CN103235973B - A kind of Diagnosis Method of Transformer Faults based on radial base neural net - Google Patents

A kind of Diagnosis Method of Transformer Faults based on radial base neural net Download PDF

Info

Publication number
CN103235973B
CN103235973B CN201310133156.2A CN201310133156A CN103235973B CN 103235973 B CN103235973 B CN 103235973B CN 201310133156 A CN201310133156 A CN 201310133156A CN 103235973 B CN103235973 B CN 103235973B
Authority
CN
China
Prior art keywords
neural net
radial
fault type
base neural
radial base
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.)
Active
Application number
CN201310133156.2A
Other languages
Chinese (zh)
Other versions
CN103235973A (en
Inventor
禹建丽
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.)
Henan Zhengshu Intelligent Technology Co Ltd
Original Assignee
Zhengzhou University of Aeronautics
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 Zhengzhou University of Aeronautics filed Critical Zhengzhou University of Aeronautics
Priority to CN201310133156.2A priority Critical patent/CN103235973B/en
Publication of CN103235973A publication Critical patent/CN103235973A/en
Application granted granted Critical
Publication of CN103235973B publication Critical patent/CN103235973B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Other Investigation Or Analysis Of Materials By Electrical Means (AREA)

Abstract

The invention discloses a kind of Diagnosis Method of Transformer Faults based on radial base neural net, can using the content of characteristic gas in insulating oil as the input of radial base neural net, output transformer fault accurately, greatly improve the accuracy rate of transformer fault diagnosis, guarantee that transformer safety is reliably run.

