CN110457750B - On-load tap-changer spring energy storage deficiency fault identification method based on neural network response surface - Google Patents
On-load tap-changer spring energy storage deficiency fault identification method based on neural network response surface Download PDFInfo
- Publication number
- CN110457750B CN110457750B CN201910616899.2A CN201910616899A CN110457750B CN 110457750 B CN110457750 B CN 110457750B CN 201910616899 A CN201910616899 A CN 201910616899A CN 110457750 B CN110457750 B CN 110457750B
- Authority
- CN
- China
- Prior art keywords
- model
- fault
- energy storage
- neural network
- spring
- 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
Links
- 230000004044 response Effects 0.000 title claims abstract description 104
- 238000004146 energy storage Methods 0.000 title claims abstract description 66
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 52
- 238000000034 method Methods 0.000 title claims abstract description 51
- 230000007812 deficiency Effects 0.000 title claims abstract description 15
- 238000013461 design Methods 0.000 claims abstract description 35
- 238000012360 testing method Methods 0.000 claims abstract description 25
- 238000004088 simulation Methods 0.000 claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 15
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 8
- 238000000611 regression analysis Methods 0.000 claims abstract description 5
- 239000007787 solid Substances 0.000 claims abstract description 4
- 238000004458 analytical method Methods 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 6
- 238000005094 computer simulation Methods 0.000 claims description 4
- 238000009434 installation Methods 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 description 7
- 238000003745 diagnosis Methods 0.000 description 6
- 239000000463 material Substances 0.000 description 6
- 230000005284 excitation Effects 0.000 description 5
- 238000004140 cleaning Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 4
- 230000000875 corresponding effect Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000010219 correlation analysis Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000002131 composite material Substances 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000005381 potential energy Methods 0.000 description 2
- 238000007670 refining Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000010079 rubber tapping Methods 0.000 description 2
- 206010023230 Joint stiffness Diseases 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000005489 elastic deformation Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000010206 sensitivity analysis Methods 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a neural network response surface-based on-load tap changer spring energy storage deficiency fault identification method. The method is characterized by comprising the following steps of: based on a finite element method, establishing an on-load tap-changer fault simulation model by using a solid finite element grid; generating training samples of a neural network response surface model by adopting a uniform design method, and obtaining output fault characteristics through simulation of the fault of insufficient energy storage of a tap changer spring; constructing a neural network response surface model through regression analysis of input parameters and output characteristics of the tap changer spring energy storage deficiency fault; and taking the output characteristics of the failure caused by insufficient energy storage of the spring as a reference, and adopting a multi-objective recognition algorithm based on a willingness function to recognize the output failure characteristics. The method is different from the traditional fault identification model based on test data, and has the advantages of high modeling efficiency, high fault identification precision and the like.
Description
Technical Field
The invention relates to a neural network response surface-based on-load tap changer spring energy storage deficiency fault identification method.
Background
An On-load Tap Changer (OLTC) is the only movable component for realizing voltage regulation of the converter transformer, not only can stabilize the load center voltage, but also is an indispensable important device for connecting a power grid, adjusting load flow, improving reactive power distribution and the like, and is subjected to huge mechanical and current impact in the frequent operation process, and mechanical and electrical faults such as switch braking failure, contact overheat burning loss, insufficient spring energy storage, loosening and falling of a fastener can occur along with the continuous increase of the number of times of actuation. Statistics data show that the number of faults of the on-load tap-changer accounts for about 40% of the total faults of the transformer, and most of the faults are caused by mechanical faults, so that the mechanical faults of the tap-changer are very necessary to be researched, and the fault generation possibility is predicted and the fault type is judged so as to improve the operation safety and stability of the power system.
Numerous engineering practices have shown that vibration caused by collisions or friction of the mechanical parts of the tap changer during operation is a major cause of mechanical failure of the tap changer. The research of mechanical faults of the on-load tap-changer based on the vibration principle is attracting general attention and importance of scientific researchers. However, the existing fault diagnosis methods of the tap switch mostly depend on test data of real objects, a large amount of fault data is needed for establishing a fault diagnosis model by adopting the methods, but because the reliability of the tap switch in the initial stage of use is higher, equipment defects and fault state data are deficient, the established model accuracy is not ideal enough, and the construction speed of part of the diagnosis model is slower, the part of the diagnosis model is easy to fall into a local optimal solution, so that the faults cannot be effectively diagnosed, therefore, a reasonable and effective method for identifying and diagnosing the typical mechanical faults of the tap switch is urgently needed, and on the basis, the operation and maintenance of the tap switch are guided, so that the method has important engineering significance for improving the operation reliability and safety of the tap switch.
