CN110082082B - GIS state identification method based on vibration signal principal component analysis method - Google Patents

GIS state identification method based on vibration signal principal component analysis method Download PDF

Info

Publication number
CN110082082B
CN110082082B CN201910350848.XA CN201910350848A CN110082082B CN 110082082 B CN110082082 B CN 110082082B CN 201910350848 A CN201910350848 A CN 201910350848A CN 110082082 B CN110082082 B CN 110082082B
Authority
CN
China
Prior art keywords
gis
vibration signal
principal component
characteristic
imf
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910350848.XA
Other languages
Chinese (zh)
Other versions
CN110082082A (en
Inventor
赵延刚
苏旭辉
王志川
黄小龙
王泽龙
龙伟
任成君
欧智乐
张大猛
高波
赵冲
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Sichuan Electric Power Co Ltd
Original Assignee
State Grid Sichuan Electric Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by State Grid Sichuan Electric Power Co Ltd filed Critical State Grid Sichuan Electric Power Co Ltd
Priority to CN201910350848.XA priority Critical patent/CN110082082B/en
Publication of CN110082082A publication Critical patent/CN110082082A/en
Application granted granted Critical
Publication of CN110082082B publication Critical patent/CN110082082B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention discloses a GIS state identification method based on a vibration signal principal component analysis method, which extracts 14 characteristic quantities of a GIS vibration signal under normal and different fault states to form 14-dimensional composite characteristic vectors; then, compressing and dimensionality reducing the feature vectors into principal component feature vectors by using a principal component analysis method; and then, obtaining a corresponding decision function between the vibration signal principal component feature vector and the GIS state through two-stage training of a deep confidence network, classifying the GIS vibration signal principal component feature vector to be state recognized by using the decision function, and determining the GIS state according to the classification. The composite eigenvector is optimized by using a principal component analysis method, so that the original information of the composite eigenvector is kept, the dimension of the eigenvector is reduced, the working efficiency of the classifier is improved, and the precision and the speed of GIS state identification are effectively improved.