Description

A kind of Diagnosis Method of Transformer Faults based on radial base neural net
Technical field
The present invention relates to a kind of Diagnosis Method of Transformer Faults, particularly relate to a kind of Diagnosis Method of Transformer Faults based on radial base neural net.
Background technology
Power transformer is one of most important equipment in State Grid's system, and be also that electric system is broken down one of maximum equipment, its running status directly affects the security of State Grid's system.Thus study transformer fault diagnosis technology, improve reliability and the security of transformer, have great importance.
In the research of transformer fault diagnosis, there is very complicated nonlinear mathematical relation in the sign of fault and fault type, and this makes the suitable mathematical model diagnosed be difficult to find.Wherein, the internal fault of transformer shows as machinery, electricity and hot three types, and below two kinds be main, and mechanical fault usually also can with electricity or heat show.Solid Insulation of Power Transformer material and oil are under the double action of electricity with heat, the gases such as hydrogen, hydro carbons, carbon monoxide and carbon dioxide can be produced, and these gases will be dissolved in the oil of inside transformer, by analyzing these gaseous species in oil and content, the fault of transformer just can be judged.The predominant gas used when wherein hydrogen, methane, acetylene, ethane, ethene, carbon monoxide, carbon dioxide are failure judgement, is called characteristic gas.
Three-ratio method is based upon on the basis of oil dissolved gas, is the diagnostic method of judgement transformer fault the most basic at present.Three-ratio method is by determining these three ratio sizes of C2H2/C2H4, CH4/H2 and C2H4/C2H6, then determines which kind of fault occurs transformer according to code of direct ratio rule (table 1) and fault type determination methods (table 2).But three-ratio method also exists significant limitation, only when each component concentration of oil dissolved gas all exceedes threshold values, three-ratio method could be adopted to carry out the fault diagnosis of transformer.In addition, in three-ratio method, lack a lot of coding, can cause in the coding rule table of three ratios, can not find corresponding ratio combination, cannot breakdown judge be carried out; Meanwhile, if the data calculated are in the border of code of direct ratio, judge that the transformer fault drawn is inaccurate by three-ratio method, be easy to diagnostic error.
The coding rule of table 1 three ratio
Table 2 fault type determination methods
Summary of the invention
The object of this invention is to provide a kind of Diagnosis Method of Transformer Faults based on radial base neural net, can using the content of characteristic gas in insulating oil as the input of radial base neural net, output transformer fault accurately, greatly improve the accuracy rate of transformer fault diagnosis, guarantee that transformer safety is reliably run.
The present invention adopts following technical proposals:
Based on a Diagnosis Method of Transformer Faults for radial base neural net, comprise the following steps:
A: collect training sample data and be used as input vector;
B: fault type is encoded, the correspondence table of establishment training sample and fault type;
C: build and train RBF Neural Network, until reach satisfied precision;
D: diagnosis sample to be tested, sample to be tested is inputted radial base neural net, and output vector after network calculations, obtains diagnostic result.
Training sample data in described step A are respectively H2, CH4, C2H4, C2H2, C2H6 and CO2 gas content.
First described training sample data input network again after normalized, and normalization formula is x i=(x i-x min)/(x max-x min), wherein, the numerical value of Xi representation feature gas, X minrepresent the numerical value that in all gas, numerical value is minimum, X maxrepresent the numerical value that in all gas, numerical value is maximum.
In described step B, the correspondence table of training sample and fault type is: if be encoded to 100000, then fault type is cryogenic overheating; If be encoded to 010000, then fault type is that middle temperature is overheated; If be encoded to 001000, then fault type is high-energy discharge; If be encoded to 000100, then fault type is hyperthermia and superheating; If be encoded to 000010, then fault type is that ground temperature is overheated; If be encoded to 000001, then fault type is shelf depreciation.
Described step C comprises according to the fault type coding schedule formulated and training sample structure radial base neural net, and training network, until reach satisfied precision; Described radial basis function adopts Gaussian function, and radial basis function is wherein, x is that n ties up input vector; C is the center of basis function, has the vector of same dimension with x, and δ determines the width of basis function around central point.
Matlab program is used in described step C, choice function newrb() carry out network design, method of calling is Net=newrb(P, T, GOAL, spread), wherein, P is input vector, T is for expecting output vector, GOAL is training precision, and spread is the dispersion constant of radial basic unit, and default value is 1.
The dispersion constant spread=10 of the radial basic unit of described radial basis function diagnosis.
The vector exported after network calculations in described D step, gets into 0 when being less than 0.5, gets into 1 when being greater than or equal to 0.5.
The present invention can using the content of characteristic gas in insulating oil as the input of radial base neural net, radial base neural net is trained, finally through training the radial base neural net that obtains can output transformer fault exactly, greatly improve the accuracy rate of transformer fault diagnosis, guarantee that transformer safety is reliably run.
Accompanying drawing explanation
Fig. 1 is process flow diagram of the present invention;
Fig. 2 is transformer fault diagnosis radial basis neural network schematic diagram of the present invention;
Fig. 3 be when precision be 0.02, the dispersion constant spread of radial basic unit be 5 time, radial base neural net training error variation diagram in MATLAB software;
Fig. 4 is 0.02 for working as precision, when the dispersion constant spread of radial basic unit is 10, and radial base neural net training error variation diagram in MATLAB software;
Fig. 5 is 0.02 for working as precision, when the dispersion constant spread of radial basic unit is 15, and radial base neural net training error variation diagram in MATLAB software.
Embodiment