Disclosure of Invention
The invention aims to provide a method for identifying the fault of insufficient energy storage of an on-load tap-changer spring based on a neural network response surface, which can quickly and effectively identify the fault of insufficient energy storage of the on-load tap-changer spring.
The on-load tap-changer spring energy storage shortage fault identification method based on the neural network response surface is characterized by comprising the following steps of:
(1) Based on a finite element method, establishing an on-load tap-changer fault simulation model by using a solid finite element grid;
(2) Generating training samples of a neural network response surface model by adopting a uniform design method, and obtaining output fault characteristics through simulation of the fault of insufficient energy storage of a tap changer spring;
(3) Constructing a neural network response surface model through regression analysis of input parameters and output characteristics of the tap changer spring energy storage deficiency fault;
(4) And taking the output characteristics of the failure caused by insufficient energy storage of the spring as a reference, and adopting a multi-objective recognition algorithm based on a willingness function to recognize the output failure characteristics.
The step (1) is specifically as follows:
(1) finite element modeling of individual components: in the iterative analysis of the sub-step grid size, the mode confidence criterion MAC is adopted to evaluate the correlation of the vibration modes of the current model and the reference model, and the MAC is defined as
In phi r,i Is the ith order vibration vector of the reference model, phi a,j The j-th order vibration mode vector of the analysis model is represented by T;
in the sub-step model convergence analysis, the average frequency error and the maximum frequency error of the current model and the reference model are used for measuring the convergence of the grid, and the average frequency error eta and the maximum frequency error delta are respectively expressed as
δ＝max|f i,j -f i+1,j |,j＝1,2,...,n；
Wherein f i,j The j-th order modal frequency of the model representing the i-th iterative calculation is represented, and n is the order of the modes;
(2) finite element modeling of assembled structures: modeling of the connecting piece is realized by adopting a connection relation between the one-dimensional spring unit and the thin layer unit simulation component, the assembly modeling of the whole machine model is completed by the connecting piece model according to the assembly relation under an assembly coordinate system, and boundary conditions are applied according to the installation condition of an actual structure;
(3) and (3) fault simulation modeling: the energy storage failure of the spring is simulated by reducing the contact exciting force, and the relation between the spring energy storage and the contact exciting force is converted according to the rigidity and the deformation of the spring.
The step (2) is to adopt a uniform test design method to provide training input samples for the neural network response surface model, wherein the uniform design samples are generated by adopting a good lattice point method, the discreteness of the samples is evaluated by adopting a divergence, the relation of the sample point divergence along with the sample test times is established, a design sample with better uniformity but fewer test times is selected as the training sample of the neural network response surface model, and the output fault characteristics corresponding to the input sample points are obtained by simulating the energy storage shortage faults of the split switch springs.
The step (3) is to establish a mapping relation between the fault input characteristics of the insufficient energy storage of the tap changer spring and the output response by adopting a BP neural network model, and to construct a neural network response surface model, wherein the fault input characteristics correspond to the mechanical parameters of the insufficient energy storage of the spring, namely the contact exciting force, the output response is the dynamic response characteristics sensitive to the comparison of the input mechanical parameters, the fitting rule of the neural network response surface model uses a gradient descent method, the weight and the threshold of the network are continuously adjusted through counter propagation, the error square sum of the actual output value and the expected output value of the network is minimum, and the decision coefficient R is selected 2 And the Root Mean Square Error (RMSE) is used as an evaluation index of the neural network response surface model precision.
Step (4) is to solve the problem of identifying a plurality of spring faults with insufficient energy storage by using a willingness function based on a neural network response surface model, wherein the willingness function is used for converting the response characteristic of the objective characteristic in the case that the quality characteristic target is the objective characteristic when a plurality of targets are identified, and the form is that
Wherein L is i ，U i The lower limit and the upper limit of the response y are respectively T i Is a target value for response y;
by converting the single willingness function, an optimized model of the composite willingness function is established, and the calculation expression is as followsCalculating a weighted geometric mean value of the model through the formula, so that the weighted geometric mean value tends to be maximized, and the recognition problem of the multi-target response is converted into a problem of single-target response; in the former, ω i (i=1, 2, l, n) represents the weight of the willingness function of the ith response; Σω i =1 is the sum of the response quality characteristic weights, where ω 1 ＝ω 2 ＝L＝ω n 。
The invention provides a method for identifying a fault with insufficient energy storage of a spring of an on-load tap-changer based on a neural network response surface based on a uniform test design technology, a neural network response surface modeling technology and a multi-target identification technology, and aims to quickly and effectively identify the fault with insufficient energy storage of the spring by means of a high-precision finite element simulation model of the tap-changer under the condition of lacking of real sample model fault test data.
