described r.m.s. is defined as follows:
Described peak index is defined as follows:
Wherein, U maxbe in an electric discharge sample sampling number according to the maximal value of absolute value;
Described kurtosis is defined as follows:
Wherein,
Figure 398600DEST_PATH_IMAGE102
for standard deviation, computing formula is as follows: ;
Described waveform index is defined as follows:
Figure 936471DEST_PATH_IMAGE106
Described pulse index is defined as follows:
Figure 288955DEST_PATH_IMAGE108
Described nargin index is defined as follows:
Figure 169187DEST_PATH_IMAGE110
Wherein,
Figure 607121DEST_PATH_IMAGE112
for root amplitude, computing formula is as follows:
Figure 965421DEST_PATH_IMAGE114
;
Described regularization trans formation is defined as: establish sample set and have nindividual sample to be sorted, each sample has mindividual characteristic index, sample set data matrix is:
Figure 437991DEST_PATH_IMAGE116
After each element of data matrix is deducted to the minimum value of this element column, then be regularization trans formation divided by the extreme difference of this column element (maximal value of this column element and minimum value poor), formula is as follows:
Figure 285861DEST_PATH_IMAGE118
Wherein:
Figure 148775DEST_PATH_IMAGE120
with
Figure 373083DEST_PATH_IMAGE122
be respectively jmaximal value and the minimum value of row.
The computing method of the described fuzzy nearness of step (2-5) are as follows:
Described fuzzy nearness is defined as: be provided with
Figure 136378DEST_PATH_IMAGE124
individual sample, each sample has mindividual index parameter, forms one n* mdata matrix, sample
Figure 420728DEST_PATH_IMAGE126
with sample
Figure 833255DEST_PATH_IMAGE128
between the computing formula of fuzzy nearness as follows:
Figure 2013105318336100002DEST_PATH_IMAGE131
In formula,
Figure 2013105318336100002DEST_PATH_IMAGE133
for sample in polymeric type kthe mean value of individual index parameter, ffor given parameters.
Vector norm described in step (1-4-3-3) adopts 2-norm,
Figure 2013105318336100002DEST_PATH_IMAGE135
computing formula as follows:
Figure 2013105318336100002DEST_PATH_IMAGE137
The feature large for BP neural computing amount, speed of convergence is slow, the present invention is applied to the diagnosis of GIS partial discharges fault by fuzzy logic clustering neuroid, the method adopts the logical operator in fuzzy set theory to complete network operations, add the feature that neural network concurrent is processed, increase substantially arithmetic speed, saved operation time.
Compared with prior art, the present invention has the following advantages and beneficial effect:
(1) the present invention utilizes fuzzy logic clustering neuroid to identify GIS Processing of Partial Discharge Ultrasonic Signals, and the method adopts Fuzzy Logic Operators to complete network operations, adds the feature that neural network concurrent is processed, and has increased substantially arithmetic speed.
(2) existing algorithm for pattern recognition needs first rule of thumb fault sample to be carried out to manual sort in learning process, and then with mode identification method, the mapping relations of this arteface are analyzed, this method lacks the scientific analysis to data structure.The present invention adopts the method for fuzzy clustering to carry out automatic classification to fault sample, compares with BP neural net method, and the method utilizes the principle of backpropagation to learn equally, but belongs to unsupervised study.