As shown in Figure 1, the Diagnosis Method of Transformer Faults based on radial base neural net of the present invention comprises the following steps:
A: collect training sample data and be used as input vector; Training sample data are respectively H2, CH4, C2H4, C2H2, C2H6 and CO2 gas content, and first training sample data input network again after normalized, and normalization formula is x i=(x i-x min)/(x max-x min), wherein, the numerical value of Xi representation feature gas, X minrepresent the numerical value that in all gas, numerical value is minimum, X maxrepresent the numerical value that in all gas, numerical value is maximum;
B: known training sample is encoded respectively with corresponding fault type, the correspondence table of establishment training sample and fault type; The correspondence table of training sample and fault type is: if be encoded to 100000, then fault type is cryogenic overheating; If be encoded to 010000, then fault type is that middle temperature is overheated; If be encoded to 001000, then fault type is high-energy discharge; If be encoded to 000100, then fault type is hyperthermia and superheating; If be encoded to 000010, then fault type is that ground temperature is overheated; If be encoded to 000001, then fault type is shelf depreciation; Wherein, six codings are respectively six neurons of the output layer of radial base neural net, respectively corresponding six kinds of fault types.
C: build and train RBF Neural Network; Radial base neural net is built according to the fault type coding schedule formulated and training sample, and training network, until reach satisfied precision; Described radial basis function adopts Gaussian function, and radial basis function is wherein, x is that n ties up input vector; C is the center of basis function, has the vector of same dimension with x, and δ determines the width of basis function around central point; The present invention uses Matlab program, choice function newrb() carry out network design, method of calling is Net=newrb(P, T, GOAL, spread), wherein, P is input vector, T is for expecting output vector, and GOAL is training precision, and spread is the dispersion constant of radial basic unit, and default value is 1;
D: diagnosis sample to be tested, sample to be tested is inputted radial base neural net, and output vector after network calculations, obtains diagnostic result; The vector exported after network calculations, gets into 0 when being less than 0.5, gets into 1 when being greater than or equal to 0.5.
Owing to adopting these 6 kinds of characteristic gas of H2, CH4, C2H4, C2H2, C2H6 and CO2 as the input vector of network in the present invention, therefore the input layer number of network is confirmed as 6.To in the Fault Identification of transformer, adopt 6 kinds of fault types: the electric discharge of cryogenic overheating, middle cryogenic overheating, hyperthermia and superheating, low energy, high-energy discharge and shelf depreciation, the nodes of such output layer is also 6.Fig. 2 is transformer fault diagnosis radial basis neural network schematic diagram of the present invention, and wherein, C is middle hidden layer.
For the feature of transformer fault, in the present invention, adopt Gaussian function as radial basis function (RBF),
Radial basis function is R i ( x ) = exp [ - ( x - c i ) 2 2 δ i 2 ] , Wherein, x is that n ties up input vector; C is the center of basis function, has the vector of same dimension with x, and δ determines the width of basis function around central point; In the present invention, use Matlab program, choice function newrb() carry out network design.During with radial primary function network approximating function, newrb() function can increase the hidden neuron number of network, automatically until mean square deviation meets precision or neuron number reaches maximum.Method of calling is as follows:
Net=newrb(P, T, GOAL, spread), wherein, P is input vector, and T is for expecting output vector, and GOAL is training precision, and spread is the dispersion constant of radial basic unit, and default value is 1.
Radial basis function has the following advantages:
1: the form of performance is simple, even multivariable input also can not increase too much complicacy;
2: radial symmetry;
3: be convenient to theoretical analysis;
4: smoothing of functions is good, the derivative of arbitrary order all exists.
In order to ensure network learning information accurately and network there will not be saturated, and the scale of network is only too large, before data are inputted network, total data is all normalized, all data are all dropped in [0,1] interval.Normalization formula is x i=(x i-x min)/(x max-x min), wherein, the numerical value of Xi representation feature gas, X minrepresent the numerical value that in all gas, numerical value is minimum, X maxrepresent the numerical value that in all gas, numerical value is maximum.
Utilize in MATLAB2008 software carry out simulation training time, experiment is chosen 269 groups and was carried out fault type established data as training sample, and 64 groups of fault datas are as test samples.Language editor is carried out in MATLAB, as follows:
p=load('f:\1.txt');
p=p’
t=load('f:\2.txt');
t=t'
p1=load('f:\3.txt');
p1=p1'
pp=load('f:\4.txt');
pp=pp’
net=newrb(p,t,GOAL,spread);
zhenduan=sim(netrbf,p1)
In language editor above, the sample matrix that p representative sample data normalization is later, the fault type matrix of t representing fault sample, p1 represents the matrix after the normalization of sample to be tested, pp represents the right fault type matrix of sample to be tested itself, and the superincumbent content of GOAL and spread was introduced.
As shown in the connection weights IW between table 3 input layer and hidden layer and the connection weights LW between table 4 hidden layer and output layer,
Table 3: the connection weights IW between input layer and hidden layer
0.2092 0.8541 0.3 0.2766 0.0144 0.0639
1 0.3833 0.037 0.0903 0.3828 0.4998
0.3102 1 0.695 1 0.0574 0.5937
0.5041 0.4042 0.341 0.168 0.0065 0.4344
0.2806 0.3416 0.133 0.2258 0.0028 0.0462
0.2112 0.2798 0.347 0.2531 0.0225 0.0324
0.3878 0.0065 0.0031 0.0005 0.0006 0.056
0.3571 0.7397 0.18 0.0508 0.0225 0.2343
0.3878 0.0065 0.0031 0.0005 0.0006 0.056
0.5918 0.6725 1 0.2446 0.0038 0.1973
0.8071 0.2623 0.169 0.1054 0.0135 0.0979
0.1306 0.2818 0.27 0.1155 0.0005 0.514
0.0143 0.0303 0.034 0.0135 0 0.5739
0.4796 0.0256 0.016 0.0096 0.0439 0.4937
0.0898 0.1013 0.3306 0.0087 0.007 0.0257
0.2942 0.1509 0.031 0.0977 1 0.5208
0.4796 0.0256 0.016 0.0096 0.0574 0.4937
0.5429 0.2791 0.171 0.1007 0.0091 0.1269
0.8 0.1769 0.285 0.5344 0.0417 0.1084
0.2735 0.3779 0.129 0.1724 0.0176 0.0569
0.0321 0.0276 0.0103 0.0051 0.0123 0.5604
0.7816 0.6678 0.116 0.1251 0.0045 0.0176
0.4031 0.0639 0.01 0.0337 0.0833 0.4628