Drawings
FIG. 1 is a flow of identifying an under-energy storage fault of an on-load tap changer spring based on a neural network response surface;
FIG. 2 is a topology of a BP neural network;
FIG. 3 is a plot of the divergence of a uniform test design sample as a function of test number;
FIG. 4 is a graph of a willingness function when the quality characteristic target is a look-ahead;
FIG. 5 is a comparison of a first set of identified fault signature response curves;
FIG. 6 is a comparison of a second set of identified fault signature response curves.
Detailed Description
The method comprehensively utilizes a uniform test design technology, a neural network response surface modeling technology and a multi-target identification technology, and an on-load tap-changer fault simulation technology.
The method is based on a neural network response surface model to identify the fault of insufficient energy storage of the tap-changer spring, and firstly, a high-precision tap-changer fault simulation model is established by a solid finite element grid based on a finite element method; then, a training sample of a neural network response surface model is generated by adopting a uniform design method, and output fault characteristics are obtained through simulation of the fault of insufficient energy storage of the tap switch; then, constructing a neural network response surface model through regression analysis of input parameters/output characteristics of the tap switch spring energy storage deficiency fault; and finally, taking the output characteristics of the failure caused by insufficient energy storage of the spring as a reference, and adopting a multi-target recognition algorithm to recognize the output failure characteristics.
The process for establishing the tap changer spring energy storage shortage fault simulation model mainly comprises 3 steps: (1) finite element modeling of individual components: the method comprises the steps of geometric model cleaning, initial model analysis, grid size iterative analysis, model convergence analysis and the like, wherein the average frequency error and the maximum frequency error of a current model and a reference model are adopted in the model convergence analysis to measure the convergence of the grid, and finally, a component finite element model with higher precision is obtained; (2) finite element modeling of assembled structures: modeling of the connecting piece is realized by adopting a connection relation between the one-dimensional spring unit and the thin-layer unit simulation component, wherein the one-dimensional spring unit mainly considers the rigidity of an actual spring structure and simulates a point-point connection relation; the thin layer unit is mainly used for simulating a surface-surface contact connection structure such as pin connection and bolt connection, is built by adopting isotropic materials with linear constitutive relation, has a normal elastic constant and a tangential elastic constant which are not independent, and determines the connection rigidity of the components by using the two elastic constants; under an assembly coordinate system, completing assembly modeling of a complete machine model by assembling a connecting piece model, and applying boundary conditions according to the installation condition of an actual structure; (3) and (3) fault simulation modeling: the energy storage failure of the spring is simulated by reducing the contact exciting force, and the relation between the spring energy storage and the contact exciting force is converted according to the rigidity and the deformation of the spring.
In addition, a uniform test design method is adopted to provide training samples for the neural network response surface model, the uniform design samples are generated by adopting a good lattice point method, the discreteness of the samples is evaluated by adopting the divergence, the relation of the divergence along with the test times of the samples is established, the selection uniformity is good, but the design samples with fewer test times are used as the training samples of the neural network response surface model, and the accuracy and the construction efficiency of the neural network response surface model can be simultaneously considered. Based on a neural network response surface model, solving a plurality of recognition problems of the spring faults with insufficient energy storage by adopting a willingness function, wherein the willingness function is mainly used for the situation that the quality characteristic targets are the hope characteristics when a plurality of targets are recognized.
Example 1:
firstly, realizing simulation modeling of a tap switch spring energy storage deficiency fault based on a finite element method; then, training sample points of a neural network response surface model are generated by adopting a uniform design test method, and input parameters of the sample points are substituted into a finite element model to perform fault simulation, so that output fault characteristics are obtained; secondly, establishing a neural network response surface model through regression analysis of sample input/output characteristics; and finally, taking the fault characteristic of insufficient energy storage of the tap changer spring as a reference, and adopting a fault identification algorithm based on a willingness function to identify the input fault parameters so as to realize diagnosis of the fault of insufficient energy storage of the tap changer spring.