(3) the present invention is after classifying to the sample identified, calculate again this sample with similar in the fuzzy nearness of all the other known sample, by fuzzy nearness is sorted according to size, judge that sample to be identified and which known sample are the most approaching, thereby realize the identification of electric discharge type.
Accompanying drawing explanation
Fig. 1 is fuzzy logic clustering neuroid structural representation of the present invention;
Fig. 2 is fuzzy logic clustering neuroid study of the present invention and diagnostic flow chart;
Fig. 3 is of the present invention
Figure 590307DEST_PATH_IMAGE138
with
Figure DEST_PATH_IMAGE139
adjustment algorithm process flow diagram.
Embodiment
Below in conjunction with embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are not limited only to this.
Embodiment
Figure 1 shows that the fuzzy logic clustering neuroid structural drawing that the present invention adopts, n to be clustered sample forms sample set
Figure DEST_PATH_IMAGE141
, each sample represents by m index eigenwert: , all samples are divided into c class, in figure,
Figure DEST_PATH_IMAGE145
for input sample, for fear of the dead unit problem in Competitive Learning Algorithm and the network parameter of introducing,
Figure DEST_PATH_IMAGE149
for cluster centre vector,
Figure 360073DEST_PATH_IMAGE032
with
Figure 259896DEST_PATH_IMAGE034
be respectively the output of hidden node and output layer node,
Figure DEST_PATH_IMAGE150
final output for neural network.Its computing formula is as follows respectively:
Figure 763690DEST_PATH_IMAGE038
Figure 799779DEST_PATH_IMAGE040
Figure 425932DEST_PATH_IMAGE042
Wherein, be iindividual sample and kindividual cluster centre is jsimilarity on dimensional feature, it is defined as follows:
Wherein,
Figure 447349DEST_PATH_IMAGE048
,
Figure 244404DEST_PATH_IMAGE050
, ,
Figure 964416DEST_PATH_IMAGE054
,
Figure 975097DEST_PATH_IMAGE056
As shown in Figure 2, the recognition methods of a kind of GIS Processing of Partial Discharge Ultrasonic Signals, comprises e-learning stage and defect recognition stage.
Described network learning procedure comprises the following steps:
(1-1) input known GIS Processing of Partial Discharge Ultrasonic Signals as learning sample;
(1-2) the GIS Processing of Partial Discharge Ultrasonic Signals of step (1-1) input is carried out to pre-service:
(a) sample quantization: gather the GIS Processing of Partial Discharge Ultrasonic Signals in 1 cycle as an electric discharge sample;
(b) maximum value normalization: the maximal value by each data point in 1 electric discharge sample divided by data point in this electric discharge sample, computing formula is:
Figure 943053DEST_PATH_IMAGE088
,, wherein
Figure 976868DEST_PATH_IMAGE090
for original sample point,
Figure 422893DEST_PATH_IMAGE092
be the maximal value of data acquisition sampling point in an electric discharge sample, nfor sampled point number.
(1-3) pretreated GIS Processing of Partial Discharge Ultrasonic Signals is extracted to following discharge characteristic parameter: average amplitude, r.m.s., peak index, kurtosis, waveform index, pulse index, nargin index;
(1) average amplitude is defined as follows:
Wherein, nfor the sampling number of one-period, i.e. the number of data point in an electric discharge sample; u i for sampling point value;
(2) r.m.s. is defined as follows:
Figure 817027DEST_PATH_IMAGE096
(3) peak index is defined as follows:
Figure 955884DEST_PATH_IMAGE098