0.251 0.7128 0.216 0.2409 0.027 0.1899
0.2173 0.2313 0.071 0.2076 0.0225 0.0602
0.2592 0.0159 0.0075 0.0028 0.0372 0.469
0.0973 0.044 0.1991 0.0481 0.005 0.0422
0.1578 0.0342 0.0038 0.0088 0.172 0.3759
0.0643 0.1318 0.119 0.1852 0.0146 0.1232
0.8265 0.1903 0.138 0.1989 0.016 0.349
0.0306 0.035 0.2 0.0036 0.0236 0.3023
0.0653 0.1681 0.146 0.1592 0.0259 0.1633
0.2085 0.008 0.0159 0.0018 0 0.0224
0.0153 0.0619 0.166 0.1129 0.018 0.0719
0.8265 0.1903 0.138 0.1989 0.016 0.349
0.0214 0.0356 0.013 0.025 0 0.5857
0.0214 0.0531 0.013 0.0788 0 0.4875
0.0214 0.0531 0.013 0.0788 0 0.4875
0.0398 0.0114 0.138 0.0062 0 0.4109
0.0398 0.0114 0.138 0.0062 0 0.4109
0.2173 0.1555 0.0141 0.0623 0.6285 0.3726
0.0286 0.2751 0.354 0.1302 0.0006 0.0614
0.0847 0.0296 0.075 0.0884 0.0053 0.0012
0.2408 0.2757 0.159 0.1537 0.0039 0.1232
0.3347 0.0491 0.0093 0.0131 0.0698 0.3615
0.2408 0.0329 0.0133 0.0089 0.0994 0.3936
0.1051 0.0982 0.031 0.0534 0.0045 0.2399
0.1031 0.0861 0.013 0.0273 0.0113 0.2627
0.299 0.026 0.0384 0.004 0.0579 0.0468
0.3735 0.0815 0.0478 0.026 0.3394 0.2282
0.2668 0.0224 0.0279 0.0037 0.052 0.0439
0.173 0.0194 0.0034 0.0122 0.0915 0.3838
0.2481 0.0199 0.0284 0.0053 0.0529 0.045
0.2599 0.1381 0.0447 0.0133 0.0032 0.1327
0.0573 0.2562 0.193 0.1844 0.0134 0.1263
0.0484 0.0236 0.0078 0.004 0.0117 0.4776
0.3408 0.0268 0.0478 0.001 0.2787 0.0668
0.1071 0.0455 0.0406 0.033 0.0017 0.4196
0.1432 0.1302 0.0821 0.0803 0.0304 0.5509
0.151 0.1607 0.113 0.2322 0.0045 0.0824
0.3837 0.0161 0.022 0.0048 0.0757 0.1164
0.099 0.0935 0.028 0.0943 0.0041 0.1349
0.1798 0.0093 0.0081 0.0025 0.031 0.0288
0.2143 0.2824 0.016 0.0696 0.6417 0.469
0.0296 0.0177 0.0018 0.0051 0.0928 0.3226
0.2337 0.0549 0.01 0.0174 0.0851 0.4227
0.0571 0.0273 0.0097 0.0045 0 0.7067
0.3245 0.0767 0.052 0.0726 0.0372 0.1639
0.3245 0.0767 0.052 0.0726 0.0372 0.1639
0.0987 0.2182 0.0568 0.007 0.0305 0.1899
0.4447 0.0471 0.0231 0.0097 0.0652 0.9037
0.2775 0.0236 0.0282 0.0037 0.0619 0.056
0.1716 0.0072 0.0049 0.0019 0.0274 0.0332
0.0348 0.0348 0.0207 0.0254 0 0.3731
0.0527 0.0235 0.0075 0.0041 0.0116 0.4863
0.4286 0.1345 0.014 0.0433 0.2026 0.6789
0.0005 0.0111 0.0031 0.0027 0.0399 0.2344
0.0224 0.0025 0.0008 0.0016 0.0469 0.0429
0.0816 0.0444 0.018 0.0333 0.5741 0.1911
0.1127 0.0096 0.0146 0 0.096 0.0256
0.1367 0.0921 0.013 0.029 0.0124 0.3202
0.0571 0.1923 0.096 0.1746 0.0081 0.0367
0.0343 0.0187 0.0109 0.0139 0.0023 0.2263
0.0781 0.0648 0.0235 0.0292 0 0.9788
0.0781 0.0648 0.0235 0.0292 0 0.9788
0.2959 0.074 0.0068 0.0088 0.1013 0.9506
0.0235 0.041 0.014 0.0288 0.005 0.3158
0.0559 0.0115 0.0044 0.0015 0.0096 0.2359
0.1194 0.1843 0.054 0.0523 0.0033 0.0275
0.0352 0.0147 0.0032 0.0085 0.0221 0.0528
0.0235 0.041 0.014 0.0288 0.005 0.3158
0.0908 0.0067 0.003 0.0034 0 0.2572
0.299 0.0336 0.013 0.0216 0.1351 0.1973
0.0352 0.0224 0.0061 0.0172 0.0012 0.1096
0.0735 0.2972 0.221 0.0868 0.0008 0.0738
0.0735 0.2972 0.221 0.0868 0.0008 0.0738
0.1327 0.1681 0.036 0.0753 0.5629 0.0491
0.0949 0.0222 0.011 0.0113 0.0004 0.0411
0.1378 0.1547 0.083 0.079 0.0016 0.2899
0.0643 0.0108 0.003 0.0043 0.0338 0.2813
0.451 0.0572 0.01 0.0194 0.1959 0.6184
0.1276 0.2625 0.2019 0.0936 0.0008 0.4554
0.1276 0.2625 0.2019 0.0936 0.0008 0.4554
0.0653 0.0484 0.103 0.0013 0 0.1189
0.0653 0.0484 0.103 0.0013 0 0.1189
0.0051 0.0042 0.0011 0.0022 0.0509 0.0367
0.1224 0.0404 0.0022 0.0014 0.4165 0.0676
0.0143 0.0044 0.026 0.0003 0 0.0861
0.0969 0.0135 0.0033 0.0024 0.0036 0.6851
0.0531 0.0028 0.001 0.0018 0.0532 0.0621
0.1122 0.0343 0.004 0.0062 0.0642 0.0127
0.0459 0.0309 0.013 0.0186 0.0007 0.2158
0.0755 0.0464 0.026 0.025 0 0.3381
0.0755 0.0464 0.026 0.025 0 0.3381
0.1416 0.0351 0.0068 0.0118 0.0108 0.4264
0.0068 0.0067 0.011 0.0134 0.0044 0.2158
0.1796 0.0894 0.014 0.0406 0.4458 0.2152
0.0617 0.0684 0.0323 0.0581 0.0015 0.2501
0.0806 0.0182 0.0068 0.0085 0.0026 0.2714
0.0576 0.0439 0.1066 0.0082 0.0493 0.1399
0.0044 0.1298 0.118 0.0233 0 0.0367
0.0044 0.1298 0.118 0.0237 0 0.0367
0.0408 0.0128 0.0037 0.0034 0.0545 0.2238
0.0949 0.0114 0.005 0.0038 0.0349 0.1263
0.0929 0.2421 0.04 0.1355 0.0032 0.038
0.1263 0.1839 0.0544 0.0152 0.0196 0.0818
0.0533 0.018 0.1158 0.183 0.4 0
0.0449 0.043 0.0232 0.0416 0 0.1158
0.0449 0.043 0.0232 0.0416 0 0.1158
0.1081 0.1153 0.04 0.0627 0.0265 0.3023
0.2245 0.0062 0.0024 0.0004 0.0002 0.1337
0 0.0451 0.108 0.0275 0 0.0602
0.1265 0.0773 0.04 0.0358 0.0023 0.1788
0.0204 0.0101 0.0032 0.0036 0 0.2405
0.0881 0.1265 0.0714 0.0748 0.0029 0.3436
0.0755 0.0128 0.008 0.0055 0.0653 0.2535
0.0316 0.0296 0.046 0.0391 0.0028 0.1528
0.0044 0.1298 0.118 0.0235 0 0.0367
0.0005 0.0076 0.0028 0.0017 0.0036 0.2279
0.3635 0.0331 0.049 0.0355 0.0587 0.0649
0.0567 0.1477 0.0235 0.0718 0.0491 0
0.1633 0.15 0.045 0.0932 0.0124 0.2399
0.0837 0.1083 0.079 0.0764 0.01 0.6777
0.0837 0.1083 0.079 0.0764 0.01 0.6777
0.0195 0.0192 0.0035 0.0076 0 0.8589
0.0571 0.0672 0.0514 0.0468 0.0026 0.031
0.0694 0.0706 0.025 0.0442 0.0006 0.0398
0.0505 0.0185 0.0062 0.0035 0.009 0.4159
0.0561 0.0323 0.006 0.0134 0.0146 0.1077
0.272 0.502 0.2094 0.2019 0.0017 0.1176
0.0486 0.0104 0.003 0.0033 0.0391 0.3489
0.0224 0.0022 0.0012 0.0004 0.0002 0.2158
0.0334 0.0098 0.0028 0.003 0 0.6678
0.1482 0.0062 0.0126 0.001 0.0239 0.0305
0.043 0.0188 0.01 0.0096 0.0065 0.125
0.0547 0.0132 0.0048 0.0017 0.0096 0.3609
0.0476 0.0156 0.0044 0.0038 0.0609 0.2655
0.0166 0.0055 0.002 0.0031 0.0133 0.053
0 0.0022 0.0016 0.0014 0.0096 0.0164
0.0041 0.0013 0.001 0.0045 0 0.2516
0.0157 0.0011 0.0009 0.001 0.0243 0.0035
0.028 0.0425 0.0327 0.0259 0.0006 0.0343
0.0439 0.2999 0.0649 0.0571 0.0057 0.0952
0.1748 0.0171 0.0284 0.0181 0.0238 0.0304
0.0649 0.0298 0.0102 0.0057 0.012 0.7246
0.1592 0.0003 0.0005 0.0102 0.1362 0.0466
0.0184 0.002 0.001 0.0006 0.0048 0.1541
0.0051 0.0008 0.0009 0.0006 0.0037 0.0713
0.3418 0.0451 0.018 0.0269 0.1914 0.1849
0.0152 0.009 0.0064 0.0108 0.0025 0.1559
0.0823 0.1141 0.0453 0.0589 0.0032 0.2998
0 0.0451 0.108 0.0275 0 0.0602
0.0106 0.0032 0.0011 0.0005 0.002 0.0404
0.001 0.0007 0.0003 0.0001 0.0001 0.0244
0.0112 0.0054 0.0005 0.0021 0.0079 0.0182
0.0163 0.1594 0.092 0.0884 0 0.0966
0.0286 0.098 0.113 0.082 0 0.1423
0.0806 0.1722 0.132 0.1058 0 0.1454
0.0806 0.1722 0.132 0.1058 0 0.1454
0.0786 0.0807 0.044 0.0395 0.0009 0.1911
0.0398 0.0841 0.121 0.0775 0 0.0849
0.0551 0.1439 0.133 0.1383 0.0146 0.1257
0.0347 0.0052 0.0025 0.0019 0.0011 0.08
0.0475 0.0156 0.0044 0.0038 0.0609 0.2655
0.0969 0.0874 0.055 0.0414 0.0014 0.222
0.0745 0.074 0.041 0.0358 0 0.117
0.1082 0.0161 0.004 0.0053 0.0417 0.3418
0.0167 0.0168 0.011 0.004 0 0.9667
0.0167 0.0168 0.011 0.004 0 0.9667
0.0433 0.0063 0.0029 0.0054 0.0092 0.0973
0.2041 0.0323 0.014 0.022 0.1475 0.1541
0.0072 0.0514 0.0171 0.0183 0 0.0849
0.0072 0.0514 0.0171 0.0183 0 0.0849
0.0184 0.0538 0.07 0.0348 0 0.1145
0.0173 0.0094 0.0049 0.0024 0.0059 0.4653
0.0534 0.0137 0.0052 0.0018 0.0104 0.3555