The specific flow of the on-load tap-changer spring energy storage shortage fault identification method based on the neural network response surface is shown in fig. 1, and the method mainly comprises the following steps:
(1) And carrying out dynamic analysis on the tap switch, taking key parameters affecting the fault characteristics of insufficient energy storage of the tap switch spring as design variables, determining the value range of the key parameters, constructing a design space, adopting a uniform design method to generate proper sample points, and taking the proper sample points as input variables for constructing neural network response surface model training sample points.
(2) Substituting the design variable combination of the sample points into a tap switch simulation model to perform fault simulation, obtaining the system dynamic response corresponding to the sample input, and taking the system dynamic response as the output variable of the training sample points for constructing the neural network response surface model.
(3) And selecting a BP neural network as a construction function of the response surface model, selecting a proper hidden layer number, taking the input variable and the corresponding system dynamic response thereof as training samples of the neural network model, and constructing a response surface model of the input characteristic parameters and the output response characteristics, thereby obtaining the corresponding relation between the tap changer spring energy storage shortage design parameters and the dynamic response.
(4) And selecting a small number of sample points as verification samples, substituting the verification samples into a simulation model, comparing errors of the output response of the actual structure and the prediction result of the response surface model, and checking the accuracy of the response surface model. And if the accuracy requirement is not met, correcting the response surface model by increasing the sample test times.
(5) Based on a neural network response surface model, selecting tap switch spring energy storage deficiency fault characteristic response as a reference, and constructing a willingness function of a multi-target recognition problem aiming at the recognition problem of a plurality of spring energy storage deficiency faults.
(6) And identifying characteristic parameters of the failure of the energy storage deficiency of the split switch spring by adopting a multi-target identification algorithm based on a willingness function, and substituting the identified characteristic parameters into a finite element model for simulation verification.
The 4 steps for obtaining the input variable of the response surface model in the step (1) are respectively
(1) Establishing a simulation model for simulating the fault of insufficient energy storage of the tap changer spring by adopting a finite element method, and solving fault characteristic response by fault input parameters based on the model;
(2) solving the sensitivity of the output characteristic response to the fault input parameters by a sensitivity analysis method, and extracting the characteristic response with larger influence on the input parameters as an output value of the fitting response surface model;
(3) converting the energy storage of the movable contact spring into contact exciting force according to an elastic potential energy formula, determining the variation range of the exciting force by the maximum energy storage, combining the exciting forces of a plurality of springs to form a sample point, and constructing a design space of input parameters according to the variation range of the exciting force;
(4) in the design space, a uniform design table is generated by adopting a good lattice point method, the input parameters of the design table correspond to the combination of excitation force, the divergence is adopted as an index for measuring the uniformity of a test sample, and the divergence is expressed as
Wherein C is k Unit cube for k-dimensional European space, P= { x i I=1, l, n } represents C k ＝[0,1] k A set of points, d (x, x i ) The euclidean distance in euclidean space is represented.
Comprehensively considering the generation efficiency and uniformity index of the sample, selecting reasonable test times, and finally generating an input variable of the fitting response surface model;
the finite element modeling for simulating the tap switch spring energy storage deficiency fault in the step (2) mainly comprises three parts of finite element modeling of single parts of the tap switch, finite element modeling of an assembly structure and modeling of the spring energy storage deficiency fault, wherein the finite element modeling of the single parts comprises the steps of geometric model cleaning, initial model analysis, grid size iterative analysis, model convergence analysis, convergence model determination and the like, and the specific steps are as follows:
(1) the method comprises the steps of cleaning a geometric model, checking and cleaning a component geometric model, eliminating errors generated in the geometric modeling process, compatibility between geometric modeling software (such as UG) and finite element software (such as ANSYS) and the like, ensuring that the geometric model is consistent with design expectations, and carrying out finite element mesh division;
(2) initial model analysis, namely dividing a finite element grid by adopting uniform grid, preliminarily giving a thicker grid size, dividing the finite element grid by using a second-order tetrahedron unit, and carrying out modal analysis under the condition of free-free boundary in a concerned frequency range;
(3) iterative analysis of grid size, namely gradually refining the grid size, carrying out correlation analysis on a model of the grid size of a previous reference model and a current refined grid model, matching mode pairs with the same vibration mode, and analyzing the frequency difference of the mode pairs. Correlation analysis is typically performed using modality confidence criteria (Modal Assurance Criterion, MAC), defined as
In phi r,i Is the ith order vibration vector of the reference model, phi a,j For the j-th order mode vector of the analytical model, T represents the transpose. Mac=1, which indicates that the reference mode shape is completely correlated with the analysis mode shape, mac=0, which indicates that the reference mode shape is uncorrelated, and the closer the value of MAC is to 1, the better the correlation between the two.