Description

GIS state identification method based on vibration signal principal component analysis method
Technical Field
The invention relates to the technical field of gas insulated metal enclosed switchgear (GIS) state recognition, in particular to a GIS state recognition method based on a vibration signal principal component analysis method.
Background
Gas Insulated metal enclosed Switchgear (GIS) appeared in the last 60 th century, and has been rapidly developed due to its advantages of small floor space, high reliability, high safety, short installation period, small labor for maintenance, and the like, and is now widely used in all levels of substations in the world. However, due to the complex fully-closed structure, once the GIS fails, the GIS can be affected in a large range and is difficult to accurately position and quickly repair.
Therefore, in order to ensure the safe and stable operation of the power grid, the reliability of the GIS is very important, and therefore, effective identification of the fault state of the GIS is urgent. At present, means for monitoring the running state of the GIS equipment include ultrasonic partial discharge detection, ultrahigh frequency detection, optical analysis, chemical analysis and the like, but although the methods are widely accepted by researchers or field workers, the methods are only suitable for detecting the discharge fault inside the GIS, and the mechanical fault in the GIS is difficult to identify. The frequency spectrum characteristics of the vibration signals are important indexes for representing mechanical characteristics, and due to the sensitivity of the frequency spectrum characteristics to mechanical faults in the GIS, GIS state identification based on the vibration signals becomes a recent research hotspot.
The existing vibration signal-based GIS state identification usually performs GIS state identification on a single characteristic quantity of a vibration signal, but the mechanical state and the characteristic quantity of the GIS are not in one-to-one correspondence, and different mechanical states may cause the change of the same characteristic quantity, so that the GIS state identification according to the single characteristic quantity often causes misjudgment. In order to improve the detection accuracy, multiple characteristic quantities are required to be collected to form a composite characteristic vector, and the GIS state is reliably judged by combining information complementary relations among different characteristic quantities, but the increase of the characteristic quantities can increase the workload of a computer and reduce the calculation speed and precision.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the existing GIS state identification based on vibration signals is characterized in that the GIS state is reliably judged by selecting different characteristic quantities and combining information complementary relations among the different characteristic quantities, but the increase of the characteristic quantities can increase the workload of a computer and reduce the calculation speed and precision. Therefore, it is important to select an appropriate feature amount when performing GIS state recognition based on a vibration signal. The invention provides a GIS state identification method based on a vibration signal principal component analysis method, which solves the problems, keeps GIS vibration signal characteristics as much as possible, has higher state identification efficiency, and can realize GIS state judgment of integrating various characteristic quantities with high efficiency and high precision.
The invention is realized by the following technical scheme:
a GIS state identification method based on a vibration signal principal component analysis method comprises the following steps:
step 1: collecting multiple groups of vibration signals of the GIS in normal and fault states by a vibration acceleration sensor arranged on the GIS;
step 2: processing the collected GIS vibration signals, respectively extracting the time domain, the frequency domain and the energy characteristics of the GIS vibration signals, and constructing a GIS vibration signal composite characteristic vector;
and step 3: optimizing the GIS vibration signal composite eigenvector by adopting a principal component analysis method to obtain a GIS vibration signal principal component eigenvector;
and 4, step 4: constructing a deep confidence network model, taking multiple groups of GIS vibration signal principal component feature vectors in known states as training samples of the deep confidence network recognition model, and obtaining a corresponding decision function between the GIS vibration signal principal component feature vectors and the GIS states through two-stage training of the deep confidence network model;
and 5: and collecting GIS vibration signals to be state recognized, calculating and analyzing principal component eigenvectors, classifying and recognizing the GIS vibration signals by using corresponding decision functions between the principal component eigenvectors and the GIS states, and recognizing the GIS states.
Further, the time domain characteristics of the GIS vibration signal extracted in the step 2 comprise a GIS vibration signal peak value, an average value, skewness and kurtosis;
the frequency domain characteristics comprise a principal component frequency, a 100Hz ratio and a 50Hz odd-order frequency multiplication ratio;
and decomposing the GIS vibration signal into a plurality of modal components IMF by using a set empirical mode decomposition algorithm, calculating the energy of each IMF, and taking the energy of the first 7 modal components IMF accounting for more than 90% of the original signal energy as the energy characteristic of the GIS vibration signal.
Further, constructing a GIS vibration signal composite characteristic vector alpha (k) according to the 14-dimensional characteristic quantity extracted from the GIS vibration signal (the 14-dimensional characteristic quantity is the energy of the first 7 modal components IMF with the peak value, the average value, the skewness and the kurtosis of the GIS vibration signal, the main component frequency, the 100Hz ratio, the 50Hz odd-order frequency multiplication ratio and the ratio exceeding 90% in the original signal energy, and numbering the characteristic quantities in sequence)1,k2，...k14)TWherein: alpha is a composite characteristic vector k of the GIS vibration signal1Is the peak-to-peak value of the first dimension characteristic quantity, k2Is the average value of the second dimension characteristic quantity, the rest are analogized in turn, and the superscript T is (k)1,k2，...k14) The transposing of (1).
Further, the step 3 specifically includes:
step 3.1: forming m groups of GIS vibration signals into an m multiplied by 14 matrix
X＝(α1,α2，...αm)T＝(x1,x2，...x14)
Step 3.2: the covariance matrix C of matrix X is calculated according to the following expression:
Figure BDA0002043890840000031
in the formula: cov (x, y) denotes the covariance of the two sets of data;
step 3.3: calculating an eigenvalue λ of the covariance matrix Ci(i 1,2.. 14) and a corresponding eigenvector matrix E, arranging the obtained eigenvalues in a descending order, rearranging the columns in the eigenvector matrix E in the descending order to obtain a transition matrix T, and taking each column in the T as a referenceThe column vector is a characteristic factor;
step 3.4: calculating the contribution rate of each characteristic factor, selecting 2 characteristic factors of which the sum of the contribution rates exceeds 95% to form a transformation matrix, wherein the calculation formula of the contribution rate of each characteristic factor is as follows:
Figure BDA0002043890840000032
in the formula: krDenotes the contribution of the r-th characteristic factor, λrDenotes the characteristic value, λ, corresponding to the r-th characteristic factorjRepresenting the characteristic value corresponding to the jth characteristic factor,
Figure BDA0002043890840000033
represents a pair ofjFrom λ1To lambdamSumming;
step 3.5: and obtaining an m × 2 feature matrix Y by matrix operation Y ═ X × U, wherein each column of the matrix Y is a principal component feature vector of a group of GIS vibration signals.
Further, decomposing the GIS vibration signal into a plurality of modal components IMF by using a set empirical mode decomposition algorithm, wherein a calculation formula for calculating the energy of each IMF is as follows:
Figure BDA0002043890840000034
in the formula: rjRepresenting the sum of the energies in each modal component IMF, nIMFRepresents the total amount of data contained in the nth modal component IMF,
Figure BDA0002043890840000035
representing an energy value of each GIS vibration signal data point;
in order to simplify the calculation, 2 norms of each modal component IMF are used to characterize the energy characteristics, and the calculation formula for calculating the energy of each IMF can be simplified as follows:
Figure BDA0002043890840000036
in the formula: v. ofnThe IMF energy is for each modal component.
Further, in step 4, the data is trained by using a Deep Belief Network (DBN), which is a deep neural network formed by stacking a plurality of unsupervised Restricted Boltzmann Machines (RBMs) and a supervised back propagation network (BP), and the two stages of training by the deep belief network model are unsupervised pre-training from a lower layer to a higher layer and supervised fine tuning from the higher layer to the lower layer.
Wherein: the first stage is to adopt a greedy algorithm to train each restricted Boltzmann machine RBM unsupervised, after the training of the RBM of the lower layer is finished, the output of the RBM is used as the input of the RBM of the upper layer, and the RBM is trained layer by layer in sequence, so that the characteristics of the higher layer are learned, and the training parameters of each layer are continuously updated, and the greedy algorithm cannot enable the learning parameters between the layers to be optimal, so that the fine adjustment of the second stage is needed; and in the second stage, a final layer of BP network is trained in a supervision mode, the error generated in the first stage is reversely transmitted to each layer of RBM below, and parameters among all RBM layers are finely adjusted according to the transmission result, so that the parameters of the whole deep belief network model are optimal. Through the adjustment of the two-stage learning parameters, the input features of the data are abstracted into higher-order features, so that a better classification effect is obtained.
The invention has the following advantages and beneficial effects:
1. the method extracts the peak value, the average value, the skewness principal component frequency, the 100Hz ratio, the 50Hz odd-order frequency multiplication ratio and the energy of the first 7 IMF components with the ratio exceeding 90 percent in the original signal energy of the GIS vibration signal as the vibration signal characteristic quantity, comprises the time domain, the frequency domain and the energy characteristic of the vibration signal, and can reflect the characteristic of the GIS vibration signal more completely;
2. the invention adopts a Deep Belief Network (DBN), classifies GIS vibration signals by a machine learning principle, and has higher accuracy and faster convergence rate;
3. the invention optimizes the composite feature vector by adopting a principal component analysis method, not only retains the original information of the composite feature vector, but also reduces the dimension of the feature vector, improves the working efficiency of the classifier and effectively improves the precision and speed of GIS state identification.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention. In the drawings:
FIG. 1 is a block flow diagram of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to examples and accompanying drawings, and the exemplary embodiments and descriptions thereof are only used for explaining the present invention and are not meant to limit the present invention.
Examples
As shown in fig. 1, a method for identifying a GIS state based on a vibration signal principal component analysis method includes:
step 1: installing a vibration acceleration sensor on the GIS, and collecting vibration signals of a plurality of groups of GIS in normal and fault states;
step 2: processing the GIS vibration signals acquired in the step 1, respectively extracting the time domain, the frequency domain and the energy characteristics of the GIS vibration signals, and constructing a GIS vibration signal composite feature vector;
step 2.1: extracting GIS vibration signal time domain characteristics including GIS vibration signal peak-to-peak value, average value, skewness and kurtosis;
(1) peak to peak value
The peak value can represent the GIS vibration intensity, and the calculation formula is as follows:
xpp＝xmax-xmin (1)
in the formula: x is the number ofmaxRepresenting the maximum value, x, of the GIS vibration signalminRepresenting the minimum value, x, of the GIS vibration signalppAnd representing the GIS vibration signal intensity.
(2) Mean value of
The average value may represent the deviation from the equilibrium position, and is calculated as:
Figure BDA0002043890840000051
in the formula: in
Figure BDA0002043890840000052
Representing the average value of the GIS vibration signals, n representing the total data amount of the GIS vibration signals, xiEach data point is represented.
(3) Deflection degree
The skewness can represent the asymmetry of the GIS vibration signal relative to the average value, and the calculation formula is as follows:
Figure BDA0002043890840000053
in the formula: skaw (x) denotes signal skewness, σ denotes data variance,
Figure BDA0002043890840000054
representing the average value of the GIS vibration signals, n representing the total data amount of the GIS vibration signals, xiEach data point is represented.
(4) Kurtosis
Kurtosis is a numerical statistic for describing the kurtosis of a signal waveform, and can represent the distribution characteristics of data, and the calculation formula is as follows:
Figure BDA0002043890840000055
in the formula: k represents the GIS vibration signal kurtosis, E (t) represents the mathematical expectation of the data,
Figure BDA0002043890840000056
and represents the GIS vibration signal average value, and sigma represents the data variance.
And calculating the peak value, the average value, the skewness and the kurtosis of the collected GIS vibration signals according to the formulas.
Step 2.2: extracting GIS vibration signal time domain characteristics, including principal component frequency, 100Hz ratio and 50Hz odd-order frequency multiplication ratio;
(1) frequency of principal component
The principal component frequency is the frequency at which the amplitude is maximum. When the GIS normally operates, the vibration is basically concentrated at 100Hz, and the frequency of the main component is 100 Hz. When the GIS fails, other frequency components may be greatly increased to become the main component frequency instead of 100 Hz.
(2)100Hz ratio
When the GIS normally operates, the fundamental frequency is 100Hz, and the amplitudes of other frequency components can be almost ignored. When a fault occurs, the amplitudes of other frequencies are correspondingly increased, and the 100Hz ratio is changed.
(3)50Hz odd-order frequency multiplication ratio
When the GIS normally operates, the vibration signal frequency spectrum of the GIS does not have 50Hz odd-order frequency multiplication components, when a fault occurs, corresponding components can appear, and the ratio of the components can also correspondingly change.
Step 2.3: decomposing the GIS vibration signal into a plurality of modal components IMF by using a set empirical mode decomposition algorithm, wherein the calculation formula for calculating the energy of each IMF is as follows:
Figure BDA0002043890840000061
in the formula: rjRepresenting the sum of the energies in each modal component IMF, nIMFRepresents the total amount of data contained in the nth modal component IMF,
Figure BDA0002043890840000062
representing an energy value of each GIS vibration signal data point;
in order to simplify the calculation, 2 norms of each modal component IMF are used to characterize the energy characteristics, and the calculation formula for calculating the energy of each IMF can be simplified as follows:
Figure BDA0002043890840000063
equation (6) is a 2-norm calculation of the energy of each IMF component based on equation (5).
According to an empirical value, the energy of the IMF of the first 7 modal components accounting for more than 90% of the original signal energy is taken as the GIS vibration signal energy characteristic.
Step 2.4: sequentially forming a vibration signal composite characteristic vector alpha (k) by the 14 characteristic quantities (14-dimensional characteristic quantities are peak value, average value, skewness and kurtosis of the GIS vibration signal, main component frequency, 100Hz ratio, 50Hz odd-order frequency multiplication ratio, energy of the first 7 modal components IMF with the ratio of more than 90% in the original signal energy, and sequentially numbering the characteristic quantities according to the sequence)1,k2，...k14)TWherein: alpha is a composite characteristic vector k of the GIS vibration signal1Is the peak-to-peak value of the first dimension characteristic quantity, k2Is the average value of the second dimension characteristic quantity, the rest are analogized in turn, and the superscript T is (k)1,k2，...k14) The transposing of (1).
And step 3: optimizing the GIS vibration signal composite eigenvector constructed in the step 2 by adopting a principal component analysis method, and compressing and reducing dimensions to obtain the GIS vibration signal principal component eigenvector;
the step 3 specifically comprises the following steps:
step 3.1: forming m groups of GIS vibration signals into m multiplied by 14 matrix of
X＝(α1,α2，...αm)T＝(x1,x2，...x14)
Step 3.2: the covariance matrix C of matrix X is calculated according to the following expression:
Figure BDA0002043890840000071
in the formula: cov (x, y) denotes the covariance of the two sets of data;
step 3.3: calculating an eigenvalue λ of the covariance matrix Ci(i 1,2.. 14) and a corresponding eigenvector matrix E, and the obtained eigenvalues are compared with each otherArranging in descending order, rearranging the columns in the eigenvector matrix E according to the order to obtain a transition matrix T, and taking each column vector in the T as an eigenfactor;
step 3.4: calculating the contribution rate of each characteristic factor, selecting 2 characteristic factors of which the sum of the contribution rates exceeds 95% to form a transformation matrix according to an empirical value, wherein the calculation formula of the contribution rate of each characteristic factor is as follows:
Figure BDA0002043890840000072
in the formula: krDenotes the contribution of the r-th characteristic factor, λrDenotes the characteristic value, λ, corresponding to the r-th characteristic factorjRepresenting the characteristic value corresponding to the jth characteristic factor,
Figure BDA0002043890840000073
represents a pair ofjFrom λ1To lambdamAnd (6) summing.
Step 3.5: and obtaining an m × 2 feature matrix Y by matrix operation Y ═ X × U, wherein each column of the matrix Y is a principal component feature vector of a group of GIS vibration signals.
And 4, step 4: according to the GIS vibration signal principal component feature vector obtained in the step 3, a deep confidence network model is constructed, multiple groups of GIS vibration signal principal component feature vectors in known states are used as training samples of the deep confidence network recognition model, and a corresponding decision function between the GIS vibration signal principal component feature vector and the GIS state is obtained through two-stage training of the deep confidence network model;
the deep belief network used in the method trains data, and the deep neural network is formed by stacking a plurality of unsupervised limited Boltzmann machines (RBMs) and a supervised back propagation network (BP), and the two stages of training through the deep belief network model are unsupervised pre-training from a low layer to a high layer and supervised fine-tuning from the high layer to the low layer respectively.
The first stage is to adopt a greedy algorithm to train each restricted Boltzmann machine RBM unsupervised, after the training of the RBM of the lower layer is finished, the output of the RBM is used as the input of the RBM of the upper layer, and the RBM is trained layer by layer in sequence, so that the characteristics of the higher layer are learned, and the training parameters of each layer are continuously updated, and the greedy algorithm cannot enable the learning parameters between the layers to be optimal, so that the fine adjustment of the second stage is needed;
and in the second stage, a final layer of BP network is trained in a supervision mode, the error generated in the first stage is reversely transmitted to each layer of RBM below, and parameters among all RBM layers are finely adjusted according to the transmission result, so that the parameters of the whole deep belief network model are optimal.
Through the adjustment of the two-stage learning parameters, the input features of the data are abstracted into higher-order features, so that a better classification effect is obtained.
And 5: and collecting GIS vibration signals to be state recognized, calculating and analyzing principal component eigenvectors, classifying and recognizing the GIS vibration signals by using corresponding decision functions between the principal component eigenvectors and the GIS states, and recognizing the GIS states.
The working principle is as follows: the GIS state recognition method based on the vibration signal principal component analysis method comprises the steps of firstly, extracting the peak value, the average value, the skewness principal component frequency, the 100Hz ratio, the 50Hz odd-order frequency multiplication ratio and the energy of the first 7 IMF components with the ratio exceeding 90% in the original signal energy as vibration signal characteristic quantities, constructing a GIS vibration signal composite characteristic vector, wherein the 14-dimensional characteristic quantities comprise the time domain, the frequency domain and the energy characteristics of the vibration signal, and can reflect the characteristics of the GIS vibration signal more completely; secondly, the composite feature vector is optimized by adopting a principal component analysis method, so that not only is the original information of the composite feature vector retained, but also the feature vector dimension is reduced, the working efficiency of the classifier is improved, and the precision and the speed of GIS state identification are effectively improved; then, a deep confidence network is adopted, and GIS vibration signals are classified according to a machine learning principle, so that the method has high accuracy and high convergence rate; and finally, collecting the GIS vibration signals to be state-recognized, calculating and analyzing the principal component characteristic vectors, classifying and recognizing the GIS vibration signals by using corresponding decision functions between the principal component characteristic vectors and the GIS states, and recognizing the GIS states, thereby being capable of realizing GIS state judgment of integrating various characteristic quantities with high efficiency and high precision.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (5)

1. A GIS state identification method based on a vibration signal principal component analysis method is characterized in that: the method comprises the following steps:
step 1: collecting multiple groups of vibration signals of the GIS in normal and fault states by a vibration acceleration sensor arranged on the GIS;
step 2: processing the collected GIS vibration signals, respectively extracting the time domain, the frequency domain and the energy characteristics of the GIS vibration signals, and constructing a GIS vibration signal composite characteristic vector;
and step 3: optimizing the GIS vibration signal composite eigenvector by adopting a principal component analysis method to obtain a GIS vibration signal principal component eigenvector;
and 4, step 4: constructing a deep confidence network model, taking multiple groups of GIS vibration signal principal component feature vectors in known states as training samples of the deep confidence network model, and obtaining a corresponding decision function between the GIS vibration signal principal component feature vectors and the GIS states through two-stage training of the deep confidence network model;
and 5: collecting GIS vibration signals to be state recognized, calculating and analyzing principal component eigenvectors, classifying and recognizing the GIS vibration signals by using corresponding decision functions between the principal component eigenvectors and the GIS states, and recognizing the GIS states;
the step 3 specifically includes:
step 3.1: forming m groups of GIS vibration signals into an m multiplied by 14 matrix
X＝(α1,α2，...αm)T＝(x1,x2，...x14)
Step 3.2: the covariance matrix C of matrix X is calculated according to the following expression:
Figure FDA0002581481590000011
in the formula: cov (x, y) denotes the covariance of the two sets of data;
step 3.3: calculating an eigenvalue λ of the covariance matrix Ci(i 1,2.. 14) and a corresponding eigenvector matrix E, arranging the obtained eigenvalues in a descending order, rearranging the columns in the eigenvector matrix E according to the order to obtain a transition matrix T, and taking each column vector in the T as an eigen factor;
step 3.4: calculating the contribution rate of each characteristic factor, selecting 2 characteristic factors of which the sum of the contribution rates exceeds 95% to form a transformation matrix, wherein the calculation formula of the contribution rate of each characteristic factor is as follows:
Figure FDA0002581481590000012
in the formula: krDenotes the contribution of the r-th characteristic factor, λrDenotes the characteristic value, λ, corresponding to the r-th characteristic factorjRepresenting the characteristic value corresponding to the jth characteristic factor,
Figure FDA0002581481590000021
represents a pair ofjFrom λ1To lambdamSumming;
step 3.5: and obtaining an m × 2 feature matrix Y by matrix operation Y ═ X × U, wherein each column of the matrix Y is a principal component feature vector of a group of GIS vibration signals.
2. The GIS state recognition method based on vibration signal principal component analysis according to claim 1, characterized in that: the GIS vibration signal time domain characteristics extracted in the step 2 comprise a GIS vibration signal peak value, an average value, skewness and kurtosis; the frequency domain characteristics comprise a principal component frequency, a 100Hz ratio and a 50Hz odd-order frequency multiplication ratio; and decomposing the GIS vibration signal into a plurality of modal components IMF by using a set empirical mode decomposition algorithm, calculating the energy of each IMF, and taking the energy of the first 7 modal components IMF accounting for more than 90% of the original signal energy as the energy characteristic of the GIS vibration signal.
3. The GIS state recognition method based on vibration signal principal component analysis according to claim 2, characterized in that: constructing a GIS vibration signal composite eigenvector alpha (k) according to 14-dimensional eigenvalues extracted from the GIS vibration signal1,k2，...k14)TWherein alpha is a GIS vibration signal composite characteristic vector, k1Is a first dimension feature quantity, k2For the second dimension feature quantity, k is followed14For the fourteenth dimension, the superscript T is (k)1,k2，...k14) The transposing of (1).
4. The GIS state recognition method based on vibration signal principal component analysis according to claim 2, characterized in that: decomposing the GIS vibration signal into a plurality of modal components IMF by using a set empirical mode decomposition algorithm, wherein the calculation formula for calculating the energy of each IMF is as follows:
Figure FDA0002581481590000022
in the formula: rjRepresenting the sum of the energies in each modal component IMF, nIMFRepresents the total amount of data contained in the nth modal component IMF,
Figure FDA0002581481590000023
representing an energy value of each GIS vibration signal data point;
in order to simplify the calculation, 2 norms of each modal component IMF are used to characterize the energy characteristics, and the calculation formula for calculating the energy of each IMF can be simplified as follows:
Figure FDA0002581481590000024
5. the GIS state recognition method based on vibration signal principal component analysis according to claim 1, characterized in that: in step 4, the two stages of training through the deep belief network model are respectively unsupervised pre-training from a low layer to a high layer and supervised fine tuning from the high layer to the low layer, wherein: the first stage is to adopt a greedy algorithm to train each restricted Boltzmann machine RBM unsupervised, after the training of the RBM of the lower layer is finished, the output of the RBM is used as the input of the RBM of the upper layer, and the RBM is trained layer by layer in sequence, so that the characteristics of the higher layer are learned, and the training parameters of each layer are updated continuously; and in the second stage, a final layer of BP network is trained in a supervision mode, the error generated in the first stage is reversely transmitted to each layer of RBM below, and parameters among all RBM layers are finely adjusted according to the transmission result, so that the parameters of the whole deep belief network model are optimal.
CN201910350848.XA 2019-04-28 2019-04-28 GIS state identification method based on vibration signal principal component analysis method Active CN110082082B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910350848.XA CN110082082B (en) 2019-04-28 2019-04-28 GIS state identification method based on vibration signal principal component analysis method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910350848.XA CN110082082B (en) 2019-04-28 2019-04-28 GIS state identification method based on vibration signal principal component analysis method

Publications (2)

Publication Number Publication Date
CN110082082A CN110082082A (en) 2019-08-02
CN110082082B true CN110082082B (en) 2020-11-17

Family

ID=67417339

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910350848.XA Active CN110082082B (en) 2019-04-28 2019-04-28 GIS state identification method based on vibration signal principal component analysis method

Country Status (1)

Country Link
CN (1) CN110082082B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111486043A (en) * 2020-04-24 2020-08-04 华能四川水电有限公司 Lower rack fault diagnosis method based on hydro-turbo generator set runout data
CN112633333A (en) * 2020-12-11 2021-04-09 广州致新电力科技有限公司 Method for identifying partial discharge defects
CN113191437A (en) * 2021-05-07 2021-07-30 国网山西省电力公司电力科学研究院 Transformer mechanical fault detection method based on vibration signal composite eigenvector
CN113378031B (en) * 2021-06-24 2022-06-28 深圳海域信息技术有限公司 WHOIS query method and system based on global network distributed processing
CN114279707A (en) * 2021-12-17 2022-04-05 哈尔滨工业大学 Large-scale rotating equipment spindle state feature extraction method based on multi-domain analysis and principal component analysis

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108663202A (en) * 2018-05-03 2018-10-16 国家电网公司 GIS mechanical failure diagnostic methods based on chaos cuckoo algorithm and system
CN109374270A (en) * 2018-09-19 2019-02-22 国网甘肃省电力公司电力科学研究院 A kind of analysis of GIS abnormal vibrations and mechanical fault diagnosis device and method
CN109447048A (en) * 2018-12-25 2019-03-08 苏州闪驰数控系统集成有限公司 A kind of artificial intelligence early warning system
CN109597342A (en) * 2019-01-16 2019-04-09 郑州轻工业学院 A kind of the sand dredger monitoring device and method of dynamic group net INTELLIGENT IDENTIFICATION
CN109635428A (en) * 2018-12-11 2019-04-16 红相股份有限公司 A kind of GIS mechanical failure diagnostic method based on the analysis of machine performance signal

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108663202A (en) * 2018-05-03 2018-10-16 国家电网公司 GIS mechanical failure diagnostic methods based on chaos cuckoo algorithm and system
CN109374270A (en) * 2018-09-19 2019-02-22 国网甘肃省电力公司电力科学研究院 A kind of analysis of GIS abnormal vibrations and mechanical fault diagnosis device and method
CN109635428A (en) * 2018-12-11 2019-04-16 红相股份有限公司 A kind of GIS mechanical failure diagnostic method based on the analysis of machine performance signal
CN109447048A (en) * 2018-12-25 2019-03-08 苏州闪驰数控系统集成有限公司 A kind of artificial intelligence early warning system
CN109597342A (en) * 2019-01-16 2019-04-09 郑州轻工业学院 A kind of the sand dredger monitoring device and method of dynamic group net INTELLIGENT IDENTIFICATION

Also Published As

Publication number Publication date
CN110082082A (en) 2019-08-02

Similar Documents

Publication Publication Date Title
CN110082082B (en) GIS state identification method based on vibration signal principal component analysis method
US11544917B2 (en) Power electronic circuit fault diagnosis method based on optimizing deep belief network
US11549985B2 (en) Power electronic circuit fault diagnosis method based on extremely randomized trees and stacked sparse auto-encoder algorithm
CN110263172B (en) Power grid monitoring alarm information evenized autonomous identification method
CN111325095B (en) Intelligent detection method and system for equipment health state based on acoustic wave signals
CN106022229B (en) The abnormal behaviour recognition methods with the Back propagation neural networks of self-adaptive enhancement algorithm is extracted based on video motion information characteristics
CN106933778A (en) A kind of wind power combination forecasting method based on climbing affair character identification
CN109974782B (en) Equipment fault early warning method and system based on big data sensitive characteristic optimization selection
CN109271975A (en) A kind of electrical energy power quality disturbance recognition methods based on big data multi-feature extraction synergetic classification
CN112327219A (en) Distribution transformer fault diagnosis method with automatic feature mining and automatic parameter optimization
CN110737976B (en) Mechanical equipment health assessment method based on multidimensional information fusion
CN104091602A (en) Speech emotion recognition method based on fuzzy support vector machine
CN105046376A (en) Reservoir group flood control scheduling scheme optimization method taking index correlation into consideration
CN112952830B (en) Method for predicting transient stability of power system after fault
CN118051827A (en) Power grid fault prediction method based on deep learning
CN115099296A (en) Sea wave height prediction method based on deep learning algorithm
CN114897138A (en) System fault diagnosis method based on attention mechanism and depth residual error network
CN112633315A (en) Electric power system disturbance classification method
CN113780432B (en) Intelligent detection method for operation and maintenance abnormity of network information system based on reinforcement learning
Krishna et al. Efficient genetic-wrapper algorithm based data mining for feature subset selection in a power quality pattern recognition applicationction in a power quality pattern recognition application.
CN112636328A (en) Medium-voltage distribution network equipment load state identification method
CN105116323B (en) A kind of electrical fault detection method based on RBF
Li et al. Fault diagnosis of ZD6 turnout system based on wavelet transform and GAPSO-FCM
CN115270857A (en) Transformer substation fault diagnosis method based on data fusion
Adaminejad et al. An algorithm for power quality events core vector machine based classification

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
TA01 Transfer of patent application right

Effective date of registration: 20200617

Address after: 637000 Nanchong city in Sichuan province Fu Road No. 228

Applicant after: NANCHONG POWER SUPPLY COMPANY, STATE GRID SICHUAN ELECTRIC POWER Co.,Ltd.

Address before: 637000 Nanchong city in Sichuan province Fu Road No. 228

Applicant before: NANCHONG POWER SUPPLY COMPANY, STATE GRID SICHUAN ELECTRIC POWER Co.,Ltd.

Applicant before: SICHUAN University

TA01 Transfer of patent application right
GR01 Patent grant
GR01 Patent grant