0.0197 0.004 0.0019 0.0025 0.0097 0.0367
0.0235 0.0027 0.002 0.0019 0.0047 0.0065
0.0422 0.0154 0.0042 0.0035 0.0672 0.2762
0.0974 0.0727 0.0292 0.0364 0 0.9751
0.0245 0.0282 0.0036 0.0292 0.0039 0.0256
0.0245 0.0282 0.0036 0.0292 0.0039 0.0256
0.0098 0.0031 0.001 0.0004 0.0025 0.0331
0.0005 0.0008 0.0005 0.0004 0.002 0.0157
0.1429 0.1614 0.085 0.0772 0.0012 0.259
0.1429 0.1614 0.085 0.0772 0.0012 0.259
0.1633 0.0605 0.027 0.0032 0.0056 0.469
0.0757 0.0692 0.0254 0.0343 0 0.9801
0.049 0.0128 0.004 0.0045 0.0006 0.3578
0.0305 0.0162 0.0343 0.0285 0.0007 0.0185
0.0827 0.0049 0.0063 0.005 0.125 0.1788
0.0541 0.0921 0.115 0.0984 0.0073 0.0775
0.0824 0.0702 0.0278 0.0354 0 0.9162
0.0356 0.0409 0.068 0.0788 0.0016 0.0333
0.1184 0.0135 0.007 0.0051 0.0608 0.1392
0.1184 0.0135 0.007 0.0051 0.0608 0.1392
0.1133 0.0135 0.006 0.0045 0.0484 0.1337
0.0102 0.0027 0.002 0.0009 0.0034 0.0448
0.0327 0.1063 0.111 0.1186 0.0135 0.0948
0.0327 0.1063 0.111 0.1186 0.0135 0.0948
0 0.0025 0.0026 0.0012 0 0.0281
0.1694 0.0438 0.0059 0.0115 0.0656 0.9907
0.066 0.0365 0.0143 0.0188 0 0.8533
0.2408 0.0426 0.0077 0.0143 0.0783 1
0.1041 0.0026 0.0008 0.0014 0.0441 0.0429
0.2738 0.0261 0.0217 0.0017 0.052 0.0434
0.0296 0.0054 0.0012 0.0007 0.0047 0.0553
0.0494 0.0118 0.0063 0.0081 0.0004 0.0119
0.301 0.0247 0.0268 0.0051 0.0575 0.0482
0.0673 0.1681 0.145 0.1648 0.0225 0.1516
0.0367 0.0034 0.0018 0.0005 0.0007 0.3393
0.0827 0.1063 0.03 0.0228 0.0003 0.5943
0.0316 0.0511 0.0345 0.0259 0 0.2467
0.0143 0.0195 0.018 0.0335 0 0.046
0.0613 0.0135 0.0055 0.0018 0.0102 0.3584
0.1755 0.0413 0.0109 0.0153 0.1531 0.3412
0.0449 0.0377 0.0096 0.0115 0.0901 0.2467
0.0316 0.0511 0.0345 0.0259 0 0.2467
0.1645 0.0258 0.0044 0.0066 0.1412 0.1837
0.1939 0.0161 0.007 0.0032 0.0529 0.6789
0.001 0.0075 0.0049 0.0105 0 0.0676
0 0 0 0.0017 0 0.0251
0.0149 0.0073 0.0031 0.0041 0.1216 0.0506
0.0592 0.0968 0.0018 0.0309 0.041 0.0929
0.0592 0.0968 0.0018 0.0309 0.041 0.0929
0.1439 0.0948 0.013 0.0258 0.0124 0.3165
0.0041 0.0009 0.0004 0.0001 0.0005 0.0337
0.0367 0.0114 0.004 0.0024 0.0061 0.8722
0.0827 0.0049 0.0063 0.005 0.125 0.1788
0.0254 0.015 0.0075 0.0023 0.0298 0.5869
0.1306 0.0686 0.039 0.0591 0.0158 0.1411
0.0346 0.0245 0.0315 0.0074 0 0.8716
0.0092 0.0056 0.0037 0.0056 0.0056 0.0386
0.1183 0.051 0.0268 0.0387 0.0165 0.0362
0.0005 0.0011 0.0005 0.0001 0 0.0275
0.0214 0.0067 0.001 0.0007 0.0052 0.0831
0.0306 0.005 0.0018 0.0016 0.0214 0.0985
0.0582 0.1513 0.133 0.1357 0.018 0.1269
0.152 0.2469 0.1016 0.108 0.0018 0.1244
0.055 0.0613 0.0282 0.0342 0 0.8701
0.0005 0.0024 0.0007 0.0011 0.0087 0.0028
0.0583 0.066 0.0275 0.0321 0.0021 0.2708
0.1653 0.0235 0.0056 0.0056 0.0495 0.6937
0.2041 0.0323 0.0141 0.022 0.1475 0.1541
0.0204 0.0081 0.0012 0.0105 0.3085 0.0213
0.0122 0.0027 0.002 0.0011 0.0045 0.0448
0.0041 0.002 0.0011 0.0003 0 0.054
0.0427 0.0652 0.157 0.1129 0.0006 0.1306
0.0609 0.0663 0.0246 0.0328 0 0.8556
Table 4: the connection weights LW between hidden layer and output layer
-1.96E+10 1.16E+10 4.99E+08 8.53E+09 -4.75E+09 3.69E+09
-5.28E+09 4.48E+08 5.53E+08 3.29E+09 -3.01E+08 1.29E+09
6.86E+08 2.70E+07 -1.17E+08 -7.05E+08 9.63E+07 1.23E+07
1.56E+10 -1.42E+10 1.34E+10 -1.97E+09 -5.62E+09 -7.13E+09
5.35E+10 1.97E+10 -2.60E+10 4.36E+09 -3.66E+10 -1.49E+10
-1.38E+10 -3.28E+10 -3.41E+09 4.75E+10 1.54E+10 -1.29E+10
0 0 0 0 0 0
3.22E+09 -4.10E+09 -1.67E+09 2.18E+09 8.43E+08 -4.71E+08
-2.30E+11 2.37E+11 -3.87E+09 -5.81E+10 -5.38E+10 1.09E+11
9.63E+08 1.16E+09 -8.80E+08 -1.72E+09 2.78E+08 1.95E+08
6.11E+10 -2.16E+10 1.23E+09 -3.72E+10 6.29E+09 -9.84E+09
-6.62E+10 -3.35E+10 -1.32E+10 1.23E+11 -2.13E+10 1.09E+10
0 0 0 0 0 0
-1.61E+11 2.85E+10 1.25E+10 1.23E+11 -3.41E+10 3.10E+10
-4.68E+10 -6.85E+10 3.40E+10 1.03E+11 -1.52E+10 -6.59E+09
-1.54E+10 -2.59E+10 1.91E+10 1.96E+10 6.59E+09 -4.09E+09
0 0 0 0 0 0
-1.94E+11 7.11E+10 -2.41E+09 1.13E+11 -2.36E+10 3.59E+10
-1.60E+09 1.18E+09 7.40E+08 1.91E+08 -1.28E+09 7.61E+08
1.12E+11 -1.19E+10 -1.62E+10 -1.13E+11 3.42E+10 -5.50E+09
0 0 0 0 0 0
3.88E+09 1.58E+09 1.49E+08 -4.02E+09 -6.99E+08 -8.89E+08
3.02E+11 1.82E+10 1.69E+10 -2.58E+11 -3.03E+10 -4.85E+10
4.17E+10 8.54E+09 -5.70E+09 -3.91E+10 -3.40E+09 -2.08E+09
-2.39E+11 -9.23E+10 7.04E+10 2.63E+11 6.31E+09 -8.97E+09
0 0 0 0 0 0
8.48E+10 2.40E+11 -8.76E+10 -2.74E+11 5.74E+09 3.05E+10
0 0 0 0 0 0
0 0 0 0 0 0
-7.86E+09 1.73E+09 -5.35E+09 7.24E+09 4.99E+09 -7.39E+08
8.88E+10 -5.45E+10 -4.41E+10 -4.52E+10 9.15E+10 -3.65E+10
0 0 0 0 0 0
3.33E+11 -2.98E+11 2.51E+10 -1.98E+10 7.25E+10 -1.13E+11
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
2.39E+10 6.31E+10 2.59E+10 -1.48E+11 3.62E+09 3.10E+10
0 0 0 0 0 0
-9.27E+10 1.20E+11 2.34E+10 2.46E+10 -1.17E+11 4.24E+10
0 0 0 0 0 0
2.09E+11 2.68E+11 -2.13E+11 -2.18E+11 -7.22E+10 2.66E+10
-1.67E+10 4.78E+10 5.75E+09 -3.98E+10 -1.32E+10 1.62E+10
-1.76E+11 -5.71E+10 4.17E+10 2.35E+11 -2.81E+10 -1.56E+10
2.18E+11 1.87E+11 -1.33E+10 -4.24E+11 -4.39E+10 7.66E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
1.27E+11 7.08E+10 -7.50E+10 -9.14E+10 -1.61E+10 -1.50E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
1.09E+10 -8.51E+10 -7.29E+10 1.09E+11 6.63E+10 -2.82E+10
-1.39E+11 1.10E+10 3.17E+10 1.31E+11 -4.36E+10 8.03E+09
0 0 0 0 0 0
-1.25E+11 3.92E+10 3.71E+10 1.78E+10 -4.85E+09 3.60E+10
0 0 0 0 0 0
-2.30E+11 2.84E+11 -1.38E+11 5.10E+09 -2.30E+10 1.02E+11
1.08E+11 3.24E+10 -5.58E+09 -1.61E+11 2.20E+10 4.85E+09
2.30E+11 -2.82E+11 2.11E+10 1.21E+11 5.53E+10 -1.46E+11
0 0 0 0 0 0
0 0 0 0 0 0
-3.62E+10 -5.90E+10 4.22E+10 4.12E+10 1.75E+10 -5.90E+09
1.60E+11 -5.97E+10 -3.36E+10 -8.78E+10 3.26E+10 -1.17E+10
0 0 0 0 0 0
0 0 0 0 0 0
9.71E+10 -7.80E+10 -1.30E+10 -4.36E+10 4.20E+10 -4.54E+09
0 0 0 0 0 0
-2.93E+10 3.51E+11 -1.33E+11 -1.29E+11 -1.11E+11 5.19E+10
6.59E+10 -1.62E+09 -2.29E+10 -4.51E+10 1.57E+10 -1.20E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
-4.79E+10 -4.61E+10 3.85E+10 5.22E+10 3.51E+09 -2.05E+08
0 0 0 0 0 0
0 0 0 0 0 0
-5.14E+10 4.22E+10 -1.22E+09 -5.94E+09 -6.34E+09 2.27E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
3.26E+10 1.28E+11 -1.14E+11 -8.72E+09 -2.31E+10 -1.41E+10
0 0 0 0 0 0
0 0 0 0 0 0
-2.71E+11 -5.10E+10 2.80E+11 1.35E+11 -8.99E+10 -2.92E+09
0 0 0 0 0 0
0 0 0 0 0 0
-1.53E+11 3.31E+11 -1.23E+11 -2.45E+10 -5.56E+10 2.62E+10
0 0 0 0 0 0
0 0 0 0 0 0
2.33E+11 -8.99E+10 -2.28E+10 -8.92E+10 2.23E+10 -5.35E+10
0 0 0 0 0 0
-2.21E+10 1.75E+10 1.16E+09 -3.48E+08 -3.81E+09 7.58E+09
0 0 0 0 0 0
-1.47E+11 -1.72E+11 -1.16E+11 5.88E+11 -4.00E+10 -1.13E+11
0 0 0 0 0 0
-7.59E+10 -4.42E+09 8.46E+09 2.39E+10 1.90E+10 2.89E+10
0 0 0 0 0 0
6.00E+10 7.92E+10 2.54E+10 -2.16E+11 5.28E+10 -1.07E+09
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
2.09E+11 -3.47E+10 -9.85E+10 -6.35E+10 1.09E+10 -2.35E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
-3.56E+11 -4.34E+11 3.64E+11 3.77E+11 1.01E+11 -5.13E+10
0 0 0 0 0 0
0 0 0 0 0 0