(4) Model convergence analysis, namely checking the convergence of the modal analysis result according to the convergence index of the model, if the model is not converged, continuously refining the grid, otherwise, stopping; the average frequency error eta and the maximum frequency error delta of the current calculation model are respectively expressed as
δ＝max|f i,j -f i+1,j |,j＝1,2,...,n (9)
Wherein f i,j The j-th order modal frequency of the model representing the i-th iterative calculation, n being the order of the modes.
(5) Determining a convergence model, determining proper grid size according to the average frequency error and the maximum frequency error of the accuracy index of the convergence model, re-dividing the grids, carrying out correlation analysis with the last iteration model, determining the convergence model after meeting the index requirement,
finite element modeling of tap changer assembly structure: modeling of the connecting piece is realized by adopting a connection relation between the one-dimensional spring unit and the thin-layer unit simulation component, wherein the one-dimensional spring unit mainly considers the rigidity of an actual spring structure and simulates a point-point connection relation; the lamellar unit is mainly used for simulating surface-surface contact connection such as pin connection and bolt connection, and lamellar unit models of two connection surfaces are established by adopting isotropic materials with linear constitutive relation, wherein constitutive equation is that
Wherein D is the elastic matrix of the material, σ is the stress matrix, E, G and u are young's modulus, shear modulus and poisson's ratio, respectively, and the isotropic material satisfies the formula g=e/2 (1+μ), and only two independent variables are found in the material parameters. For thin layer units, it can be considered that the thickness is much smaller than the feature sizes in the other two directionsN i (i=1, 2, L, 8) is a plane 8-node isoparametric function, and ε is known from the relationship between cell strain and cell node displacement x ＝ε y ＝γ xy Approximately 0, in which case the internal strain component (. Epsilon.) of the two characteristic dimension directions of the lamellar unit x ,ε y ,γ xy ) And internal stress component (sigma) x ,σ y ,τ xy ) Will be ignored. The contact surface normal and the two tangential directions are respectively defined as the z, x and y directions of the overall coordinates of the thin layer unit, and the constitutive equation of the material is degenerated as follows:
where λ is the lame constant, λ=g (2G-E)/(E-3G), where the normal and tangential elastic constants are non-independent, which together determine the joint stiffness of the component contact surface.
And under the assembly coordinate system, carrying out connection modeling on the sub-assembly connection model according to the assembly relation to complete the whole assembly modeling of the tapping switch, and applying boundary conditions according to the actual installation condition of the tapping switch.
Modeling of spring energy storage shortage faults: the shortage of spring energy storage is represented by the reduction of the excitation force of the moving contact, the deformation of the spring is converted into the excitation force of the contact according to an elastic deformation potential energy formula, and the excitation force is applied to the contact positions of the moving contact and the fixed contact.
The construction method of the neural network response surface model of the tap changer spring energy storage shortage fault in the step (3) comprises the following steps: and establishing a mapping relation between the input characteristics and the output response of the tap changer spring energy storage deficiency fault by adopting a BP neural network model, and constructing a neural network response surface model, wherein the topological structure of the BP neural network is shown in figure 2. Firstly, sampling a neural network model by adopting a uniform design method to provide training samples, assuming that n springs in m moving contact energy storage springs in a tap switch are insufficient in energy storage, and m is larger than or equal to n, the number of design variables involved in the uniform design is n, the number of units of an input layer of the neural network model is n, the variation range of the design variables is defined according to fault characteristics, the uniform test design is carried out by a good lattice method, the efficiency of a simulation test and the precision of a response surface model are comprehensively considered, the simulation test is carried out by selecting reasonable test times, and fig. 3 shows the rule that the test times in an example change along with the divergence of a sample design point, and the divergence gradually decreases and gradually tends to converge along with the increase of the test times. The fault input characteristics correspond to mechanical parameters of insufficient energy storage of the spring, namely contact excitation force, and the output response is a dynamic response characteristic which is sensitive to the input mechanical parameters. The fitting rule of the neural network response surface model uses a gradient descent method, and the weight and the threshold value of the network are continuously adjusted through back propagation, so that the error square sum of the actual output value and the expected output value of the network is minimum.
