CN112307963A - Converter transformer running state identification method based on vibration signals - Google Patents

Converter transformer running state identification method based on vibration signals Download PDF

Info

Publication number
CN112307963A
CN112307963A CN202011190818.6A CN202011190818A CN112307963A CN 112307963 A CN112307963 A CN 112307963A CN 202011190818 A CN202011190818 A CN 202011190818A CN 112307963 A CN112307963 A CN 112307963A
Authority
CN
China
Prior art keywords
converter transformer
signal
learning machine
modal
extreme learning
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011190818.6A
Other languages
Chinese (zh)
Inventor
邓军
潘志城
楚金伟
张晋寅
周海滨
谢志成
孙勇
张占龙
武雍烨
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.)
Maintenance and Test Center of Extra High Voltage Power Transmission Co
Original Assignee
Maintenance and Test Center of Extra High Voltage Power Transmission Co
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 Maintenance and Test Center of Extra High Voltage Power Transmission Co filed Critical Maintenance and Test Center of Extra High Voltage Power Transmission Co
Priority to CN202011190818.6A priority Critical patent/CN112307963A/en
Publication of CN112307963A publication Critical patent/CN112307963A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Signal Processing (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention discloses a converter transformer running state identification method based on vibration signals, which is used for identifying the running state of a converter transformer. The method comprises the steps of collecting vibration signals of the surface box body of the converter transformer, selecting a variational modal decomposition algorithm by a signal characteristic extraction method, completing parameter optimization work of the variational modal decomposition algorithm through a longicorn beard search algorithm, determining the optimal parameters of decomposition, performing modal decomposition reconstruction on the vibration signals of the surface box body of the converter transformer obtained through measurement by using the optimized variational modal decomposition algorithm, calculating the energy ratio of each reconstructed signal after decomposition in the original signal, and constructing a corresponding state characteristic vector. The method comprises the steps of constructing a self-encoder on the basis of an extreme learning machine, establishing a deep extreme learning machine network model by taking the self-encoder as a basic unit, dividing a state feature vector into a training set and a testing set, importing the training set and the testing set into the deep extreme learning machine network model, and effectively identifying the running state of the converter transformer through model training and testing.