-3.97E+10 -1.59E+11 1.45E+11 1.09E+11 -5.55E+10 4.02E+08
-9.60E+10 1.27E+11 -1.25E+11 -3.18E+09 5.28E+10 4.44E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
1.19E+11 1.40E+11 -9.18E+10 -1.93E+11 2.45E+08 2.52E+10
0 0 0 0 0 0
3.59E+10 2.35E+09 -1.70E+10 -1.68E+10 -1.08E+09 -3.35E+09
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
1.99E+11 4.14E+10 -1.28E+11 -1.42E+11 3.10E+10 -9.19E+08
0 0 0 0 0 0
0 0 0 0 0 0
2.56E+11 -2.01E+11 7.37E+10 -8.07E+10 4.51E+10 -9.30E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
-1.10E+11 -8.63E+10 2.03E+10 1.47E+11 3.74E+10 -8.45E+09
0 0 0 0 0 0
0 0 0 0 0 0
-3.77E+10 -6.91E+10 -4.70E+10 1.83E+11 4.80E+09 -3.43E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
-7.19E+10 -1.91E+11 8.56E+10 1.51E+11 4.43E+10 -1.74E+10
0 0 0 0 0 0
0 0 0 0 0 0
-2.02E+11 3.93E+11 -1.37E+11 -1.22E+11 -8.32E+10 1.51E+11
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
6.22E+10 2.46E+10 5.74E+09 -1.09E+11 1.64E+10 1.75E+08
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
-1.22E+11 7.61E+10 -6.24E+10 1.04E+11 -2.46E+10 2.86E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
2.02E+11 -3.13E+11 2.51E+11 -3.56E+11 1.79E+11 3.73E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
8.65E+09 -1.65E+11 1.58E+11 -2.36E+10 1.63E+10 5.03E+09
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
7.98E+09 -9.49E+10 9.74E+10 4.28E+10 -4.74E+10 -5.91E+09
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
-1.80E+11 1.41E+11 -1.18E+11 2.02E+11 -4.35E+10 -1.56E+09
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
2.56E+10 -1.41E+11 8.82E+10 2.54E+10 2.25E+10 -2.02E+10
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
3.82E+08 1.42E+10 2.55E+10 9.32E+07 -9.52E+09 -3.06E+10
0 0 0 0 0 0
0 0 0 0 0 0
2.70E+10 -1.39E+11 4.22E+10 5.10E+10 2.90E+10 -9.96E+09
0 0 0 0 0 0
Be 0.02 when choosing precision, when the dispersion constant spread of radial basic unit is 5, in MATLAB software, radial base neural net training error variation diagram as shown in Figure 3; The diagnostic result of radial base neural net to test samples is as follows:
By comparing with test samples physical fault, have 53 to check correct, accuracy is 82.81%.
Be 0.02 when choosing precision, when the dispersion constant spread of radial basic unit is 10, the radial base neural net training error variation diagram produced in MATLAB software as shown in Figure 4; The result of radial base neural net to test samples fault diagnosis is as follows:
By comparing with test samples physical fault, 55 inspections are correct, and accuracy is 85.94%.
Be 0.02 when choosing precision, when the dispersion constant spread of radial basic unit is 15, the radial base neural net training error variation diagram produced in MATLAB software as shown in Figure 5: the result that radial base neural net is diagnosed test samples is as follows:
By comparing with test samples physical fault, 52 inspections are correct, and accuracy is 81.25%, and the mean square deviation of l-G simulation test is 0.0479.
By to the comparative result of spread=5, spread=10, spread=14 as shown in rate of correct diagnosis table:
Rate of correct diagnosis table
Test samples number Diagnose positive exact figures Accuracy rate
SPread＝5 64 53 82.81％
SPread＝10 64 55 85.94％
SPread＝15 64 52 81.25％
Can be drawn by rate of correct diagnosis table and mean square deviation table, RBF neuron function when the present invention chooses spread=10 is diagnosed as the neural network model of transformer fault diagnosis, weight matrix IW (as table) between its input layer and hidden layer, the weight matrix LW between hidden layer and output layer are shown in annex, the threshold value equal 0.083255 of hidden layer, the threshold value of output layer is respectively-72129,250000,74263 ,-62261 ,-145000 ,-44636.
Carrying out in diagnosis sample to be tested process, sample to be tested is inputted radial base neural net, and the vector exported after network calculations, presses f ( x ) = 0 x < 0.5 1 x &GreaterEqual; 0.5 Process, get into 0 when the vector exported is less than 0.5, when being greater than or equal to 0.5, get into 1.
The Diagnosis Method of Transformer Faults based on radial base neural net of the present invention and traditional three-ratio method is used to compare:
Use three-ratio method to carry out fault diagnosis result to 60 groups of Sample data as shown in the table, cannot refer in table has no idea to determine fault type:
Three-ratio method diagnostic result
1 l l Cannot
l l 0 Low energy is discharged
l 2 2 Cannot
l 0 l Low energy is discharged
1 0 2 High-energy discharge
l 0 2 High-energy discharge
1 0 2 High-energy discharge
0 0 2 Cannot
0 0 2 Cannot
l 0 0 Cannot
0 0 l Cryogenic overheating
0 0 2 Cannot
l 0 l Low energy is discharged
l 0 l Low energy is discharged
l 0 1 Low energy is discharged
l 0 l Low energy is discharged
l 0 l Low energy is discharged
1 0 l Low energy is discharged
l 0 l Low energy is discharged
0 2 2 Hyperthermia and superheating
0 0 2 Cannot
0 2 2 Hyperthermia and superheating
0 2 2 Hyperthermia and superheating
0 2 2 Hyperthermia and superheating
l 0 2 High-energy discharge
0 0 2 Cannot
0 0 2 Cannot
0 0 2 Cannot
l 0 l Low energy is discharged
l 0 1 Low energy is discharged
l 0 0 Cannot
l 0 1 Low energy is discharged
1 0 l Low energy is discharged
l 0 l Low energy is discharged
1 0 2 High-energy discharge
0 0 2 Cannot
0 0 2 Cannot
0 0 2 Cannot
0 2 2 Hyperthermia and superheating
0 2 2 Hyperthermia and superheating
0 0 2 Cannot
0 0 2 Cannot
0 0 2 Cannot
0 0 2 Cannot
0 0 2 Cannot
0 0 2 Cannot
0 0 2 Cannot
l 0 0 Cannot
1 1 0 High-energy discharge
1 0 0 Cannot
1 0 l Low energy is discharged
1 0 0 Cannot
2 0 0 Cannot
1 0 2 High-energy discharge
0 0 2 Cannot
0 l 0 Low energy is discharged
0 0 l Cryogenic overheating
1 0 l Low energy is discharged
1 0 l Low energy is discharged
1 0 1 Low energy is discharged
1 0 l Low energy is discharged
1 0 l Low energy is discharged
l 0 l Low energy is discharged
2 0 1 Low energy is discharged
There are 30 groups of data to judge fault type as can be seen from the above table, have 10 groups of fault type diagnostic error, therefore have 32 groups of data can not correct failure judgement type.
And utilize the Diagnosis Method of Transformer Faults based on radial base neural net of the present invention to carry out fault type judgement, as shown in the rate of correct diagnosis table above as spread=10, rate of accuracy reached is to 85.94%.
Therefore, the Diagnosis Method of Transformer Faults based on radial base neural net of the present invention is utilized to carry out fault type judgement, can using the content of characteristic gas in insulating oil as the input of radial base neural net, output transformer fault exactly, greatly improve the accuracy rate of transformer fault diagnosis, guarantee that transformer safety is reliably run.