The decision coefficient R is selected in the step (4) 2 And the root mean square error RMSE is used as an evaluation index of the neural network response surface model precision to determine a coefficient R 2 Expressed as
In the method, in the process of the invention,a predicted value for the i-th sample; y is i (i=1, 2, l, n) is the true value of the i-th sample; n is the number of samples. The determination coefficient ranges from [0,1]In, the closer to 1, the better the fitting accuracy of the model is shown; conversely, the closer to 0, the worse the fitting accuracy
In the middle ofThe smaller the root mean square error is, the better the prediction accuracy of the model is indicated.
The mathematical form of constructing the multi-objective optimization problem for identifying a plurality of fault characteristics in the step (5) is that
Wherein F (x) is a vector objective function, F i (x) G as the ith objective function j (x) For the kth inequality constraint, h k (x) For the kth equality constraint, m, n, j, k are the number of objective functions, design variables, inequality constraints, equality constraints, and D is the feasible domain of the design variables, respectively.
The willingness function in the step (6) is mainly used for carrying out willingness function conversion on the response characteristic of the hopeful characteristic aiming at the situation that the response quality characteristic is the hopeful characteristic, and the form is that
Wherein L is i ，U i The lower limit and the upper limit of the response y are respectively T i Is a target value for response y.
As shown in FIG. 4, when the response characteristic is eye-looking, the response y can be close to the target value from double directions, and the closer the response y is to the target value, the maximum will is, and 0 is less than or equal to d i ≤1。
By converting the single willingness function, an optimized model of the composite willingness function is established, a calculation expression of the optimized model is shown as a formula (16), and a weighted geometric average value of the model is calculated so as to tend to be maximized, so that the recognition problem of the multi-target response is converted into the problem of the single target response.
Wherein omega is i (i=1, 2, l, n) represents the weight of the willingness function of the ith response; Σω i =1 is the sum of the response quality characteristic weights, usually ω 1 ＝ω 2 ＝L＝ω n 。
And randomly extracting fault samples with insufficient energy storage of the two groups of switch springs, taking fault response characteristics as a reference, identifying exciting force caused by the insufficient energy storage of the springs based on a neural network response surface model by adopting a multi-target identification algorithm based on a willingness function, and substituting the identified fault exciting force into a finite element model to perform fault simulation to obtain a response characteristic curve. Fig. 5 and 6 show the vibration displacement response curves of the identified exciting force and the reference exciting force, and it can be seen that the vibration response curves generated by the identified fault exciting force are identical to those of the reference exciting force, and besides the peak responses at the resonance frequencies are identical, the responses at the other frequencies are basically identical, which indicates that the identified fault parameters are reasonable, can reflect the change of the actual fault response characteristics, and also indicates that the established neural network response surface model has high prediction precision, and can obtain more satisfactory results in combination with a multi-objective identification algorithm based on a willingness function, thereby verifying the rationality of the method.
The beneficial effects of the invention are as follows: the invention opens up a new research path for the identification of the fault of insufficient energy storage of the spring of the on-load tap-changer, provides a beneficial reference for the identification and diagnosis of other typical mechanical faults of the tap-changer, and has wide application prospect.