Description

Converter transformer running state identification method based on vibration signals
Technical Field
The invention relates to the field of power systems, in particular to a method for identifying the running state of a converter transformer based on a vibration signal.
Background
The converter transformer is core equipment for ensuring stable operation of the converter station, is an important component in ultrahigh voltage direct current transmission engineering, and the operation of whether the converter transformer can be safely and stably directly relates to the operation reliability of the whole direct current transmission system. When the converter transformer has a fault, the converter valve is blocked, so that local and even large-scale power failure of a power system is caused, and great loss is caused to the development of social economy. At present, a plurality of direct current converter stations are finished and put into operation at home and abroad, and according to CIGRE statistical results, about 60% of faults generated by a direct current transmission system are caused by faults of a converter transformer, wherein mechanical faults caused by windings, iron cores and tap switches in the converter transformer account for about 50% of the faults of the converter transformer. According to the data statistics result of the CIGRE, the failure rate of the converter transformer is continuously improved every year, wherein most of the failure rate is caused by the abnormality of the mechanical structure of the converter transformer, so that the method is very important for identifying the operation state of the mechanical parts in the converter transformer, and has important significance for improving the state maintenance level of direct-current transmission equipment and ensuring the safe and reliable operation of a direct-current transmission system.
At present, the state monitoring methods for the converter transformer are few, and if the running state of mechanical components in the converter transformer cannot be identified in time, potential safety hazards in the converter transformer cannot be found easily, so that the safety, stability and fixed running of a direct current transmission system are threatened. Most of the existing converter transformer state identification means are based on electrical signals, and the converter transformer is required to be overhauled only by power failure, so that the detection cannot be carried out under the working condition of live operation of equipment, and the conventional planned overhaul and physical and financial resources are consumed, so that the service life of the equipment is influenced while the power failure loss is caused.
The vibration analysis method is a state identification method capable of being realized outside a transformer body, and can acquire vibration signals of the converter transformer in the operation process at different time and different positions by using a sensor, extract time domain and frequency domain characteristic information of the vibration signals, identify the operation state of the transformer and timely find latent defects causing abnormal vibration of the transformer. However, in the existing stage of transformer vibration technology research, the alternating current transformer is mainly focused on, the vibration of the converter transformer is obviously different from the vibration characteristics of the common alternating current transformer, the signal characteristics are more complex and changeable, the identification accuracy of the signal analysis method adopted by the traditional vibration analysis method is low, the current requirement on the state identification of the converter transformer is difficult to meet, and a new method must be introduced to realize the state identification of the converter transformer based on the vibration signals.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a converter transformer running state identification method based on a vibration signal, which is used for identifying the running state of a converter transformer.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a method for identifying the running state of a converter transformer based on vibration signals is realized by adopting the following steps:
collecting a vibration acceleration signal of the converter transformer;
dividing the vibration acceleration signal into a training sample set and a test sample set; wherein the training sample set is used for seeking the optimal set number of modal functions and a penalty factor;
according to a variational modal decomposition algorithm, establishing a variational model with constraint conditions by using the optimal set number of the modal functions and the penalty factors;
extracting energy characteristics of the variation model with the constraint condition to obtain a vibration sample signal characteristic set and a test sample signal characteristic set, and training the vibration sample signal characteristic set through a depth limit learning machine to obtain a trained model;
and inputting the test sample signal characteristic set into the trained model to obtain a classification result of the vibration acceleration signal, and identifying the running state of the converter transformer according to the classification result.
The method for identifying the operational status of the converter transformer based on the vibration signal as claimed in claim 1, wherein the method for seeking the optimal set number of the modal functions and the penalty factor by using the training sample set is as follows:
establishing an objective function model;
initializing an optimizing initial position in a parameter space;
converting the position of the parameter space into the set number and the penalty factor of a preset modal function;
processing according to a variational modal decomposition algorithm by using the set number of preset modal functions and a penalty factor;
calculating objective function values of all training sample sets;
judging whether the objective function value reaches the maximum value; if so, taking the set number and the penalty factor of the preset modal function as the set number and the penalty factor of the optimal modal function, otherwise, updating the position of the next parameter space by using a longicorn searching algorithm, converting the position of the updated parameter space into the set number and the penalty factor of the preset modal function, and repeating the steps until the set number and the penalty factor of the optimal modal function are obtained.
The method for identifying the operating state of the converter transformer based on the vibration signal further acquires the vibration acceleration signal of the converter transformer, and specifically includes the following steps:
mounting a multi-channel vibration sensor on the surface of a converter transformer box body; and transmitting the vibration acceleration signals output by the surface of the converter transformer box body to a computer through an integrated measuring device, and carrying out state recognition on the collected vibration acceleration signals in the computer.
The method for identifying the operation state of the converter transformer based on the vibration signal further includes the following steps:
defining each sub-signal as a mode function Ui(t), the original signal may be represented as:
Figure BDA0002752691130000031
wherein K denotes the number of modes, Ai(t) denotes the instantaneous amplitude value,
Figure BDA0002752691130000032
indicating phase, instantaneous frequency
Figure BDA0002752691130000033
Variational model with constraint conditions:
Figure BDA0002752691130000034
Figure BDA0002752691130000035
wherein: [ U ]i]Represented is a set of K modal functions, [ omega ] omegai]A set of corresponding center frequencies is represented,
Figure BDA00027526911300000318
denoted by gradient operation, R (t) isThe real part of the modal function;
in order to obtain the optimal solution of the model, a Lagrange multiplier lambda and a penalty factor alpha are adopted aiming at the constraint variation problem,
it is changed to an unconstrained case and an augmented lagrange function is established:
Figure BDA0002752691130000036
the saddle point of the above formula is solved by an alternative direction multiplier method, and the mode component in the frequency domain can be obtained by converting the time domain signal into the frequency domain signal by Fourier transformation
Figure BDA0002752691130000037
And center frequency
Figure BDA0002752691130000038
The expression of (1);
Figure BDA0002752691130000039
Figure BDA00027526911300000310
iterative updating in operation process
Figure BDA00027526911300000311
ωiAnd
Figure BDA00027526911300000312
the optimal solution can be obtained by solving, and the specific flow is as follows:
1) initialization
Figure BDA00027526911300000313
And
Figure BDA00027526911300000314
λ1＝n＝0；
2) iterative update based on equation (4)
Figure BDA00027526911300000315
And ωi；
3) Updating Lagrange multiplication operator on the basis of formula (6)
Figure BDA00027526911300000316
Wherein τ represents the update parameter of the Lagrangian multiplier;
Figure BDA00027526911300000317
4) and (3) repeating the steps (1) and (3) until the constraint condition that the operation is stopped is taken as an equation (7), wherein epsilon represents the iteration precision, and the result obtained after the iteration is finished is the modal component and the center frequency of the variation modal decomposition.
Figure BDA0002752691130000041
The method for identifying the operating state of the converter transformer based on the vibration signal further includes the following steps:
the position of the longicorn can be defined as sn(N ═ 1,2, ·, N), where N denotes the number of iterations of the search, and the food smell is constituted by the location-dependent objective function f(s), the extreme of which is the final location of the food;
the global optimization of the algorithm in the multidimensional data space can be represented by the following formula:
Figure BDA0002752691130000042
the global optimization summary of the algorithm is exploration behavior and traveling behavior, wherein the orientation of the longicorn whiskers can be represented by a normalized random vector D
Figure BDA0002752691130000043
In formula (9), Random is a Random function; d is the dimension of the data space. And the left and right palpus positions slAnd srThen can be defined as
Figure BDA0002752691130000044
In the formula, anThe length of the whisker; the advancing behavior is to determine the stepping length and the direction of the longicorn according to the magnitude relation of the objective function values corresponding to the left and right tentacle positions, so as to update the position of the longicorn; its update location can be expressed as
sn＝sn-1+ξnDsign[f(xr)-f(xl)] (11)
Xi in formula (11)nSign is a sign function for the step length of each iteration; each time the exploration behavior and the advancing behavior are executed, an iteration process of the algorithm can be defined, and a is carried out after each iterationnAnd xinNeeds to be updated, the updating formula is
Figure BDA0002752691130000045
Update coefficient c in equation (12)1And c3Set to 0.95, c2Set to 0.01; and after the algorithm reaches the preset iteration times, the data space position corresponding to the optimal objective function value in the past iteration is the optimizing result.
The method for identifying the operating state of the converter transformer based on the vibration signal further includes the following steps:
setting the length of a vibration signal y (t) as L, and if the proportion of the reconstructed signal energy after the variation modal decomposition in the original signal energy is defined as E, the mathematical model can be expressed as
Figure BDA0002752691130000051
In formula (13), Ui(t) represents the reconstructed signal amplitude of the ith mode, and K represents the number of modes;
the central frequency variation range of the modal bandwidth estimate is called b, which can be defined as
Figure BDA0002752691130000052
In the formula, max and min are respectively functions for solving the maximum value and the minimum value; f. ofsIs the signal sampling frequency; omegai(n) is the estimated center frequency of the ith mode in an iterative process; m is the number of estimated center frequencies in the mode;
Figure BDA0002752691130000053
wherein s ═ p, q](p belongs to K, q belongs to alpha) is any parameter combination in the optimizing space; n is the number of qualified training samples; ejAnd bjE and b corresponding to the j training sample vibration signal; 0.5 can control the calculation result to be [0, 100 ]]The weight coefficient of (2).
The method for identifying the operating state of the converter transformer based on the vibration signal further comprises the following steps:
constructing an extreme learning machine:
t different marked samples are obtained and obtained,
Figure BDA0002752691130000054
denotes the ith sample, corresponding to label li＝[liz,liz,···,liz]TWherein z represents the number of classes; if sample
Figure BDA0002752691130000055
If the tag belongs to the m-th class, the m-th value of the corresponding tag vector is set to be 1, and the rest (m-1) values are set to be-1; the extreme learning machine having H hidden nodes can be represented by equation (16):
Figure BDA0002752691130000056
wherein Q isi＝[Qi1,Qi2,···,Qiz]TDenoted is the output weight connecting the ith hidden layer and the output layer, Ai＝[Ai1,Ai2,···,Aiz]TIs an input weight connecting the input layer and the i-th hidden layer, BiIs the bias of the i-th hidden layer, G (A)i,Bi,Ej) Is the output of the ith hidden layer; assuming that the activation function of the ith hidden layer is g (x), the output of the hidden layer is:
G(Ai,Bi,Ej)＝g(Ai·Ej+Bi) (17)
the formula (16) is summarized in a matrix form as
Figure BDA0002752691130000061
Wherein the content of the first and second substances,
Figure BDA0002752691130000062
in order to satisfy the expected value output after network training, the parameter A needs to be obtainedi、Bi、QiOptimum value of (2)
Figure BDA0002752691130000063
Such that:
Figure BDA0002752691130000064
then the minimization loss function is
Figure BDA0002752691130000065
In extreme learning machine, parameter (A)i,Bi) Is randomly initialized, so QiIs uniquely determined; the solution can be converted into:
Figure BDA0002752691130000066
the norm of the solution Q that is sought is minimal and unique; to make the model have better generalization performance, L can be added1Regular terms, the above solution problem can be transformed into equation (21):
Figure BDA0002752691130000067
wherein, CLIs a regularization coefficient; solving is essentially a ridge regression problem whose solution is shown below:
Figure BDA0002752691130000068
wherein I is an identity matrix;
constructing an extreme learning machine-based self-encoder: applying the idea of an automatic encoder to an extreme learning machine, so that the input data is also used for output, namely the output Y is equal to X;
the output of the extreme learning machine-based self-encoder can be converted by equation (16) to:
Figure BDA0002752691130000069
wherein A isTA＝I，BTB＝1；
Wherein A is AiA composed matrix of BiA vector of components; for sparse representation and dimension compressionThe hidden layer output weight Q can be converted by equation (23):
Figure BDA0002752691130000071
wherein E ═ E1,E2,···,E8]The feature vector is formed by the proportion of each modal energy after the variation modal decomposition of the input vibration signal; for the equal-dimensional feature mapping, the weight Q can be calculated by equation (26):
Q＝C-1l (26)
wherein Q isTQ＝I；
Constructing a depth extreme learning machine:
taking the feature representation capability of an extreme learning machine-based self-encoder as a basic unit of a depth extreme learning machine;
taking the input data sample E as the target output of the 1 st self-encoder based on the extreme learning machine, and further solving the output weight Q1(ii) a Then the output matrix C of the 1 st hidden layer of the depth limit learning machine is used1When the input and the target output of the next 1 self-encoder based on the extreme learning machine are considered, the training is carried out by analogy, the last 1 layer is trained by the extreme learning machine, and the output weight Q of the last 1 hidden layer of the deep extreme learning machine is solved by using an equation (21)i+1。