Claims (7)

1. based on a Diagnosis Method of Transformer Faults for radial base neural net, it is characterized in that: comprise the following steps:
A: collect training sample data and be used as input vector;
B: fault type is encoded, the correspondence table of establishment training sample and fault type;
C: build and train RBF Neural Network, until reach satisfied precision;
D: diagnosis sample to be tested, sample to be tested is inputted radial base neural net, and output vector after network calculations, obtains diagnostic result;
Training sample data in described step A are respectively H 2, CH 4, C 2h 4, C 2h 2, C 2h 6and CO 2gas content.
2. the Diagnosis Method of Transformer Faults based on radial base neural net according to claim 1, is characterized in that: first described training sample data input network again after normalized, and normalization formula is x i=(x i-x min)/(x max-x min), wherein, the numerical value of Xi representation feature gas, Xmin represents the numerical value that in all gas, numerical value is minimum, and Xmax represents the numerical value that in all gas, numerical value is maximum.
3. the Diagnosis Method of Transformer Faults based on radial base neural net according to claim 2, is characterized in that: in described step B, the correspondence table of training sample and fault type is: if be encoded to 100000, then fault type is cryogenic overheating; If be encoded to 010000, then fault type is that middle temperature is overheated; If be encoded to 001000, then fault type is high-energy discharge; If be encoded to 000100, then fault type is hyperthermia and superheating; If be encoded to 000010, then fault type is low energy electric discharge; If be encoded to 000001, then fault type is shelf depreciation.
4. the Diagnosis Method of Transformer Faults based on radial base neural net according to claim 3, it is characterized in that: described step C comprises according to the fault type coding schedule formulated and training sample structure radial base neural net, and training network, until reach satisfied precision; Described radial basis function adopts Gaussian function, and radial basis function is wherein, x is that n ties up input vector; C is the center of basis function, has the vector of same dimension with x, and δ determines the width of basis function around central point.
5. the Diagnosis Method of Transformer Faults based on radial base neural net according to claim 4, it is characterized in that: use Matlab program in described step C, choice function newrb () carries out network design, method of calling is Net=newrb (P, T, GOAL, spread), wherein, P is input vector, and T is for expecting output vector, and GOAL is training precision, spread is the dispersion constant of radial basic unit, and default value is 1.
6. the Diagnosis Method of Transformer Faults based on radial base neural net according to claim 5, is characterized in that: the dispersion constant spread=10 of the radial basic unit of described radial basis function diagnosis.
7. the Diagnosis Method of Transformer Faults based on radial base neural net according to claim 6, is characterized in that: the vector field homoemorphism exported after network calculations in described D step, gets into 0, get into 1 when being greater than or equal to 0.5 when being less than 0.5.
CN201310133156.2A 2013-04-16 2013-04-16 A kind of Diagnosis Method of Transformer Faults based on radial base neural net Active CN103235973B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201310133156.2A CN103235973B (en) 2013-04-16 2013-04-16 A kind of Diagnosis Method of Transformer Faults based on radial base neural net

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201310133156.2A CN103235973B (en) 2013-04-16 2013-04-16 A kind of Diagnosis Method of Transformer Faults based on radial base neural net

Publications (2)

Publication Number Publication Date
CN103235973A CN103235973A (en) 2013-08-07
CN103235973B true CN103235973B (en) 2016-04-27

Family

ID=48884012

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201310133156.2A Active CN103235973B (en) 2013-04-16 2013-04-16 A kind of Diagnosis Method of Transformer Faults based on radial base neural net

Country Status (1)

Country Link
CN (1) CN103235973B (en)

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103810328B (en) * 2014-01-16 2017-02-22 国家电网公司 Transformer maintenance decision method based on hybrid model
CN104063577B (en) * 2014-05-12 2017-02-15 吉林省电力科学研究院有限公司 Method for forecasting characteristic gas development tendency in transformer oil based on generalized recurrent neural network
CN105335759B (en) * 2015-11-12 2019-07-16 南方电网科学研究院有限责任公司 Transformer fault detection method based on probability model generation
CN105930861B (en) * 2016-04-13 2019-08-06 西安西拓电气股份有限公司 A kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm
CN106067066A (en) * 2016-05-31 2016-11-02 西安工程大学 Method for diagnosing fault of power transformer based on genetic algorithm optimization pack algorithm
CN106526370A (en) * 2016-10-27 2017-03-22 国家电网公司 RVM transformer fault diagnosing method based on chemical reaction algorithm
CN107024267A (en) * 2017-04-12 2017-08-08 无锡研测技术有限公司 Weighing-appliance Transducer-fault Detecting Method based on radial base neural net
CN110174255B (en) * 2019-06-03 2021-04-27 国网上海市电力公司 Transformer vibration signal separation method based on radial basis function neural network
CN110879373B (en) * 2019-12-12 2021-09-03 国网电力科学研究院武汉南瑞有限责任公司 Oil-immersed transformer fault diagnosis method with neural network and decision fusion
CN111076927A (en) * 2019-12-19 2020-04-28 佛山科学技术学院 Gear fault diagnosis method, device and system based on deep neural network
CN111060221A (en) * 2019-12-31 2020-04-24 云领电气智能科技(苏州)有限公司 Transformer overheating fault early warning method based on cyclic neural network
CN111947795B (en) * 2020-08-18 2022-07-12 广东电网有限责任公司广州供电局 Transformer overheating early warning method

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102221651A (en) * 2011-03-11 2011-10-19 太原理工大学 Fault on-line diagnosis and early warning method of flameproof dry-type transformer for mine
KR20110122238A (en) * 2010-05-04 2011-11-10 남창현 Fault diagnosis method and apparatus using headspace gases for transformer

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20110122238A (en) * 2010-05-04 2011-11-10 남창현 Fault diagnosis method and apparatus using headspace gases for transformer
CN102221651A (en) * 2011-03-11 2011-10-19 太原理工大学 Fault on-line diagnosis and early warning method of flameproof dry-type transformer for mine

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
The fault diagnosis of transformer based on BP neural Network;Jianli Yu etal;《2012 international conference on Information management,innovation management and industrial engineering》;20121020;第487-489页 *
基于RBF神经网络的变压器故障诊断及matlab仿真研究;吴宏岐等;《科学技术与工程》;20100228;第10卷(第5期);第1249-1251,1275页 *
基于径向基函数网络的大型电力变压器故障诊断;魏星等;《江苏电机工程》;20051130;第24卷(第6期);第19-21页 *

Also Published As

Publication number Publication date
CN103235973A (en) 2013-08-07

Similar Documents

Publication Publication Date Title
CN103235973B (en) A kind of Diagnosis Method of Transformer Faults based on radial base neural net
CN103062076B (en) A kind of urban water plant pump station separate unit characteristic curve of pump rating method
CN109978079A (en) A kind of data cleaning method of improved storehouse noise reduction self-encoding encoder
CN113011481B (en) Electric energy meter function abnormality assessment method and system based on decision tree algorithm
CN103218662A (en) Transformer fault diagnosis method based on back propagation (BP) neural network
RU2014130786A (en) SYSTEMS AND METHODS FOR EVALUATING MOMENTS OF A BREAKTHROUGH OF A FLUID IN LOCATIONS OF PRODUCING WELLS
CN105117512A (en) Transformer early-warning value estimation method and apparatus
CN105447646A (en) Health index assessment method for power distribution system
Xue et al. Prediction of gas drainage changes from nitrogen replacement: A study of a TCN deep learning model with integrated attention mechanism
CN105359003A (en) Hybrid approach to assisted history matching in large reservoirs
CN105092457A (en) Injection-production string corrosion evaluation method under combined action of alternating load and corrosion medium
Shaohu et al. Prediction of drilling plug operation parameters based on incremental learning and CNN-LSTM
Xue et al. Deep learning based production prediction for an enhanced geothermal system (EGS)
CN114186489A (en) Finished oil pipeline abnormity detection method, system and equipment based on sequencing network
Stewart et al. Carbon accounting for carbon dioxide enhanced oil recovery
Li et al. Economic benefit of shale gas exploitation based on back propagation neural network
Wu et al. Optimizing a monitoring scheme for CO2 geological sequestration: An environmentally sustainable and cost-efficient approach
Shah et al. Applications of Artificial Intelligence (AI) and Machine Learning (ML) in the Petroleum Industry
Cheng et al. Sensitivity of injection costs to input petrophysical parameters in numerical geologic carbon sequestration models
Bist et al. Artificial intelligence based optimizing solutions for the geothermal power plants
Katterbauer et al. A Sensor Based Hydrogen Volume Assessment Deep Learning Framework–A Pohokura Field Case Study
Jiang et al. Life Cycle CO 2 Emission Estimation of CCS-EOR System Using Different CO 2 Sources.
Kuczynski et al. Application of Monte Carlo simulations for economic efficiency evaluation of carbon dioxide enhanced oil recovery projects
CN104832210B (en) The gas absolute discharge Forecasting Methodology of Based PC A-FIG-SVM
CN111181172A (en) Power grid frequency disturbance source positioning method for scheduling master station

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
TR01 Transfer of patent right

Effective date of registration: 20191219

Address after: Room 2301-2312, 23 / F, block B, Weiye international, intersection of Zhengkai Avenue and Jinxiu Road, Baisha Town, Zhengdong New District, Zhengzhou City, Henan Province

Patentee after: Henan Zhengshu Intelligent Technology Co., Ltd

Address before: 450000 Henan Province, Zhengzhou District, No. 27 University Road, No. 2

Patentee before: Zhengzhou Institute of Aeronautical Industry Management

TR01 Transfer of patent right