Claims (3)
1. The on-load tap-changer spring energy storage shortage fault identification method based on the neural network response surface is characterized by comprising the following steps:
(1) Based on a finite element method, establishing an on-load tap-changer fault simulation model by using a solid finite element grid;
(2) Generating training samples of a neural network response surface model by adopting a uniform design method, and obtaining output fault characteristics through simulation of the fault of insufficient energy storage of a tap changer spring;
(3) Constructing a neural network response surface model through regression analysis of input parameters and output characteristics of the tap changer spring energy storage deficiency fault;
(4) Taking the output characteristics of the failure caused by insufficient energy storage of the spring as a reference, and adopting a multi-objective recognition algorithm based on a willingness function to recognize the output failure characteristics;
the step (1) is specifically as follows:
(1) finite element modeling of individual components: in the iterative analysis of the sub-step grid size, the mode confidence criterion MAC is adopted to evaluate the correlation of the vibration modes of the current model and the reference model, and the MAC is defined as
In phi r,i Is the ith order vibration vector of the reference model, phi a,j For analyzing the j-th order mode vector of the modelT represents a transpose;
in the sub-step model convergence analysis, the average frequency error and the maximum frequency error of the current model and the reference model are used for measuring the convergence of the grid, and the average frequency error eta and the maximum frequency error delta are respectively expressed as
δ＝max|f i,j -f i+1,j |,j＝1,2,...,n；
Wherein f i,j The j-th order modal frequency of the model representing the i-th iterative calculation is represented, and n is the order of the modes;
(2) finite element modeling of assembled structures: modeling of the connecting piece is realized by adopting a connection relation between the one-dimensional spring unit and the thin layer unit simulation component, the assembly modeling of the whole machine model is completed by the connecting piece model according to the assembly relation under an assembly coordinate system, and boundary conditions are applied according to the installation condition of an actual structure;
(3) and (3) fault simulation modeling: the energy storage failure of the spring is simulated by reducing the contact exciting force, and the relation between the spring energy storage and the contact exciting force is converted according to the rigidity and the deformation of the spring.
2. The method for identifying the fault of insufficient energy storage of the spring of the on-load tap-changer based on the response surface of the neural network as claimed in claim 1, wherein the method comprises the following steps:
the step (2) is to adopt a uniform test design method to provide training input samples for the neural network response surface model, wherein the uniform design samples are generated by adopting a good lattice point method, the discreteness of the samples is evaluated by adopting a divergence, the relation of the sample point divergence along with the sample test times is established, a design sample with better uniformity but fewer test times is selected as the training sample of the neural network response surface model, and the output fault characteristics corresponding to the input sample points are obtained by simulating the energy storage shortage faults of the split switch springs.
3. The method for identifying the fault of insufficient energy storage of the spring of the on-load tap-changer based on the response surface of the neural network as claimed in claim 1, wherein the method comprises the following steps:
the step (3) is to establish a mapping relation between the fault input characteristics of the insufficient energy storage of the tap changer spring and the output response by adopting a BP neural network model, and to construct a neural network response surface model, wherein the fault input characteristics correspond to the mechanical parameters of the insufficient energy storage of the spring, namely the contact exciting force, the output response is the dynamic response characteristics sensitive to the comparison of the input mechanical parameters, the fitting rule of the neural network response surface model uses a gradient descent method, the weight and the threshold of the network are continuously adjusted through counter propagation, the error square sum of the actual output value and the expected output value of the network is minimum, and the decision coefficient R is selected 2 And the Root Mean Square Error (RMSE) is used as an evaluation index of the neural network response surface model precision.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910616899.2A CN110457750B (en) | 2019-07-09 | 2019-07-09 | On-load tap-changer spring energy storage deficiency fault identification method based on neural network response surface |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910616899.2A CN110457750B (en) | 2019-07-09 | 2019-07-09 | On-load tap-changer spring energy storage deficiency fault identification method based on neural network response surface |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110457750A CN110457750A (en) | 2019-11-15 |
| CN110457750B true CN110457750B (en) | 2023-10-24 |
Family
ID=68482524