Compared with the prior art, the invention has the beneficial effects that: when the converter transformer is in different running states, the vibration signals of the converter transformer can be changed, but due to the fact that the internal structure of the converter transformer is complex, the obvious characteristics of the converter transformer are difficult to determine when a traditional vibration analysis method is adopted, and therefore accurate classification and recognition cannot be achieved. After the optimized variational modal decomposition algorithm is adopted to decompose the vibration signals of the converter transformer, effective characteristic vectors of the vibration signals can be effectively obtained, and therefore a depth limit learning machine can be utilized to carry out accurate signal classification to determine the running state of the converter transformer, maintenance is facilitated, and the safety and stability of a power system are improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a converter transformer vibration measurement system;
FIG. 2 is a diagram of a multi-channel sensor layout;
FIG. 3 is a block diagram of an extreme learning machine;
FIG. 4 is a view showing the structure of ELM-AE;
FIG. 5 is a diagram of a model of a deep extreme learning machine;
FIG. 6 is a general method flow diagram;
FIG. 7 is a graph of deep extreme learning machine training results;
FIG. 8 is a graph of depth-extreme learning machine test results.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Example (b):
it should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Referring to fig. 1 to 8, fig. 1 is a converter transformer vibration measurement system; FIG. 2 is a diagram of a multi-channel sensor layout; FIG. 3 is a block diagram of an extreme learning machine; FIG. 4 is a view showing the structure of ELM-AE; FIG. 5 is a diagram of a model of a deep extreme learning machine; fig. 6 is an overall method flowchart fig. 7 is a deep extreme learning machine training result fig. 8 is a deep extreme learning machine test result diagram.
The invention aims to provide a converter transformer running state identification method based on vibration signals, which is used for identifying the running state of a converter transformer. The method comprises the steps of measuring and obtaining a vibration signal of a converter transformer surface box body by adopting a multi-channel vibration acceleration sensor through multi-point layout, selecting a variation modal decomposition algorithm by a signal characteristic extraction method, completing parameter optimization work of the variation modal decomposition algorithm through a longicorn stigma search algorithm, determining an optimal parameter of decomposition, carrying out modal decomposition reconstruction on the vibration signal of the converter transformer surface box body obtained through measurement by using the optimized variation modal decomposition algorithm, calculating the energy proportion of each reconstructed signal after decomposition in an original signal, and constructing a corresponding state characteristic vector. The method comprises the steps of constructing a self-encoder on the basis of an extreme learning machine, establishing a deep extreme learning machine network model by taking the self-encoder as a basic unit, dividing a state feature vector into a training set and a testing set, importing the training set and the testing set into the deep extreme learning machine network model, and effectively identifying the running state of the converter transformer through model training and testing.
The invention provides a vibration signal-based converter transformer running state identification method, which is characterized by comprising the following steps of:
(1) measurement and acquisition of vibration signals
The multi-channel vibration sensor is arranged on the surface of the converter transformer box body, the vibration acceleration signal output by the multi-channel vibration sensor is transmitted to a computer through the integrated measuring device, and the whole vibration measuring system is shown in figure 1. The mounting position of the vibration sensor on the surface of the converter transformer tank is shown in fig. 2.
(2) Vibration signal energy feature extraction
Variation modal decomposition algorithm (VMD)
The variational modal decomposition is a new adaptive quasi-orthogonal decomposition technology in recent years. The algorithm can decompose an original signal X (t) containing a plurality of signal components into a limited number of sub-signals, wherein the sub-signals have special sparsity and can realize amplitude modulation and frequency modulation, and each sub-signal is defined as a mode function Ui(t), the sparsity of which is generally determined by the bandwidth of the frequency domain, the original signal can be represented as:
Figure BDA0002752691130000091
where K denotes the number of modes, i.e. the decomposition scale of the signal, Ai(t) denotes the instantaneous amplitude value,
Figure BDA0002752691130000092
phase is indicated. Instantaneous frequency
Figure BDA0002752691130000093
And the change rate of the instantaneous frequency and the instantaneous amplitude is slower compared with the phase, so the mode function Ui(t) at [ t-delta, t + delta]Can be regarded as a harmonic signal and its instantaneous amplitude is Ai(t) center frequency is ωi(t)。
The essence of the variational modal decomposition algorithm is that a variational model with constraint conditions is established, and the optimal solution of the model is obtained under the constraint conditions of modal estimation bandwidth and minimum constraint conditions, so that each modal component and the corresponding central frequency thereof are determined. In order to effectively realize the estimation of modal bandwidth, firstly, hilbert transform is carried out on each modal component to obtain a single-sided frequency spectrum, an exponential correction method is used to modulate the center frequency obtained by modal estimation to a fundamental frequency band, a gaussian smoothing method is used to obtain the bandwidth of each band, so that the whole process is defined as a variation problem with bandwidth constraint, and then the mathematical model is as follows:
Figure BDA0002752691130000094
wherein: [ U ]i]Represented is a set of K modal functions, [ omega ] omegai]A set of corresponding center frequencies is represented,
Figure BDA0002752691130000095
gradient operations are shown, and R (t) represents the real part of the mode function.
In order to obtain the optimal solution of the model, a Lagrangian multiplier lambda and a penalty factor alpha are adopted for the constraint variation problem, the constraint variation problem is changed into an unconstrained condition, and an augmented Lagrangian function is established.
Figure BDA0002752691130000096
The saddle point of the above formula is solved by an alternative direction multiplier method, and the mode component in the frequency domain can be obtained by converting the time domain signal into the frequency domain signal by Fourier transformation
Figure BDA0002752691130000101
And center frequency
Figure BDA0002752691130000102
Is described in (1).
Figure BDA0002752691130000103
Figure BDA0002752691130000104
Iterative updating in operation process
Figure BDA0002752691130000105
ωiAnd
Figure BDA0002752691130000106
the optimal solution can be obtained by solving, and the specific flow is as follows:
1) initialization
Figure BDA0002752691130000107
And
Figure BDA0002752691130000108
λ1＝n＝0；
2) iterative update based on equation (4)
Figure BDA0002752691130000109
And ωi；
3) Updating Lagrange multiplication operator on the basis of formula (6)
Figure BDA00027526911300001010
Wherein τ represents the update parameter of the Lagrangian multiplier;
Figure BDA00027526911300001011
4) and (3) repeating the steps (1) and (3) until the constraint condition that the operation is stopped is taken as an equation (7), wherein epsilon represents the iteration precision, and the result obtained after the iteration is finished is the modal component and the center frequency of the variation modal decomposition.
Figure BDA00027526911300001012
As can be seen from the principle of the variational modal decomposition, a plurality of parameters need to be preset before decomposition operation decomposition is performed. Wherein tau and epsilon mainly determine the iteration number and the precision, and K and alpha determine the decomposition effect. The optimal [ K, alpha ] needs to be screened for signal analysis. Therefore, the optimization is carried out by introducing a longicorn whisker search algorithm.
Optimization of decomposition algorithm
The principle of the longicorn whisker search algorithm simulates the first left and right equidistant tentacles to randomly explore the odor intensity of nearby food when the longicorn is foraging, then the food is sent to the direction of the tentacle with stronger food odor, and the food position is determined through circulation until the odor is the strongest point.
In constructing a mathematical model of the algorithm, the position of the longicorn can be defined as sn(N ═ 1,2, ·, N), where N denotes the number of iterations of the search, and the food smell is constituted by the location-dependent objective function f(s), the extreme of which is the final location of the food. The global optimization of the algorithm in the multidimensional data space can be represented by the following formula:
Figure BDA00027526911300001013
the global optimization of the algorithm is summarized as exploration behavior and traveling behavior. The orientation of the long-horned beard in the former can be represented by a normalized random vector D
Figure BDA0002752691130000111
In formula (9), Random is a Random function; d is the dimension of the data space. And the left and right palpus positions slAnd srThen can be defined as
Figure BDA0002752691130000112
In the formula, anThe whisker length. The advancing behavior is to determine the stepping length and the direction of the longicorn according to the magnitude relation of the objective function values corresponding to the left and right tentacle positions, so as to update the position of the longicorn. Its update location can be expressed as
sn＝sn-1+ξnDsign[f(xr)-f(xl)] (11)
Xi in formula (11)nSign is a sign function for the step length of each iteration. Each time holdingThe line exploration behavior and the traveling behavior can be defined as an iterative process of the algorithm, and a is carried out after each iterationnAnd xinNeeds to be updated, the updating formula is
Figure BDA0002752691130000113
Update coefficient c in equation (12)1And c3Is generally set to 0.95, c2It is generally set to 0.01. And after the algorithm reaches the preset iteration times, the data space position corresponding to the optimal objective function value in the past iteration is the optimizing result.
Construction of feature vectors
The target function used by the optimized longicorn whisker search algorithm needs to be capable of fully embodying the characteristics of the transformer flow vibration signal under the variation modal decomposition, and the extreme value of the target function also needs to represent the optimal average decomposition effect. According to theory, the metamorphic modal decomposition can divide the signal into several components existing in different bandwidths, so that the signal energy is naturally distributed into modes one by one, and each modal bandwidth can determine the estimation effect of the center frequency. Therefore, the variation of the center frequency of the modal energy and bandwidth estimation can represent the decomposition performance of the variation modal decomposition, so that the objective function is designed based on the principle.
Firstly, setting a vibration signal y (t) with length L, and if the proportion of the reconstructed signal energy after the variation modal decomposition in the original signal energy is defined as E, then the mathematical model can be expressed as
Figure BDA0002752691130000114
In formula (13), Ui(t) denotes the reconstructed signal amplitude of the ith mode, and K denotes the number of modes. The larger E is, the decomposed signal is closer to the original signal, and the original signal information is more sufficient.
Secondly, each mode has a certain bandwidth, which can determine the center frequency, and the lower the bandwidth span, the smaller the fluctuation of the center frequency in the iterative process, the stronger the signal resolution and the more sufficient the decomposition, so the central frequency variation of the mode bandwidth estimation is called b, which can be defined as the central frequency variation of the mode bandwidth estimation
Figure BDA0002752691130000121
In the formula, max and min are respectively functions for solving the maximum value and the minimum value; f. ofsIs the signal sampling frequency; omegai(n) is the estimated center frequency of the ith mode in an iterative process; m is the number of center frequencies estimated in that mode. A larger b means a smaller percentage of the estimated center frequency fluctuation with respect to the entire signal frequency range, i.e. a narrower bandwidth the decomposition is more refined.
The sum of E and b of each signal represents the balance between the decomposition degree of the signal in the decomposition process of the variation mode and the information content of the signal, and the larger the result is, the more sufficient the decomposition is, the more information content is, and the better the signal processing performance is; and the average result of the sum of all signals E and b in the whole qualified training sample set represents the degree of commonality achieved by the decomposition performance of the whole sample. Each group of parameters corresponds to an average result, the maximum value of the average result in the parameter space reflects the highest variational modal decomposition average processing performance of the whole sample, and the uniform parameter suitable for the variational modal decomposition of each signal is the parameter combination corresponding to the maximum value.
In summary, the objective function F(s) for parameter optimization can be summarized as follows
Figure BDA0002752691130000122
Wherein s ═ p, q](p belongs to K, q belongs to alpha) is any parameter combination in the optimizing space; n is the number of qualified training samples; ejAnd bjE and b corresponding to the j training sample vibration signal; 0.5 can control the calculation result to be [0, 100 ]]The weight coefficient of (2). According to the past historical experience, the frequency spectrum of the vibration signal of the converter transformer is mainly distributed in 0-800 Hz, therefore, the K value is set to be 8, the optimal alpha is calculated and solved according to the objective function,and solving to obtain a characteristic vector E ═ E of each sample1,E2,···,EK](K＝8)。
(3) Classification identification of vibration signals
Basic principle of Extreme Learning Machine (ELM)
The extreme learning machine is a supervised learning method, belonging to a single hidden layer feedforward neural network. The input weight and bias of the extreme learning machine network training are randomly generated, so that the extreme learning machine network training has higher learning speed, input data can be randomly mapped to a Hilbert space, the generalization capability is strong, and the basic network structure is shown in FIG. 3.
(4) Obtaining T different marked samples,
Figure BDA0002752691130000123
denotes the ith sample, corresponding to label li＝[liz,liz,···,liz]TWherein z represents the number of classes. If sample
Figure BDA0002752691130000124
Belonging to class m, the m-th value of the corresponding tag vector is set to 1, and the remaining (m-1) values are set to-1. The extreme learning machine having H hidden nodes can be represented by equation (16):
Figure BDA0002752691130000125
wherein Q isi＝[Qi1,Qi2,···,Qiz]TDenoted is the output weight connecting the ith hidden layer and the output layer, Ai＝[Ai1,Ai2,···,Aiz]TIs an input weight connecting the input layer and the i-th hidden layer, BiIs the bias of the i-th hidden layer, G (A)i,Bi,Ej) Is the output of the ith hidden layer. Assuming that the activation function of the ith hidden layer is g (x), the output of the hidden layer is:
G(Ai,Bi,Ej)＝g(Ai·Ej+Bi) (17)
the formula (16) is summarized in a matrix form as
Figure BDA0002752691130000131
Wherein the content of the first and second substances,
Figure BDA0002752691130000132
in order to satisfy the expected value output after network training, the parameter A needs to be obtainedi、Bi、QiOptimum value of (2)
Figure BDA0002752691130000133
Such that:
Figure BDA0002752691130000134
since the goal of the extreme learning machine is to minimize the error between the actual output and the expected output, the minimization loss function is
Figure BDA0002752691130000135
In extreme learning machine, parameter (A)i,Bi) Is randomly initialized, so QiIs uniquely determined. The solution can be converted into:
Figure BDA0002752691130000136
the norm of the solution Q is found to be minimal and unique. To make the model have better generalization performance, L can be added1Regular terms, the above solution problem can be transformed into equation (21):
Figure BDA0002752691130000137
wherein, CLIs a regularization coefficient. Solving is essentially a ridge regression problem whose solution is shown below:
Figure BDA0002752691130000138
wherein I is an identity matrix.
② self-encoder (ELM-AE) based on extreme learning machine
An Auto Encoder (AE) is trained to copy input to output. Because no marker data is needed, training the autoencoder is unsupervised. Therefore, the idea of the auto-encoder is applied to the extreme learning machine, and the input data is used for output as well, i.e., the output Y is equal to X. The network structure of the self-encoder based on the extreme learning machine is shown in FIG. 4
If w is greater than H in FIG. 2, ELM-AE realizes dimension compression, and high-dimension data is mapped into low-dimension feature expression; if w ═ H, then ELM-AE achieves equal dimensional feature expression; if w is less than H, the ELM-AE realizes sparse expression, namely high-dimensional characteristic expression of the original data. In summary, ELM-AE is a general approximator, and features that the output and input of the network are the same, and the input parameters (A) of the hidden layeri,Bi) And after random generation, the orthogonal is realized.
The output of ELM-AE can be converted by equation (16) to:
Figure BDA0002752691130000141
wherein A isTA＝I，BTB＝1。
Wherein A is AiA composed matrix of BiThe vectors of the components. For sparse representation and dimension compression, the hidden layer output weight Q can be converted by equation (23):
Figure BDA0002752691130000142
wherein E ═ E1,E2,···,E8]And the characteristic vector is formed by the ratio of the modal energy of the input vibration signal after the variation modal decomposition. For the equal-dimensional feature mapping, the weight Q can be calculated by equation (26):
Q＝C-1l (26)
wherein Q isTQ＝I。
③ Deep Extreme Learning Machine (DELM)
And (4) taking the ELM-AE as a basic unit of a depth limit learning machine according to the feature representation capability of the ELM-AE. The depth limit learning machine is also used for training a network by a layer-by-layer greedy training method, the input weight of each hidden layer of the depth limit learning machine is initialized by using ELM-AE, and layered unsupervised training is performed, but the depth limit learning machine is different from the traditional depth learning algorithm in that the ELM-AE does not need a reverse fine tuning process. The idea of the deep extreme learning machine is that the output can be infinitely close to the original input by reducing the reconstruction error to the maximum extent, and the high-level characteristics of the original data can be learned through the training of each layer. FIG. 5 depicts the training process of the extreme depth learning machine model with input data sample E as the target output for the 1 st ELM-AE (E)1E), and further obtain an output weight Q1(ii) a Then the output matrix C of the 1 st hidden layer of the depth limit learning machine is used1As input and target output for the next 1 ELM-AE (E)2＝C1) And the analogy is that training layer by layer is repeated, the last 1 layer is trained by ELM, and the output weight Q of the last 1 hidden layer of the depth limit learning machine is solved by using an equation (21))i+1. C in FIG. 5i+1Is the output matrix of the last 1 hidden layer and l is the sample label. The input weight matrix of each 1 hidden layer of the deep extreme learning machine is
Figure BDA0002752691130000143
Therefore, the operation state of the converter transformer can be effectively identified after the characteristic vector after the vibration signal decomposition is input into the depth limit learning machine and classified, and the flow of the overall method is shown in fig. 6.
Under the support of the theory, the vibration signals of the converter transformer are measured, and a group of vibration signals are obtained through measurement. And decomposing by adopting an optimized variational modal decomposition algorithm, wherein the solved characteristic vectors are shown in table 1.
TABLE 1 eigenvectors of the vibration signal of the converter transformer
Figure BDA0002752691130000151
The result shows that the eigenvalues of different components in the eigenvector of the separated vibration signal have obvious difference, and are suitable for signal classification, therefore, the collected vibration signal of the converter transformer totally 1834 samples, the training set sample set is divided according to the proportion of 8:2, and the corresponding label is calibrated according to the running state of the converter transformer. The training set is input into a DELM model for training, the training result is shown in FIG. 7, and it can be seen from the graph that after 100 iterations, the classification accuracy of the training set reaches 99.4548%, indicating that the training result is better. And inputting the test set into the trained model, wherein the test result is shown in fig. 8, the classification accuracy of the test set reaches 99.4536%, and the model can effectively realize the identification of the operation state of the converter transformer based on the vibration signal.
In summary, when the converter transformer is in different operating states, the vibration signal of the converter transformer changes, but due to the complex internal structure of the converter transformer, it is difficult to determine the significant features of the converter transformer by using the traditional vibration analysis method, so that accurate classification and identification cannot be performed. After the optimized variational modal decomposition algorithm is adopted to decompose the vibration signals of the converter transformer, effective characteristic vectors of the vibration signals can be effectively obtained, and therefore a depth limit learning machine can be utilized to carry out accurate signal classification to determine the running state of the converter transformer, maintenance is facilitated, and the safety and stability of a power system are improved.
The method disclosed by the invention is based on the variational modal decomposition optimization algorithm and the deep extreme learning machine, realizes the state identification of the converter transformer by analyzing the vibration signal, reduces the risk of abnormal operation of the converter transformer, and improves the safety and reliability of the operation of the power system.
The above embodiments are only for illustrating the technical concept and features of the present invention, and the purpose thereof is to enable those skilled in the art to understand the contents of the present invention and implement the present invention accordingly, and not to limit the protection scope of the present invention accordingly. All equivalent changes or modifications made in accordance with the spirit of the present disclosure are intended to be covered by the scope of the present disclosure.

Claims (7)

1. A method for identifying the running state of a converter transformer based on a vibration signal is characterized by comprising the following steps:
collecting a vibration acceleration signal of the converter transformer;
dividing the vibration acceleration signal into a training sample set and a test sample set; wherein the training sample set is used for seeking the optimal set number of modal functions and a penalty factor;
according to a variational modal decomposition algorithm, establishing a variational model with constraint conditions by using the optimal set number of the modal functions and the penalty factors;
extracting energy characteristics of the variation model with the constraint condition to obtain a vibration sample signal characteristic set and a test sample signal characteristic set, and training the vibration sample signal characteristic set through a depth limit learning machine to obtain a trained model;
and inputting the test sample signal characteristic set into the trained model to obtain a classification result of the vibration acceleration signal, and identifying the running state of the converter transformer according to the classification result.
2. The method for identifying the operational status of the converter transformer based on the vibration signal as claimed in claim 1, wherein the method for seeking the optimal set number of the modal functions and the penalty factor by using the training sample set is as follows:
establishing an objective function model;
initializing an optimizing initial position in a parameter space;
converting the position of the parameter space into the set number and the penalty factor of a preset modal function;
processing according to a variational modal decomposition algorithm by using the set number of preset modal functions and a penalty factor;
calculating objective function values of all training sample sets;
judging whether the objective function value reaches the maximum value; if so, taking the set number and the penalty factor of the preset modal function as the set number and the penalty factor of the optimal modal function, otherwise, updating the position of the next parameter space by using a longicorn searching algorithm, converting the position of the updated parameter space into the set number and the penalty factor of the preset modal function, and repeating the steps until the set number and the penalty factor of the optimal modal function are obtained.
3. The method for identifying the operating state of the converter transformer based on the vibration signal according to claim 1, wherein the collecting of the vibration acceleration signal of the converter transformer is performed by the following method:
mounting a multi-channel vibration sensor on the surface of a converter transformer box body; and transmitting the vibration acceleration signals output by the surface of the converter transformer box body to a computer through an integrated measuring device, and carrying out state recognition on the collected vibration acceleration signals in the computer.
4. The method for identifying the operation state of the converter transformer based on the vibration signal according to claim 2, wherein the variational modal decomposition algorithm comprises the following steps:
defining each sub-signal as a mode function Ui(t), the original signal may be represented as:
Figure FDA0002752691120000011
wherein K denotes the number of modes, Ai(t) denotes the instantaneous amplitude value,
Figure FDA0002752691120000021
indicating phase, instantaneous frequency
Figure FDA0002752691120000022
Variational model with constraint conditions:
Figure FDA0002752691120000023
wherein: [ U ]i]Represented is a set of K modal functions, [ omega ] omegai]A set of corresponding center frequencies is represented,
Figure FDA0002752691120000024
the gradient operation is shown, and the real part of the mode function is shown in R (t);
in order to obtain the optimal solution of the model, a Lagrangian multiplier lambda and a penalty factor alpha are adopted for the constraint variation problem, the Lagrangian multiplier lambda and the penalty factor alpha are changed into an unconstrained condition, and an augmented Lagrangian function is established:
Figure FDA0002752691120000025
the saddle point of the above formula is solved by an alternative direction multiplier method, and the mode component in the frequency domain can be obtained by converting the time domain signal into the frequency domain signal by Fourier transformation
Figure FDA0002752691120000026
And center frequency
Figure FDA0002752691120000027
The expression of (1);
Figure FDA0002752691120000028
Figure FDA0002752691120000029
iterative updating in operation process
Figure FDA00027526911200000210
ωiAnd
Figure FDA00027526911200000211
the optimal solution can be obtained by solving, and the specific flow is as follows:
1) initialization
Figure FDA00027526911200000212
And
Figure FDA00027526911200000213
λ1＝n＝0；
2) iterative update based on equation (4)
Figure FDA00027526911200000214
And ωi；
3) Updating Lagrange multiplication operator on the basis of formula (6)
Figure FDA00027526911200000215
Wherein τ represents the update parameter of the Lagrangian multiplier;
Figure FDA00027526911200000216
4) and (3) repeating the steps (1) and (3) until the constraint condition that the operation is stopped is taken as an equation (7), wherein epsilon represents the iteration precision, and the result obtained after the iteration is finished is the modal component and the center frequency of the variation modal decomposition.
Figure FDA00027526911200000217
5. The method for identifying the running state of the converter transformer based on the vibration signal as claimed in claim 4, wherein the longicorn whisker searching algorithm comprises the following steps:
the position of the longicorn can be defined as sn(N-1, 2, …, N), where N denotes the number of iterations of the search, and the food smell is formed by an objective function f(s) associated with the position, the extreme value of which is the final position of the food;
the global optimization of the algorithm in the multidimensional data space can be represented by the following formula:
Figure FDA0002752691120000031
the global optimization summary of the algorithm is exploration behavior and traveling behavior, wherein the orientation of the longicorn whiskers can be represented by a normalized random vector D
Figure FDA0002752691120000032
In formula (9), Random is a Random function; d is the dimension of the data space. And the left and right palpus positions slAnd srThen can be defined as
Figure FDA0002752691120000033
In the formula, anThe length of the whisker; the advancing behavior is to determine the stepping length and the direction of the longicorn according to the magnitude relation of the objective function values corresponding to the left and right tentacle positions, so as to update the position of the longicorn; its update location can be expressed as
sn＝sn-1+ξnDsign[f(xr)-f(xl)] (11)
Xi in formula (11)nSign is a sign function for the step length of each iteration; each time the exploration behavior and the advancing behavior are executed, an iteration process of the algorithm can be defined, and a is carried out after each iterationnAnd xinNeeds to be updated, the updating formula is
Figure FDA0002752691120000034
Update coefficient c in equation (12)1And c3Set to 0.95, c2Set to 0.01; and after the algorithm reaches the preset iteration times, the data space position corresponding to the optimal objective function value in the past iteration is the optimizing result.
6. The method for identifying the operation state of the converter transformer based on the vibration signal as claimed in claim 5, wherein the objective function model F(s) of the parameter optimization is implemented by the following steps:
setting the length of a vibration signal y (t) as L, and if the proportion of the reconstructed signal energy after the variation modal decomposition in the original signal energy is defined as E, the mathematical model can be expressed as
Figure FDA0002752691120000041
In formula (13), Ui(t) represents the reconstructed signal amplitude of the ith mode, and K represents the number of modes;
the central frequency variation range of the modal bandwidth estimate is called b, which can be defined as
Figure FDA0002752691120000042
In the formula, max and min are respectively functions for solving the maximum value and the minimum value; f. ofsIs the signal sampling frequency; omegai(n) is the ith modality in an iterative processAn estimated center frequency; m is the number of estimated center frequencies in the mode;
Figure FDA0002752691120000043
wherein s ═ p, q](p belongs to K, q belongs to alpha) is any parameter combination in the optimizing space; n is the number of qualified training samples; ejAnd bjE and b corresponding to the j training sample vibration signal; 0.5 can control the calculation result to be [0, 100 ]]The weight coefficient of (2).
7. The method for identifying the operating state of the converter transformer based on the vibration signal as claimed in claim 6, wherein the deep extreme learning machine is constructed by the following steps:
constructing an extreme learning machine:
t different marked samples are obtained and obtained,
Figure FDA0002752691120000044
denotes the ith sample, corresponding to label li＝[liz,liz,…,liz]TWherein z represents the number of classes; if sample
Figure FDA0002752691120000045
If the tag belongs to the m-th class, the m-th value of the corresponding tag vector is set to be 1, and the rest (m-1) values are set to be-1; the extreme learning machine having H hidden nodes can be represented by equation (16):
Figure FDA0002752691120000046
wherein Q isi＝[Qi1,Qi2,…,Qiz]TDenoted is the output weight connecting the ith hidden layer and the output layer, Ai＝[Ai1,Ai2,…,Aiz]TIs an input connecting an input layer and an i-th hidden layerWeight, BiIs the bias of the i-th hidden layer, G (A)i,Bi,Ej) Is the output of the ith hidden layer; assuming that the activation function of the ith hidden layer is g (x), the output of the hidden layer is:
G(Ai,Bi,Ej)＝g(Ai·Ej+Bi) (17)
the formula (16) is summarized in a matrix form as
Figure FDA0002752691120000051
Wherein the content of the first and second substances,
Figure FDA0002752691120000052
in order to satisfy the expected value output after network training, the parameter A needs to be obtainedi、Bi、QiOptimum value of (2)
Figure FDA0002752691120000053
Such that:
Figure FDA0002752691120000054
then the minimization loss function is
Figure FDA0002752691120000055
In extreme learning machine, parameter (A)i,Bi) Is randomly initialized, so QiIs uniquely determined; the solution can be converted into:
Figure FDA0002752691120000056
the norm of the solution Q is minimal and uniqueFirstly, performing primary filtration; to make the model have better generalization performance, L can be added1Regular terms, the above solution problem can be transformed into equation (21):
Figure FDA0002752691120000057
wherein, CLIs a regularization coefficient; solving is essentially a ridge regression problem whose solution is shown below:
Figure FDA0002752691120000058
wherein I is an identity matrix;
constructing an extreme learning machine-based self-encoder: applying the idea of an automatic encoder to an extreme learning machine, so that the input data is also used for output, namely the output Y is equal to X;
the output of the extreme learning machine-based self-encoder can be converted by equation (16) to:
Figure FDA0002752691120000059
wherein A isTA＝I，BTB＝1；
Wherein A is AiA composed matrix of BiA vector of components; for sparse representation and dimension compression, the hidden layer output weight Q can be converted by equation (23):
Figure FDA0002752691120000061
wherein E ═ E1,E2,…,E8]The feature vector is formed by the proportion of each modal energy after the variation modal decomposition of the input vibration signal; for the equal-dimensional feature mapping, the weight Q can be calculated by equation (26):
Q＝C-1l (26)
wherein Q isTQ＝I；
Constructing a depth extreme learning machine:
taking the feature representation capability of an extreme learning machine-based self-encoder as a basic unit of a depth extreme learning machine;
taking the input data sample E as the target output of the 1 st self-encoder based on the extreme learning machine, and further solving the output weight Q1(ii) a Then the output matrix C of the 1 st hidden layer of the depth limit learning machine is used1When the input and the target output of the next 1 self-encoder based on the extreme learning machine are considered, the training is carried out by analogy, the last 1 layer is trained by the extreme learning machine, and the output weight Q of the last 1 hidden layer of the deep extreme learning machine is solved by using an equation (21)i+1。
CN202011190818.6A 2020-10-30 2020-10-30 Converter transformer running state identification method based on vibration signals Pending CN112307963A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011190818.6A CN112307963A (en) 2020-10-30 2020-10-30 Converter transformer running state identification method based on vibration signals

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011190818.6A CN112307963A (en) 2020-10-30 2020-10-30 Converter transformer running state identification method based on vibration signals

Publications (1)

Publication Number Publication Date
CN112307963A true CN112307963A (en) 2021-02-02

Family

ID=74332730

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011190818.6A Pending CN112307963A (en) 2020-10-30 2020-10-30 Converter transformer running state identification method based on vibration signals

Country Status (1)

Country Link
CN (1) CN112307963A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112785080A (en) * 2021-02-03 2021-05-11 燕山大学 Energy consumption optimization method of real-time dynamic cement grinding system based on cement industry
CN113269082A (en) * 2021-05-21 2021-08-17 合肥工业大学 Partial discharge threshold denoising method based on improved variational modal decomposition
CN113537402A (en) * 2021-08-13 2021-10-22 重庆大学 Vibration signal-based converter transformer multi-scale fusion feature extraction method
CN116680623A (en) * 2023-08-03 2023-09-01 国网安徽省电力有限公司超高压分公司 State monitoring method and system for extra-high voltage converter transformer
CN117674199A (en) * 2024-02-01 2024-03-08 西安热工研究院有限公司 Novel power system frequency modulation method and system for super-capacitor coupled lithium battery

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112785080A (en) * 2021-02-03 2021-05-11 燕山大学 Energy consumption optimization method of real-time dynamic cement grinding system based on cement industry
CN113269082A (en) * 2021-05-21 2021-08-17 合肥工业大学 Partial discharge threshold denoising method based on improved variational modal decomposition
CN113537402A (en) * 2021-08-13 2021-10-22 重庆大学 Vibration signal-based converter transformer multi-scale fusion feature extraction method
CN116680623A (en) * 2023-08-03 2023-09-01 国网安徽省电力有限公司超高压分公司 State monitoring method and system for extra-high voltage converter transformer
CN116680623B (en) * 2023-08-03 2023-10-27 国网安徽省电力有限公司超高压分公司 State monitoring method and system for extra-high voltage converter transformer
CN117674199A (en) * 2024-02-01 2024-03-08 西安热工研究院有限公司 Novel power system frequency modulation method and system for super-capacitor coupled lithium battery
CN117674199B (en) * 2024-02-01 2024-04-23 西安热工研究院有限公司 Power system frequency modulation method and system of super-capacitor coupling lithium battery

Similar Documents

Publication Publication Date Title
CN112307963A (en) Converter transformer running state identification method based on vibration signals
Sharma et al. Review of artificial intelligence techniques application to dissolved gas analysis on power transformer
CN104966161B (en) A kind of power quality recorder data calculation and analysis methods based on gauss hybrid models
CN112508442B (en) Transient stability assessment method and system based on automatic and interpretable machine learning
Hassani et al. Fault location in smart grids through multicriteria analysis of group decision support systems
CN109657882A (en) Short-term power load prediction model establishment method based on VMD-PSO-LSSVM
Rezaeian Koochi et al. New phasor‐based approach for online and fast prediction of generators grouping using decision tree
CN116245033A (en) Artificial intelligent driven power system analysis method and intelligent software platform
Lin et al. Intelligent prediction of the construction cost of substation projects using support vector machine optimized by particle swarm optimization
CN115225516A (en) Least square support vector machine network flow prediction method based on improved ABC-VMD
Li et al. An optimized GRNN‐enabled approach for power transformer fault diagnosis
Milosavljevic et al. Integrated transformer health estimation methodology based on Markov chains and evidential reasoning
CN112926620B (en) Defect type identification method and system for partial discharge of transformer bushing
Gao et al. Mechanical fault diagnosis of an on‐load tap changer by applying cuckoo search algorithm‐based fuzzy weighted least squares support vector machine
Haiqiang et al. Ultra‐short‐term wind speed forecasting method based on spatial and temporal correlation models
Xu et al. An improved ELM-WOA–based fault diagnosis for electric power
Khalyasmaa et al. Training samples construction for energy utilities operational assets management
Wang et al. Cloud computing and extreme learning machine for a distributed energy consumption forecasting in equipment-manufacturing enterprises
Wong et al. Classification of the harmonic load types using multi-layer extreme learning machine
Wilson et al. Uncertainty Quantification of Capacitor Switching Transient Location Using Machine Learning
CN115577290A (en) Distribution network fault classification and source positioning method based on deep learning
CN113128130B (en) Real-time monitoring method and device for judging stability of direct-current power distribution system
Panapongpakorn et al. Short-Term Load Forecast for Energy Management System Using Neural Networks with Mutual Information Method of Input Selection
Chen et al. An Improved AdaBoost-based Ensemble Learning Method for Data-Driven Dynamic Security Assessment of Power Systems
Hailu et al. Application of Data‐Driven Tuned Fuzzy Inference System for Static Equivalencing of Power Systems with High Penetration of Renewable Energy

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