Wherein, U maxbe in an electric discharge sample sampling number according to the maximal value of absolute value;
(4) kurtosis is defined as follows:
Figure DEST_PATH_IMAGE152
Wherein,
Figure 476995DEST_PATH_IMAGE102
for standard deviation, computing formula is as follows:
Figure 664394DEST_PATH_IMAGE104
;
(5) waveform index is defined as follows:
Figure 649667DEST_PATH_IMAGE106
(6) pulse index is defined as follows:
Figure 897109DEST_PATH_IMAGE108
(7) nargin index is defined as follows:
Figure 967833DEST_PATH_IMAGE110
Wherein,
Figure 21240DEST_PATH_IMAGE112
for root amplitude, computing formula is as follows:
(1-4) the discharge characteristic parameter of extracting with step (1-3) is carried out modeling, specifically comprises the following steps:
(1-4-1) the discharge characteristic parameter of all samples is carried out to regularization trans formation, using the discharge characteristic parameter after conversion as learning sample, form study sample set;
Regularization trans formation is defined as: establish sample set and have nindividual sample to be sorted, each sample has mindividual characteristic index, sample set data matrix is:
Figure 777898DEST_PATH_IMAGE116
After each element of data matrix is deducted to the minimum value of this element column, then be regularization trans formation divided by the extreme difference of this column element (maximal value of this column element and minimum value poor), formula is as follows:
Figure DEST_PATH_IMAGE153
Wherein:
Figure DEST_PATH_IMAGE154
with be respectively jmaximal value and the minimum value of row.
(1-4-2) set up fuzzy logic clustering neuroid preference pattern parameter, comprise cluster centre vector v, network parameter w, cluster numbers cwith study end condition
Figure 945705DEST_PATH_IMAGE004
.
(1-4-3) output to each learning sample computational grid, and adjust network parameter according to learning algorithm, when meeting the condition of convergence, learning process finishes, and obtains final network parameter, has set up GIS Processing of Partial Discharge Ultrasonic Signals model of cognition, as shown in Figure 2, comprise following steps:
(1-4-3-1) initialization, initialization network parameter
Figure 802803DEST_PATH_IMAGE006
with
Figure 434773DEST_PATH_IMAGE008
, and hard clustering is counted c and study end condition
Figure 86334DEST_PATH_IMAGE004
, for
Figure 570798DEST_PATH_IMAGE010
can choose arbitrarily, but for
Figure 966007DEST_PATH_IMAGE012
must be initialized as very little value, this is to be avoided dead unit problem necessary, sets study algebraically
Figure 452484DEST_PATH_IMAGE014
;
(1-4-3-2) to each learning sample
Figure 540525DEST_PATH_IMAGE016
, calculate
Figure 807559DEST_PATH_IMAGE018
, then according to learning algorithm, adjust network parameter
Figure 678563DEST_PATH_IMAGE020
with ;
(1-4-3-3) judgement
Figure 606384DEST_PATH_IMAGE024
whether set up, if set up, learning process finishes,
Figure 32818DEST_PATH_IMAGE026
be exactly cluster centre,
Figure 35409DEST_PATH_IMAGE028
which kind of the sample data that shows input belongs to, and obtains GIS Processing of Partial Discharge Ultrasonic Signals model of cognition; If be false, , turn to step (1-4-3-2), wherein
Figure 995012DEST_PATH_IMAGE135
for vector norm, adopt 2-norm, computing formula is as follows: .
Described defect recognition process comprises:
(2-1) input GIS Processing of Partial Discharge Ultrasonic Signals to be identified;
(2-2) the GIS Processing of Partial Discharge Ultrasonic Signals to be identified of step (2-1) input is carried out to pre-service, disposal route is identical with step (1-2);
(2-3) characteristic parameter of the GIS Processing of Partial Discharge Ultrasonic Signals to be identified that extraction step (2-2) obtains: average amplitude, r.m.s., peak index, kurtosis, waveform index, pulse index, nargin index;
(2-4) to comprising that the discharge characteristic parameter of all samples of sample to be identified carries out regularization trans formation, using the discharge characteristic parameter after conversion as sample set, the GIS Processing of Partial Discharge Ultrasonic Signals model of cognition obtaining by step (1-4-3) calculates sample set, obtains its correspondence
Figure 338586DEST_PATH_IMAGE002
thereby, obtain treating the classification of diagnostic sample;
(2-5) according to the classification results of step (2-4), calculate the fuzzy nearness of other known sample in sample to be identified and same class, computing formula is as follows:
Described fuzzy nearness is defined as: be provided with individual sample, each sample has mindividual index parameter, forms one n* mdata matrix, sample with sample between the computing formula of fuzzy nearness as follows:
In formula,
Figure DEST_PATH_IMAGE158
for sample in polymeric type kthe mean value of individual index parameter, ffor given parameters.
(2-6) fuzzy nearness step (2-5) being calculated sorts according to size, and exchange premium degree is larger, and defect classification is more similar, thereby determines the defect classification of sample to be identified.
As shown in Figure 3, network parameter
Figure 980176DEST_PATH_IMAGE020
with
Figure 478154DEST_PATH_IMAGE022
adjustment algorithm is as follows:
In network
Figure 984221DEST_PATH_IMAGE032
with be respectively the output of hidden node and output layer node,
Figure 20627DEST_PATH_IMAGE036
for the final output of neural network, its computing formula is as follows respectively:
Figure 310795DEST_PATH_IMAGE038
Figure 987764DEST_PATH_IMAGE040
Wherein,
Figure 49577DEST_PATH_IMAGE044
be i sample and the similarity of k cluster centre on j dimensional feature, it is defined as follows:
Figure 256568DEST_PATH_IMAGE046
Wherein,
Figure DEST_PATH_IMAGE159
,
Figure 806236DEST_PATH_IMAGE050
,
Figure 731466DEST_PATH_IMAGE052
,
Figure 893457DEST_PATH_IMAGE054
,
Figure 220534DEST_PATH_IMAGE056
If
Figure 504884DEST_PATH_IMAGE058
, so
Figure DEST_PATH_IMAGE160
, the triumph of k class is described,
Figure DEST_PATH_IMAGE161
with
Figure 527198DEST_PATH_IMAGE139
to adjust, now target output will be defined as if k class is won, so
Figure DEST_PATH_IMAGE163
target output should be 1, so error may be defined as:
Figure 994345DEST_PATH_IMAGE070
If
Figure 175928DEST_PATH_IMAGE072
, judgement
Figure 631180DEST_PATH_IMAGE074
whether set up, if be false,
Figure 468686DEST_PATH_IMAGE076
, wherein
Figure 34796DEST_PATH_IMAGE078
for custom parameter; If set up, continue judgement whether set up, if set up,
Figure 946252DEST_PATH_IMAGE082
if, be false,
Figure 572405DEST_PATH_IMAGE084
, wherein
Figure DEST_PATH_IMAGE165
for custom parameter.
Above-described embodiment is preferably embodiment of the present invention; but embodiments of the present invention are not limited by the examples; other any do not deviate from change, the modification made under Spirit Essence of the present invention and principle, substitutes, combination, simplify; all should be equivalent substitute mode, within being included in protection scope of the present invention.

Claims (8)

1. a GIS Processing of Partial Discharge Ultrasonic Signals recognition methods, comprises network learning procedure and defect recognition process, it is characterized in that,
Described network learning procedure comprises the following steps:
(1-1) input known GIS Processing of Partial Discharge Ultrasonic Signals as learning sample;
(1-2) the GIS Processing of Partial Discharge Ultrasonic Signals of step (1-1) input is carried out to pre-service;
(1-3) pretreated GIS Processing of Partial Discharge Ultrasonic Signals is extracted to following discharge characteristic parameter: average amplitude, r.m.s., peak index, kurtosis, waveform index, pulse index, nargin index;
(1-4) the discharge characteristic parameter of extracting with step (1-3) is carried out modeling, specifically comprises the following steps:
(1-4-1) the discharge characteristic parameter of all samples is carried out to regularization trans formation, using the discharge characteristic parameter after conversion as learning sample, form study sample set;
(1-4-2) set up fuzzy logic clustering neuroid preference pattern parameter;
(1-4-3) output to each learning sample computational grid, and adjust network parameter according to learning algorithm, when meeting the condition of convergence, learning process finishes, and obtains final network parameter, has set up GIS Processing of Partial Discharge Ultrasonic Signals model of cognition;
Described defect recognition process comprises:
(2-1) input GIS Processing of Partial Discharge Ultrasonic Signals to be identified;
(2-2) the GIS Processing of Partial Discharge Ultrasonic Signals to be identified of step (2-1) input is carried out to pre-service;
(2-3) characteristic parameter of the GIS Processing of Partial Discharge Ultrasonic Signals to be identified that extraction step (2-2) obtains: average amplitude, r.m.s., peak index, kurtosis, waveform index, pulse index, nargin index;
(2-4) to comprising that the discharge characteristic parameter of all samples of sample to be identified carries out regularization trans formation, using the discharge characteristic parameter after conversion as sample set, the GIS Processing of Partial Discharge Ultrasonic Signals model of cognition obtaining by step (1-4-3) calculates sample set, obtains its correspondence thereby, obtain treating the classification of diagnostic sample;
(2-5), according to the classification results of step (2-4), calculate the fuzzy nearness of other known sample in sample to be identified and same class;
(2-6) fuzzy nearness step (2-5) being calculated sorts according to size, and exchange premium degree is larger, and defect classification is more similar, thereby determines the defect classification of sample to be identified.
2. GIS Processing of Partial Discharge Ultrasonic Signals according to claim 1 recognition methods, is characterized in that, the described model parameter of step (1-4-2) comprises cluster centre vector v , network parameter w , cluster numbers cwith study end condition .
3. GIS Processing of Partial Discharge Ultrasonic Signals according to claim 2 recognition methods, it is characterized in that, the described output to each learning sample computational grid of step (1-4-3), and adjust network parameter according to learning algorithm, when meeting the condition of convergence, learning process finishes, and obtains final network parameter, thereby set up GIS Processing of Partial Discharge Ultrasonic Signals model of cognition, be specially:
(1-4-3-1) initialization, initialization network parameter
Figure 218639DEST_PATH_IMAGE006
with
Figure 953377DEST_PATH_IMAGE008
, and hard clustering number cwith study end condition
Figure 562213DEST_PATH_IMAGE004
, for can choose arbitrarily, but for must be initialized as very little value, this is to be avoided dead unit problem necessary, sets study algebraically
Figure 324130DEST_PATH_IMAGE014
;
(1-4-3-2) to each learning sample
Figure 420262DEST_PATH_IMAGE016
, calculate
Figure 559295DEST_PATH_IMAGE018
, then according to learning algorithm, adjust network parameter
Figure 690062DEST_PATH_IMAGE020
with
Figure 828919DEST_PATH_IMAGE022
;
(1-4-3-3) judgement whether set up, if set up, learning process finishes,
Figure 599746DEST_PATH_IMAGE026
be exactly cluster centre, which kind of the sample data that shows input belongs to, and obtains GIS Processing of Partial Discharge Ultrasonic Signals model of cognition; If be false,
Figure 832461DEST_PATH_IMAGE030
, turn to step (1-4-3-2).
4. GIS Processing of Partial Discharge Ultrasonic Signals according to claim 1 recognition methods, is characterized in that, step (1-4-3-2) is described adjusts network parameter according to learning algorithm
Figure 637606DEST_PATH_IMAGE020
with
Figure 628696DEST_PATH_IMAGE022
method of adjustment is as follows:
In network with
Figure 385354DEST_PATH_IMAGE034
be respectively the output of hidden node and output layer node, for the final output of neural network, its computing formula is as follows respectively:
Figure 370125DEST_PATH_IMAGE040
Figure 21686DEST_PATH_IMAGE042
Wherein,
Figure 739106DEST_PATH_IMAGE044
be i sample and the similarity of k cluster centre on j dimensional feature, it is defined as follows:
Figure 134316DEST_PATH_IMAGE046
Wherein,
Figure 387836DEST_PATH_IMAGE048
,
Figure 210298DEST_PATH_IMAGE050
, ,
Figure 613915DEST_PATH_IMAGE054
,
Figure 282794DEST_PATH_IMAGE056
If , so
Figure 968170DEST_PATH_IMAGE060
, the triumph of k class is described, with
Figure 431829DEST_PATH_IMAGE064
to adjust, now target output will be defined as
Figure 596094DEST_PATH_IMAGE066
if k class is won, so
Figure 273938DEST_PATH_IMAGE068
target output should be 1, so error may be defined as:
Figure 814641DEST_PATH_IMAGE070
If
Figure 192533DEST_PATH_IMAGE072
, judgement
Figure 730961DEST_PATH_IMAGE074
whether set up, if be false,
Figure 194304DEST_PATH_IMAGE076
, wherein
Figure 476381DEST_PATH_IMAGE078
for custom parameter; If set up, continue judgement
Figure 708779DEST_PATH_IMAGE080
whether set up, if set up, if, be false,
Figure 368747DEST_PATH_IMAGE084
, wherein
Figure 251253DEST_PATH_IMAGE086
for custom parameter.
5. GIS Processing of Partial Discharge Ultrasonic Signals according to claim 1 recognition methods, is characterized in that, step (1-2) and the described data pre-service of step (2-2) comprise the following steps:
(a) sample quantization: gather the GIS Processing of Partial Discharge Ultrasonic Signals in 1 cycle as an electric discharge sample;
(b) maximum value normalization: the maximal value by each data point in 1 electric discharge sample divided by data point in this electric discharge sample, computing formula is: ,, wherein for original sample point,
Figure 423367DEST_PATH_IMAGE092
be the maximal value of data acquisition sampling point in an electric discharge sample, nfor sampled point number.
6. GIS Processing of Partial Discharge Ultrasonic Signals according to claim 1 recognition methods, is characterized in that,
Described average amplitude is defined as follows:
Figure 47247DEST_PATH_IMAGE094
Wherein, nfor the sampling number of one-period, i.e. the number of data point in an electric discharge sample; u i for sampling point value;
Described r.m.s. is defined as follows:
Figure 254237DEST_PATH_IMAGE096
Described peak index is defined as follows:
Figure 305370DEST_PATH_IMAGE098
Wherein, U maxbe in an electric discharge sample sampling number according to the maximal value of absolute value;
Described kurtosis is defined as follows:
Wherein,
Figure 454908DEST_PATH_IMAGE102
for standard deviation, computing formula is as follows:
Figure 719668DEST_PATH_IMAGE104
;
Described waveform index is defined as follows:
Figure 738439DEST_PATH_IMAGE106
Described pulse index is defined as follows:
Figure 150966DEST_PATH_IMAGE108
Described nargin index is defined as follows:
Figure 615183DEST_PATH_IMAGE110
Wherein,
Figure 796766DEST_PATH_IMAGE112
for root amplitude, computing formula is as follows:
Figure 189701DEST_PATH_IMAGE114
;
Described regularization trans formation is defined as: establish sample set and have nindividual sample to be sorted, each sample has mindividual characteristic index, sample set data matrix is:
Figure 823945DEST_PATH_IMAGE116
After each element of data matrix is deducted to the minimum value of this element column, then be regularization trans formation divided by the extreme difference of this column element (maximal value of this column element and minimum value poor), formula is as follows:
Wherein:
Figure 629407DEST_PATH_IMAGE120
with
Figure 255560DEST_PATH_IMAGE122
be respectively jmaximal value and the minimum value of row.
7. GIS Processing of Partial Discharge Ultrasonic Signals according to claim 1 recognition methods, is characterized in that, the computing method of the described fuzzy nearness of step (2-5) are as follows:
Described fuzzy nearness is defined as: be provided with
Figure 314783DEST_PATH_IMAGE124
individual sample, each sample has mindividual index parameter, forms one n* mdata matrix, sample
Figure 684584DEST_PATH_IMAGE126
with sample
Figure 840759DEST_PATH_IMAGE128
between the computing formula of fuzzy nearness as follows:
Figure 76962DEST_PATH_IMAGE130
Figure DEST_PATH_IMAGE131
In formula, for sample in polymeric type kthe mean value of individual index parameter, ffor given parameters.
8. GIS Processing of Partial Discharge Ultrasonic Signals according to claim 1 recognition methods, is characterized in that, vector norm described in step (1-4-3-3) adopts 2-norm, computing formula as follows:
。
CN201310531833.6A 2013-11-02 2013-11-02 GIS partial discharge ultrasonic signal identification method Pending CN103558519A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201310531833.6A CN103558519A (en) 2013-11-02 2013-11-02 GIS partial discharge ultrasonic signal identification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201310531833.6A CN103558519A (en) 2013-11-02 2013-11-02 GIS partial discharge ultrasonic signal identification method

Publications (1)

Publication Number Publication Date
CN103558519A true CN103558519A (en) 2014-02-05

Family

ID=50012821

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201310531833.6A Pending CN103558519A (en) 2013-11-02 2013-11-02 GIS partial discharge ultrasonic signal identification method

Country Status (1)

Country Link
CN (1) CN103558519A (en)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104407278A (en) * 2014-11-07 2015-03-11 国家电网公司 Extraction method and device of ultrasonic wave characteristic parameters
CN105842588A (en) * 2016-03-18 2016-08-10 深圳供电局有限公司 Method of correcting supersonic wave partial discharge detection and system thereof
CN106546892A (en) * 2016-11-10 2017-03-29 华乘电气科技（上海）股份有限公司 The recognition methodss of shelf depreciation ultrasonic audio and system based on deep learning
CN107132459A (en) * 2017-03-31 2017-09-05 国网浙江省电力公司电力科学研究院 A kind of partial discharge of transformer ultrasound locating method
CN110033451A (en) * 2019-04-17 2019-07-19 国网山西省电力公司电力科学研究院 A kind of power components defect inspection method based on SSD framework
CN110175508A (en) * 2019-04-09 2019-08-27 杭州电子科技大学 A kind of Eigenvalue Extraction Method applied to ultrasonic partial discharge detection
CN110837028A (en) * 2019-09-27 2020-02-25 中国船舶重工集团公司第七一九研究所 Method for rapidly identifying partial discharge mode
CN110907769A (en) * 2019-11-28 2020-03-24 深圳供电局有限公司 Defect detection method of closed gas insulated switchgear based on neural network
CN111239263A (en) * 2020-01-19 2020-06-05 国网宁夏电力有限公司电力科学研究院 Method and system for detecting foreign matter defects inside GIS equipment
CN112198400A (en) * 2020-09-30 2021-01-08 国网福建省电力有限公司漳州供电公司 High-voltage switch cabinet partial discharge online detection method based on spectrum sensing characteristic
CN112287789A (en) * 2020-10-20 2021-01-29 国网四川省电力公司电力科学研究院 GIS ultrahigh frequency method partial discharge map identification method based on deep learning
CN113514557A (en) * 2021-07-09 2021-10-19 华北电力科学研究院有限责任公司 GIS shell weld defect type identification method and device based on ultrasonic waves
CN114783011A (en) * 2022-06-22 2022-07-22 广东惠丰达电气设备有限公司 Ultrasonic imaging identification method for internal defects of GIS

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1825345A (en) * 2006-04-07 2006-08-30 上海电力学院 Load density index acquiring method based on fuzzy maxiumum approximation degree theory
KR100853725B1 (en) * 2007-06-20 2008-08-22 (주) 피에스디테크 Gis partial discharge diagnostic method and system thereof with prps algorithm
CN102539527A (en) * 2010-12-27 2012-07-04 湖北省电力试验研究院 GIS (gas insulated substation) partial discharge mode identification method based on ultrasonic testing
CN102809718A (en) * 2012-07-25 2012-12-05 华南理工大学 Ultra-high-frequency partial discharge signal identification method for gas insulated switchgear (GIS)

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1825345A (en) * 2006-04-07 2006-08-30 上海电力学院 Load density index acquiring method based on fuzzy maxiumum approximation degree theory
KR100853725B1 (en) * 2007-06-20 2008-08-22 (주) 피에스디테크 Gis partial discharge diagnostic method and system thereof with prps algorithm
CN102539527A (en) * 2010-12-27 2012-07-04 湖北省电力试验研究院 GIS (gas insulated substation) partial discharge mode identification method based on ultrasonic testing
CN102809718A (en) * 2012-07-25 2012-12-05 华南理工大学 Ultra-high-frequency partial discharge signal identification method for gas insulated switchgear (GIS)

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
岳璐等: "基于模糊逻辑聚类神经网络算法的中长期负荷预测", 《华北电力大学学报》, vol. 35, no. 2, 31 March 2008 (2008-03-31) *

Cited By (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104407278A (en) * 2014-11-07 2015-03-11 国家电网公司 Extraction method and device of ultrasonic wave characteristic parameters
CN105842588A (en) * 2016-03-18 2016-08-10 深圳供电局有限公司 Method of correcting supersonic wave partial discharge detection and system thereof
CN105842588B (en) * 2016-03-18 2018-09-28 深圳供电局有限公司 A kind of method and system for correcting ultrasonic wave Partial Discharge Detection
CN106546892A (en) * 2016-11-10 2017-03-29 华乘电气科技（上海）股份有限公司 The recognition methodss of shelf depreciation ultrasonic audio and system based on deep learning
CN107132459A (en) * 2017-03-31 2017-09-05 国网浙江省电力公司电力科学研究院 A kind of partial discharge of transformer ultrasound locating method
CN107132459B (en) * 2017-03-31 2019-07-09 国网浙江省电力有限公司电力科学研究院 A kind of partial discharge of transformer ultrasound locating method
CN110175508A (en) * 2019-04-09 2019-08-27 杭州电子科技大学 A kind of Eigenvalue Extraction Method applied to ultrasonic partial discharge detection
CN110033451A (en) * 2019-04-17 2019-07-19 国网山西省电力公司电力科学研究院 A kind of power components defect inspection method based on SSD framework
CN110837028A (en) * 2019-09-27 2020-02-25 中国船舶重工集团公司第七一九研究所 Method for rapidly identifying partial discharge mode
CN110837028B (en) * 2019-09-27 2021-08-31 中国船舶重工集团公司第七一九研究所 Method for rapidly identifying partial discharge mode
CN110907769A (en) * 2019-11-28 2020-03-24 深圳供电局有限公司 Defect detection method of closed gas insulated switchgear based on neural network
CN111239263A (en) * 2020-01-19 2020-06-05 国网宁夏电力有限公司电力科学研究院 Method and system for detecting foreign matter defects inside GIS equipment
CN111239263B (en) * 2020-01-19 2022-10-28 国网宁夏电力有限公司电力科学研究院 Method and system for detecting foreign matter defects in GIS equipment
CN112198400A (en) * 2020-09-30 2021-01-08 国网福建省电力有限公司漳州供电公司 High-voltage switch cabinet partial discharge online detection method based on spectrum sensing characteristic
CN112198400B (en) * 2020-09-30 2024-04-09 国网福建省电力有限公司漳州供电公司 High-voltage switch cabinet partial discharge online detection method based on frequency spectrum sensing characteristics
CN112287789A (en) * 2020-10-20 2021-01-29 国网四川省电力公司电力科学研究院 GIS ultrahigh frequency method partial discharge map identification method based on deep learning
CN113514557A (en) * 2021-07-09 2021-10-19 华北电力科学研究院有限责任公司 GIS shell weld defect type identification method and device based on ultrasonic waves
CN114783011A (en) * 2022-06-22 2022-07-22 广东惠丰达电气设备有限公司 Ultrasonic imaging identification method for internal defects of GIS
CN114783011B (en) * 2022-06-22 2022-09-06 广东惠丰达电气设备有限公司 Ultrasonic imaging identification method for internal defects of GIS

Similar Documents

Publication Publication Date Title
CN103558519A (en) GIS partial discharge ultrasonic signal identification method
CN109933881A (en) A kind of Fault Diagnosis of Power Electronic Circuits method based on optimization deepness belief network
CN106197999B (en) A kind of planetary gear method for diagnosing faults
CN109213121B (en) Method for diagnosing clamping cylinder fault of fan braking system
CN111680875B (en) Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model
CN107766816A (en) A kind of Mechanical Failure of HV Circuit Breaker recognition methods based on LVQ neutral nets
CN110108992B (en) Cable partial discharge fault identification method and system based on improved random forest algorithm
CN109597401A (en) A kind of equipment fault diagnosis method based on data-driven
CN110399854B (en) Rolling bearing fault classification method based on mixed feature extraction
CN112557826A (en) Ship electric power system fault diagnosis method
CN102279358A (en) MCSKPCA based neural network fault diagnosis method for analog circuits
CN110796120A (en) Time domain feature-based circuit breaker mechanical fault XGboost diagnosis method
CN107942210A (en) The classification of transformer pulse electric current Partial Discharge and denoising method and system
CN114325256A (en) Power equipment partial discharge identification method, system, equipment and storage medium
CN108709744A (en) Motor bearings method for diagnosing faults under a kind of varying load operating mode
CN115587309A (en) Method, device and equipment for extracting key features of short-circuit resistance of transformer
CN115600088A (en) Distribution transformer fault diagnosis method based on vibration signals
Li et al. Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy
CN112462355A (en) Sea target intelligent detection method based on time-frequency three-feature extraction
CN115587290A (en) Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network
CN108508297A (en) A kind of fault arc detection method based on change coefficient and SVM
CN114330430A (en) Elevator fault judgment method and system based on big data characteristic analysis
CN110020637A (en) A kind of analog circuit intermittent fault diagnostic method based on more granularities cascade forest
CN116466408B (en) Artificial neural network superbedrock identification method based on aeromagnetic data
CN113378652A (en) Disturbance classification method based on EWT-MPE-PSO-BP

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
C02 Deemed withdrawal of patent application after publication (patent law 2001)
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20140205