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910616899.2A Active CN110457750B (en) | 2019-07-09 | 2019-07-09 | On-load tap-changer spring energy storage deficiency fault identification method based on neural network response surface |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110457750B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110909503B (en) * | 2019-11-25 | 2023-08-18 | 中国船舶重工集团公司第七一九研究所 | Prediction method for flange leakage of pipeline system |
| CN111859676B (en) * | 2020-07-23 | 2022-11-22 | 西南交通大学 | Intelligent detection method for damage of floating slab track steel spring |
| CN113011055B (en) * | 2021-02-03 | 2023-01-03 | 中国船级社 | Motor high-temperature fault simulation method and system based on finite element technology |
| CN113779820A (en) * | 2021-07-26 | 2021-12-10 | 国网辽宁省电力有限公司电力科学研究院 | On-load tap-changer fault identification method based on neural network response surface model |
| CN114636900B (en) * | 2022-05-11 | 2022-09-30 | 广东电网有限责任公司东莞供电局 | Power distribution network multiple fault diagnosis method |
| CN115238560B (en) * | 2022-09-21 | 2023-01-20 | 广东电网有限责任公司佛山供电局 | Transformer on-load switch fault identification method and system based on finite element simulation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104200004A (en) * | 2014-07-28 | 2014-12-10 | 东北大学 | Optimized bridge damage identification method based on neural network |
| WO2017024583A1 (en) * | 2015-08-13 | 2017-02-16 | 华为技术有限公司 | Method and apparatus for model predictive control |
| CN108490345A (en) * | 2018-03-26 | 2018-09-04 | 盐城博鸣信息科技有限公司 | A kind of fault simulation diagnostic method of converter power transformer tap switch |
| CN108920776A (en) * | 2018-06-11 | 2018-11-30 | 国网宁夏电力有限公司检修公司 | A kind of load ratio bridging switch kinetic model layering confirmation method based on hyper-model |
-
2019
- 2019-07-09 CN CN201910616899.2A patent/CN110457750B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104200004A (en) * | 2014-07-28 | 2014-12-10 | 东北大学 | Optimized bridge damage identification method based on neural network |
| WO2017024583A1 (en) * | 2015-08-13 | 2017-02-16 | 华为技术有限公司 | Method and apparatus for model predictive control |
| CN108490345A (en) * | 2018-03-26 | 2018-09-04 | 盐城博鸣信息科技有限公司 | A kind of fault simulation diagnostic method of converter power transformer tap switch |
| CN108920776A (en) * | 2018-06-11 | 2018-11-30 | 国网宁夏电力有限公司检修公司 | A kind of load ratio bridging switch kinetic model layering confirmation method based on hyper-model |
Non-Patent Citations (2)
| Title |
|---|
| AGA-BP神经网络的变压器分接开关机械故障诊断;王福忠等;《现代制造工程》;20170418(第04期);第11页-第15页 * |
| 基于振动信号的有载分接开关故障模拟试验研究;张静等;《高压电器》;20181016(第10期);第103-第111页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110457750A (en) | 2019-11-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110457750B (en) | On-load tap-changer spring energy storage deficiency fault identification method based on neural network response surface | |
| CN105608263B (en) | A kind of adaptive processing method towards turbine blade structural life-time probability analysis | |
| CN109659933A (en) | A kind of prediction technique of power quality containing distributed power distribution network based on deep learning model | |
| Fei et al. | Vectorial surrogate modeling method for multi-objective reliability design | |
| CN110880062B (en) | Method for determining state maintenance time of power distribution equipment | |
| CN110443417A (en) | Multi-model integrated load prediction method based on wavelet transformation | |
| CN110543932A (en) | air conditioner performance prediction method and device based on neural network | |
| CN116305522A (en) | Digital twin-based aircraft structure reliability simulation test model calibration method, device and medium | |
| CN112084709B (en) | Large-scale generator insulation state evaluation method based on genetic algorithm and radial basis function neural network | |
| CN113609700A (en) | Simulation method for completely non-stable wind field of mountainous bridge | |
| CN114266396A (en) | Transient stability discrimination method based on intelligent screening of power grid characteristics | |
| CN108920776A (en) | A kind of load ratio bridging switch kinetic model layering confirmation method based on hyper-model | |
| US20050234839A1 (en) | Neural network for aeroelastic analysis | |
| Kül et al. | Coreloss Estimation via Long Short-Term Memory Model (LSTM) of Dry-Type Transformer based on FEA | |
| CN110825583B (en) | Energy efficiency qualitative assessment technology for multi-index fusion of cloud data center | |
| CN115762653B (en) | Fuel combustion mechanism optimization method based on evolutionary algorithm and deep learning | |
| Li et al. | Predicting Software Quality by Optimized BP Network Based on PSO. | |
| CN108959721B (en) | Method for calculating reliability of long-span bridge member under action of windmill load | |
| CN109886316B (en) | Transformer state parameter combination prediction method based on cloud system similarity weight distribution | |
| Zhou et al. | Uncertainty propagation in structural reliability with implicit limit state functions under aleatory and epistemic uncertainties | |
| CN107292045B (en) | Complex modal evaluation method of finite element model with damping structure | |
| Romanska-Zapala et al. | Can artificial neuron networks be used for control of HVAC in environmental quality management systems? | |
| Zhao et al. | A fast and accurate transient stability assessment method based on deep learning: Wecc case study | |
| CN115048857A (en) | CNN-based transient stability limit transmission power determination method | |
| CN114491730A (en) | Dynamic stability analysis iteration method and device for high-speed railway roadbed structure |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |