CN117471252A - DC fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM - Google Patents
DC fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM Download PDFInfo
- Publication number
- CN117471252A CN117471252A CN202311442981.0A CN202311442981A CN117471252A CN 117471252 A CN117471252 A CN 117471252A CN 202311442981 A CN202311442981 A CN 202311442981A CN 117471252 A CN117471252 A CN 117471252A
- Authority
- CN
- China
- Prior art keywords
- fault
- kelm
- arc
- soa
- model
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000003745 diagnosis Methods 0.000 title claims abstract description 24
- 238000000605 extraction Methods 0.000 title claims abstract description 18
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 41
- 238000001514 detection method Methods 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 22
- 238000010219 correlation analysis Methods 0.000 claims abstract description 4
- 238000004422 calculation algorithm Methods 0.000 claims description 39
- 230000006870 function Effects 0.000 claims description 28
- 238000005457 optimization Methods 0.000 claims description 20
- 238000012360 testing method Methods 0.000 claims description 15
- 239000011159 matrix material Substances 0.000 claims description 14
- 239000013598 vector Substances 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 9
- 230000009467 reduction Effects 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 8
- 238000011156 evaluation Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 241000272168 Laridae Species 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 4
- 230000005484 gravity Effects 0.000 claims description 4
- 230000000737 periodic effect Effects 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000010891 electric arc Methods 0.000 claims description 3
- 238000000513 principal component analysis Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 2
- 238000013459 approach Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000012512 characterization method Methods 0.000 claims description 2
- 230000001186 cumulative effect Effects 0.000 claims description 2
- 230000002431 foraging effect Effects 0.000 claims description 2
- 230000010354 integration Effects 0.000 claims description 2
- 230000009191 jumping Effects 0.000 claims description 2
- 238000013508 migration Methods 0.000 claims description 2
- 230000005012 migration Effects 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 238000004088 simulation Methods 0.000 claims description 2
- 230000006399 behavior Effects 0.000 claims 1
- 238000010276 construction Methods 0.000 claims 1
- 238000005259 measurement Methods 0.000 claims 1
- 238000012216 screening Methods 0.000 claims 1
- 239000000523 sample Substances 0.000 description 11
- 238000002474 experimental method Methods 0.000 description 10
- 230000007547 defect Effects 0.000 description 3
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 2
- 229910052802 copper Inorganic materials 0.000 description 2
- 239000010949 copper Substances 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000009413 insulation Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 206010000369 Accident Diseases 0.000 description 1
- 206010002515 Animal bite Diseases 0.000 description 1
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 241000287127 Passeridae Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000012847 principal component analysis method Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
- G01R31/1263—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation
- G01R31/1272—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation of cable, line or wire insulation, e.g. using partial discharge measurements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
- G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
-
- 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
- G06F18/2131—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on a transform domain processing, e.g. wavelet transform
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the field of direct current fault diagnosis, in particular to a direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM. The direct-current fault arc detection model is constructed, so that the characteristic difference of the direct-current fault arc caused in the actual application scene is analyzed, and the diagnosis of the direct-current fault arc is carried out. 1) Simulating arc faults of the low-voltage direct-current line in an actual application scene, and extracting current signals of the low-voltage direct-current line; 2) Constructing a complete set empirical mode decomposition ICEEMDAN decomposition method based on improved self-adaptive noise to obtain an intrinsic mode component IMF containing fault characteristic information; 3) Performing correlation analysis; 4) Denoising by using a wavelet soft threshold; 5) Extracting fault characteristics; 6) Extracting nonlinear characteristic principal elements of data; 7) Constructing a detection model of a KELM core extreme learning machine of the direct current fault arc; 8) Training a KELM detection model; 9) Optimizing; 10 Constructing a direct current fault arc detection model.
Description
Technical Field
The invention relates to the field of direct current fault diagnosis, in particular to a direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM.
Background
The direct current distributed power generation system has high randomness, so that the duty ratio of the direct current power supply system in a power grid is continuously improved. After the DC power supply system is used for a certain period of time, problems such as joint loosening, poor contact, wire insulation aging, device faults and the like easily occur in a circuit, and then DC fault arc is generated at a contact point. As the direct current power supply does not have zero crossing points, once a fault arc is generated and is not processed timely and effectively, the fault arc can continuously burn and release a large amount of energy to the outside, thereby seriously affecting the normal operation of the direct current power supply and distribution system and even causing electric fire accidents.
The DC system has rich and extensive application scenes and complicated and various lines, so that the occurrence time and the occurrence position of the fault arc have randomness, and the DC fault arc can occur in different application scenes. The 'good arc' of the plug switch, the periodic 'good arc' of the brush motor, the topology structure of other load changing circuits connected to the circuit in the normal state and the types of different loads in the circuit are generated, and the different positions of the fault arc can cause interference to the fault arc detection characteristic.
The method is important to research the influence on the fault arc detection characteristics in the actual application scene and pertinently propose a fault arc detection algorithm suitable for the actual application scene, and can further improve the application range and the detection accuracy of direct current fault arc detection, which also belongs to the aspect that related research has not been carried out at home and abroad.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention provides a direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM. The invention analyzes the direct-current fault arc characteristic difference caused in the actual application scene by constructing a direct-current fault arc detection model, and provides a direct-current fault arc diagnosis method based on ICEEMDAN-KPCA characteristic extraction and SOA-KELM.
In order to achieve the purpose, the invention adopts the following technical scheme that the method comprises the following steps of 1, constructing a direct current series fault arc experimental platform. Wherein: the output range of the main loop DC power supply current is 0-20A, the voltage output range is 0-300V, the stepping motor is controlled by a DQ-2HD542 driver and a TC-5510 controller, the stepping motor drives a screw rod to control a movable electrode to move, the minimum movement speed is 0.1mm/s, a DC voltage source is a 24V regulated voltage source, a stationary electrode is a carbon rod, the diameter of the stationary electrode is 6.35mm, the movable electrode is a copper rod, the diameter of the movable electrode is 6.4mm, the front sharpening 19mm accords with the standard UL1699B, and the DC series arc fault voltage and current are measured by the device to explore the characteristics of the movable electrode. The experiment also simulates the fact that direct current fault arc is caused by the damage of the lead, the two electrodes are copper bars, and the contact surfaces are cutting surfaces.
And 2, simulating arc faults of the low-voltage direct current lines in an actual application scene, and extracting current signals of the low-voltage direct current lines in different types. Wherein: in order to study fault arc detection characteristics under different conditions, a direct current fault arc experiment loop is built. The experimental loop is mainly powered by a direct-current power supply, and fault arcs are simulated and generated in an arc-striking mode through fault arc generating devices connected in series in the loop. The arc discharge simulates an arc generating gap which is continuously increased in the fault arc generating process. As insulation ages and the degree of loosening of the connection terminals increases, visible arc gaps may suddenly appear during operation of the dc system due to a decrease in contact pressure. Accidental external damage factors such as animal bite or environmental impact may also introduce visible arc gaps into the dc system. In order to enable the collected direct current series fault arc data to have diversity and universality, the experiment considers not only the 'good arc' of the plug switch, the periodic 'good arc' of the brush motor, the influence of other loads connected into the circuit in a normal state on loop current, the type of the load in the circuit, but also the influence of different positions of fault arc on the current in the circuit. The invention adopts a combined parallel model of the direct current motor, the direct current fan and the direct current lamp to simulate the actual application scene, thereby collecting different types of current data.
And 3, constructing a complete set empirical mode decomposition (ICEEMDAN) decomposition technology based on improved self-adaptive noise, and performing data decomposition and filtering on the arc signals to obtain an intrinsic mode component IMF containing fault characteristic information.
Aiming at the characteristic of nonlinear and non-stable direct-current fault arc current signals, EEMD, CEEMDAN and ICEEMDAN decomposition processing is respectively carried out on 6 types of original signals of normal operation, resistance load fault arc, resistance-inductance load fault arc, motor load fault arc and branch fault arc. The three parameters of the algorithm, noise amplitude, total integration and overall iteration number, are set to 0.2, 100, 500, respectively. The IMF components obtained by decomposition represent frequency components in the original signal respectively, and are sequentially arranged in order from high frequency to low frequency, and signal characteristics in the original arc fault are reflected from different angles.
The EMD algorithm and the EEMD algorithm have the defects of large reconstruction error, modal aliasing and the like, therefore, a CEEMDAN algorithm for complete empirical mode decomposition of self-adaptive noise is provided, self-adaptive non-Gaussian white noise is added, the modal aliasing is further reduced, the reconstruction error is greatly reduced, partial residual noise is still contained, and a pseudo component is generated due to delay in the decomposition process. Based on this study, an ICEEMDAN algorithm was proposed to further attenuate residual noise. Different from the conventional mode of directly adding Gaussian white noise, the improved algorithm defines the modal component as the difference between the residual quantity and the local mean value, and adds special white noise to the kth IMF function containing Gaussian white noise EMD decomposition, so that the residual noise is greatly reduced.
The ICEEMDAN carries out EMD decomposition on Gaussian white noise, and then adds an IMF component obtained after the decomposition into an original signal x (t); and then calculating to obtain a specific residual error, calculating the average value of the specific residual error, and defining the IMF as the difference between the existing residual error signal and the local average value of the specific residual error. Define a (·) as the remaining component after EMD, ek (·) as the kth IMF using the EMD output, and ζi (i=1, 2, …, I) as a set of gaussian white noise. The decomposition process of ICEEMDAN is described below.
(1) Adding I noise components into the original harmonic signal x (t) to obtain
X i (t)＝x(t)+β 0 E 1 (ξ i (t)),i＝1,2,...,I
Wherein: β0—raw signal to noise ratio.
(2) Solving a first order residual component of the signal to obtain a first residual:
(3) By r 1 (t) obtaining a first intrinsic mode function IMF1, which is specifically as follows:
c 1 (t)＝x(t)-r 1 (t)
(4) At r 1 Adding I Gaussian white noise to (t) to obtain a second residual error r of the signal 2 (t); and according to c 2 (t)＝r 1 (t)-r 2 (t) obtaining a second intrinsic mode function IMF2, which is specifically as follows:
wherein: beta 1 Signal to noise ratio of the first stage.
(5) Similarly, it can be according to formula c k (t)＝r k-1 (t)-r k (t) determining the value of the remaining IMF component, wherein
Wherein: beta k-1 Signal to noise ratio at stage k-1, k=3, 4,5, ….
(6) And (5) continuously repeating the step (5) until the residual error becomes a monotonic function or the standard deviation of the adjacent IMF is less than 0.2, and stopping iteration.
(7) Finally, IMF components of each order are obtained, and the original signal can be expressed as
And 4, calculating the correlation coefficient and the variance contribution rate between the IMF and the original signal respectively, and carrying out correlation analysis on the IMF and the original signal.
The correlation coefficient between each order of modal components obtained by decomposition and the original signal is used as an index for measuring the correlation degree between data, and real and meaningful components can be screened and judged to be used as sensitive IMF components. The correlation coefficient is the ratio of the covariance between the two time sequences and the standard deviation of each time sequence, so that the misjudgment caused by interference to a final result due to the difference of the magnitudes of the two signals is eliminated. Let the original signal be X and the components of each order be Y i The correlation coefficient between each order component and the original signal is:
the absolute value range of the correlation coefficient is [0,1], and the larger the correlation coefficient is, the stronger the correlation between the two signals is, which means that the IMF component obtained by decomposition contains more original characteristic values; conversely, the smaller the correlation coefficient, the less false the IMF component can be determined to be.
The variance contribution rate is the ratio of the IMF variance to the original sequence variance, and the variance contribution rate is larger, which means that the influence of the component on the original data is larger than that of other components.
And 5, taking the signal with high correlation as a target signal, and denoising the target signal by using a wavelet soft threshold.
Wavelet decomposition is a reduced order decomposition of the target data by using the Mallat tower algorithm. The Mallat tower algorithm may decompose the data into an approximation component and a detail component. The low frequency information is stored in the approximation component, representing the high scale of the signal; high frequency information is stored in the detail component, representing the low scale of the signal. When a noise signal is present in the processed signal, the main component of the decomposed noise is stored in the detail component. The wavelet threshold denoising method is to screen out the noise part of the target signal to suppress and strengthen the effective part. Wavelet threshold denoising can be simply divided into three steps: the first step is wavelet transformation decomposition, namely selecting a wavelet to carry out n layers of wavelet decomposition on signals; the second part is threshold processing, namely, threshold processing is carried out on each decomposed coefficient of the layers to obtain an estimated wavelet coefficient; and thirdly, carrying out wavelet inverse transformation, namely carrying out wavelet reconstruction on the effective signals to obtain denoised signals.
The threshold value suitable for processing the target signal is selected, so that the characteristic quantity in the target signal can be better reserved, and the purpose of the threshold function is to filter the wavelet coefficient containing noise. The invention selects unbiased risk estimation threshold and soft threshold function. The formula is as follows:
wherein,as a soft threshold function, w represents the signal after wavelet decomposition, and σ represents the wavelet threshold selected in the previous step.
And 6, extracting fault characteristics, respectively calculating time domain statistical characteristics and frequency domain statistical characteristics, simultaneously calculating power entropy of each IMF component, and constructing a multi-domain high-dimensional fault characteristic set.
The time domain, frequency domain and time domain characteristics of the fault arc signal can be used for describing the fault state from different angles, and the formula is concise and convenient to calculate, and can be used as an initial characteristic for describing the fault state. Meanwhile, the power entropy is used for measuring the complexity of the time sequence, so that the power entropy of each IMF component is calculated, and a multi-domain high-dimensional fault feature set is constructed.
Taking a current fault signal as an example, calculating the multi-domain fault characteristic indexes by using an ICEEDAN method for each decomposed IMF component, and finally calculating the principal component contribution rate of the multi-domain characteristics by using a KPCA algorithm to screen out the fused nuclear principal component fault characteristics. The formula is as follows:
variance:
average value:
kurtosis:
peak-to-peak value: x is X p-p ＝max(x i )-min(x i )
Skewness factor:
kurtosis factor:
maximum value: x is X max ＝max{|x i |}
Absolute average:
root mean square value:
peak factor:
waveform factor:
minimum value: x is X min ＝min{x i }
Skew degree:
margin factor:
pulse factor:
power entropy of each IMF component:
and 7, mapping fault sample data to a high-dimensional space based on KPCA, eliminating the space correlation and redundant data generated on different fault characteristics due to irradiance temperature and the like, further extracting nonlinear characteristic principal elements of the data, and dividing the acquired data into a training data set and a test data set. Wherein:
(1) KPCA maps the original fault sample data to a high dimensional space phi to form new data phi (e i )＝{φ(e 1 ),φ(e 2 ),...,φ(e n ) I=1, 2,..n. Assuming that the samples have been centered in the high dimensional space, the covariance matrix is:
(2) Introducing a kernel function of K * ＝φ T And phi, carrying out principal component analysis and solving on the data in S:
K * η＝λη
wherein: lambda is a characteristic value; η is the eigenvector.
(3) Setting the cumulative contribution rate as 95%, and taking the first s eigenvalues lambda in descending order j (j=1, 2,., s) and its corresponding feature vector η j (j＝1,2,...,s)：
(4) When the accumulated contribution rate reaches the set requirement, calculating a nonlinear sample G after dimension reduction mapping:
and 8, constructing a KELM direct current fault arc detection model based on a seagull optimization algorithm. And 8.1, constructing a detection model of the KELM core extreme learning machine of the direct current fault arc.
In ELM algorithm:
wherein: x is the given training sample input; f (x) is the actual output of the network; h (x) is a sample matrix; h is a network hidden layer output matrix; ρ is the connection weight vector of the hidden layer and the output layer; t is a training sample class vector matrix; i is a diagonal matrix; c is a regularization coefficient.
Defining a KELM kernel matrix Ω ELM Replacing HH T The formula is as follows:
wherein: x is x i And x j For the input samples.
K(x i ,x j ) Selected as an RBF kernel function:
wherein: gamma is the nuclear parameter.
The output of the KELM model is therefore:
and 8.2, training the KELM detection model.
(1) Firstly, determining the number of hidden layer neurons, and randomly determining the connection weight omega between an input layer and the hidden layer and the value of the threshold value b of the hidden layer neurons.
(2) The activation function of hidden layer neurons is chosen, which must be infinitely differentiable, and H is found.
(3) Calculating the weight of the output layerWherein->A least squares solution for the minimum norm of the set of hβ=t equations; h+ is the Moore-Pen-rose generalized inverse of the hidden layer output matrix H.
(4) Updating parameters, and sequentially iterating until convergence conditions are met, so that training of the KELM model is completed. And inputting the test set data with the labels removed into the trained model for recognition.
And 8.3, optimizing the KELM detection model parameters through an SOA seagull optimization algorithm.
In order to avoid collision among different seagulls in the migration process and approach of all the seagulls to the direction of the optimal seagull position, the updating of the seagull position is realized by the following formula.
In the formula D S (t) is a new position updated by the seagull; p (P) S (t) is the current position of the seagull; c (C) S (t) is a new location where no collision occurs; p (P) best (t) is the optimal gull position; m is M S (t) is the direction in which the optimal position of the seagull is located; t is the iteration number; a is an additional variable used for simulating the motion of seagulls in a search space, and the size of the additional variable is controlled by the following formula; b is a random parameter for the local and global search of the balance algorithm, and the size is controlled by the following formula.
B＝2×A 2 ×rd
In which t is MAX The maximum iteration number; f (f) c The variation from 2 to 0 is linear as a function of the number of iterations. The seagull produces an attack action when foraging, the angle and the flight spiral radius can be continuously changed, and the position of the seagull in the three-dimensional space is updated by the following formula.
Wherein: k is [0,2 pi ]]Random angle values within the range; r is the gull spiral radius, r=u×e kv U and v are correlation constants of a spiral shape, and e is a base of natural logarithm.
And 8.4, constructing a KELM direct current fault arc detection model based on a seagull optimization algorithm.
(1) Setting the upper and lower boundaries of the SOA model. The upper boundary and the lower boundary of the SOA model are set through the number of the weight values and the threshold values in the KELM model, so that the optimal initial weight values and the optimal threshold values can be obtained when the KELM model is calculated each time, the convergence speed of calculated data is improved, and the model detection and identification precision is improved. And (3) carrying out ICEEMDAN signal decomposition on the original data, and extracting a plurality of modal components such as a time domain, a frequency domain, an IMF and the like.
(2) And carrying out standard normalization processing on the feature data, carrying out dimension reduction on the feature data set by using KPCA, extracting sample feature main components, and dividing a training set and a testing set.
(3) Initializing parameters of a KELM model and an SOA algorithm, and setting the number of gull population, the maximum iteration number, the upper and lower limits of independent variables and the dimension.
(4) And calculating the fitness value of each seagull, continuously updating the seagull position by the equation (11), calculating the fitness value of the new position, comparing with the fitness value before updating, and storing smaller fitness value as the current optimal value.
(5) If the maximum iteration number of the seagull algorithm is reached, the optimal parameters are given to the KELM and training is carried out, and the predicted value of each component is output; otherwise, jumping to the step (4).
(6) And updating the fault identification model by using the KELM optimal super parameters.
Setting the iteration times, population number, initial position, initial search rate and the like of the model. Assuming that the number of neurons in the hidden layer is l and the number of elements contained in each feature vector to be trained is R during KELM training and learning, d in the SOA model is:
d＝l×R+l。
and setting a fitness function. Because the SOA-KELM model is used for detecting and identifying the fault arc and the detection and identification accuracy of the model is improved by optimizing parameters, the fitness function is designed as the error rate (P T ) And error rate (P) generated by SOA-KELM model during test C ) The sum can enable the SOA-KELM model to have more accurate results in training data and sample testing. The fitness function may be expressed as:
F＝argmin(P T +P C )。
and 9, evaluating the fault model.
For classification problems, commonly used evaluation criteria include Accuracy (Accuracy), precision (Precision), recall (Recall), and comprehensive evaluation index.
The calculation methods of the above four evaluation criteria are defined as follows:
in this test, F is introduced 1 Score concept to characterize anti-interference ability of diagnostic model, F 1 Score is a comprehensive characterization of the accuracy and recall of the model, and can more fully evaluate the performance of the diagnostic model.
Comprehensive evaluation index F 1 The formula of Score is:
p in the formula c For the precision, R z Is the recall rate.
For the series fault arc detection problem, the invention selects the basic accuracy as an index and simultaneously examines other two indexes. Precision indicates the specific gravity of the actual fault arc predicted to be in a series fault, an indicator that is particularly important for data sets with unbalanced sample distributions. Recall represents the specific gravity of the predicted series fault arc in the actual series fault arc.
Compared with the prior art, the invention has the beneficial effects.
According to the invention, the arc generating device is designed by constructing a low-voltage direct-current arc fault data acquisition experimental platform, so that accurate acquisition of loop current and voltage data at two ends of the arc generator is realized. In order to solve the problems of insufficient accuracy and poor universality of the traditional arc fault detection method in a low-voltage direct-current system to practical application scenes, the invention provides a direct-current series fault arc diagnosis method based on improved adaptive noise complete set empirical mode decomposition (ICEEMDAN) -Kernel Principal Component Analysis (KPCA) and Haiou optimization algorithm (SOA) optimization Kernel Extreme Learning Machine (KELM). Firstly, a direct current series fault arc experiment platform containing a mixed load is built, an actual application scene is simulated, current signals under multiple working conditions are collected, and a fault arc database is built. And secondly, calculating fault characteristics of IMF components such as time domain, frequency domain, power entropy and the like decomposed by ICEEMDAN, extracting nonlinear characteristics of fault data by using KPCA dimension reduction, and reducing redundant data generated by external conditions. And optimizing the nuclear parameter gamma and the regularization coefficient C in the KELM through an SOA optimization algorithm, and finally inputting the fault characteristics after the dimension reduction into an SOA-KELM model through KPCA, so as to realize the identification of the direct current fault arc in the actual application scene.
Drawings
The invention is further described below with reference to the drawings and the detailed description. The scope of the present invention is not limited to the following description.
Fig. 1 is a schematic block diagram of an arc fault experimental circuit.
Fig. 2 is a KPCA dimension reduction flow chart.
Fig. 3 is a diagram of the decomposition results of icemdan.
Fig. 4 shows the contribution rates of the features after the KPCA dimension reduction.
FIG. 5 is an SOA optimization KELM flow.
FIG. 6 is a KELM fault arc identification model test result.
FIG. 7 is a test result of an SOA-KELM fault arc identification model.
FIG. 8 is a graph of fitness curves for different algorithm models.
Detailed Description
1-8, in order to make the objects, technical solutions and advantages of the present invention more apparent, the following detailed description of the low-voltage direct current arc fault detection method based on KPCA-SOA-KELM will be made with reference to specific embodiments and accompanying drawings.
(1) And building a direct current series fault arc experimental platform.
And manufacturing a direct current series fault arc generator according to UL1699-2018AFCI standard, building an experimental platform, and acquiring data required by the direct current series fault arc through an oscilloscope and a voltage and current probe.
In order to enable the collected direct current series fault arc data to be more practical, the practical application scene is simulated through a combined parallel model of a direct current lamp, a direct current fan and a direct current motor in an experiment, and the direct current series fault arc data under different conditions are collected.
(2) And simulating arc faults of the low-voltage direct current lines in an actual application scene, and extracting current signals of the low-voltage direct current lines in different types.
The direct current series fault arc signals are collected through experiments, so that the collected direct current series fault arc data have diversity and universality, the effects of good arcs of the plug switch, periodic good arcs of the brush motor, the influence of other loads connected into the circuit on loop current in a normal state, the types of loads in the circuit are increased, and the influence of different positions of fault arcs on the current in the circuit are considered in the experiments.
(3) And (3) carrying out ICEEMDAN decomposition on the current fault signal, respectively calculating a time domain feature index, a frequency domain feature index and a time domain feature index, and constructing a high-dimensional fault feature.
And mapping fault sample data to a high-dimensional space based on KPCA, eliminating the space correlation and redundant data generated on different fault characteristics due to irradiance temperature and the like, further extracting nonlinear characteristic principal elements of the data, and dividing the acquired data into a training data set and a test data set.
And the direct current series fault arc signals are acquired through experiments, and the characteristic extraction is carried out on the direct current fault arc data by using a nuclear principal component analysis method, so that the data dimension is reduced.
(4) And constructing a KELM direct current fault arc detection model based on a seagull optimization algorithm.
The invention discloses a KELM direct-current fault arc detection model based on a seagull optimization algorithm.
The KELM model is optimized through the SOA seagull optimization algorithm, so that the problems that the KELM identification model is low in convergence speed and easy to sink into a local optimal solution are solved, and the accuracy of KELM model identification is improved.
Specific embodiments are given below:
the invention adopts the current widely used arc fault simulation experiment and design as the basis, and refers to the UL1699-2018AFCI standard to manufacture the direct current series fault arc generator and build an experimental platform, so as to collect the electric signals of the circuit under different conditions as the characteristic values, improve the accuracy of data collection, and provide an experimental data basis for the extraction of the characteristic vectors of the follow-up fault arc.
The direct current series fault arc signals of various electric lines are collected through experiments, nonlinear characteristics of fault data are extracted through KPCA dimension reduction, redundant data generated by external conditions are reduced, and the accuracy rate of complex fault identification is effectively improved.
The contribution rates of the characteristics after the KPCA is reduced in dimension are shown in figure 4.
The fault arc feature vector extracted by the method solves the problem of interference generated by redundant data, ensures the integrity of the feature vector, has higher identification degree, and provides favorable conditions for the identification and classification of the follow-up model.
And finally, detecting, identifying and researching the direct current series fault arc by using a seagull optimization algorithm optimization nuclear extreme learning machine.
Compared with other recognition algorithms, the recognition model based on ICEEMDAN-KPCA and SOA-KELM constructed by the invention can accurately and rapidly recognize the direct current fault arc, and is more suitable for detecting and recognizing the direct current series fault arc in a direct current system.
The nuclear extreme learning machine model can obtain higher accuracy rate when used for fault diagnosis, but has some defects: the setting of the KELM part parameters has a great influence on the overall diagnosis accuracy; the optimization algorithm used in the partial improvement model, such as particle swarm optimization algorithm and sparrow search algorithm, has the problems of low convergence speed, local optimum sinking and the like.
Table 1 comparison of test results for different models
Therefore, the invention provides a method for optimizing a kernel extreme learning machine based on a seagull algorithm. Aiming at the problem that the nuclear parameters and regularization coefficients have great influence on the identification result of the nuclear extreme learning machine, a seagull optimization algorithm is introduced to perform parameter optimization, and then a fault diagnosis model based on SOA-KELM is established. And the results of experiments are compared with the KELM, PSO-KELM and SSA-KELM models, so that the SOA-KELM model has higher convergence iteration speed and higher fault diagnosis accuracy, and can effectively perform direct current fault arc diagnosis under multiple scenes.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art will understand that; the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced with equivalents; such modifications and substitutions do not depart from the spirit of the invention, which is defined by the following claims.
Claims (9)
1. The DC fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM is characterized in that:
1) Simulating arc faults of the low-voltage direct-current line in an actual application scene, and extracting current signals of the low-voltage direct-current line;
2) Constructing a complete set empirical mode decomposition ICEEMDAN decomposition method based on improved self-adaptive noise, and carrying out data decomposition and filtering on an arc signal to obtain an intrinsic mode component IMF containing fault characteristic information;
3) Calculating the correlation coefficient and variance contribution rate between the IMF and the original signal respectively, and carrying out correlation analysis on the IMF and the original signal;
4) Taking the signal with high correlation as a target signal, and denoising by using a wavelet soft threshold;
5) Extracting fault characteristics, respectively calculating time domain statistical characteristics and frequency domain statistical characteristics, simultaneously calculating power entropy of each IMF component, and constructing a multi-domain high-dimensional fault characteristic set;
6) Mapping fault sample data to a high-dimensional space based on KPCA, eliminating space correlation and redundant data generated by irradiance temperature on different fault characteristics, further extracting nonlinear characteristic principal elements of the data, and dividing acquired data into a training data set and a test data set;
7) Constructing a detection model of a KELM core extreme learning machine of the direct current fault arc;
8) Training a KELM detection model;
9) Optimizing the KELM detection model parameters through an SOA seagull optimization algorithm;
10 A KELM direct current fault arc detection model based on a seagull optimization algorithm is constructed;
11 A) evaluating the fault model.
2. The direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM as claimed in claim 1, wherein the method comprises the following steps: the simulating the arc fault of the low-voltage direct current circuit in the actual application scene comprises the following steps:
simulating to generate a fault arc in an arc discharge mode, wherein the arc discharge simulation is an arc generation gap which is continuously increased in the fault arc generation process; the fault arc data not only considers the 'good arc' of the plug switch, the periodic 'good arc' of the brush motor, the influence of other loads connected into the circuit in a normal state on the loop current, the type of the load in the circuit is increased, but also considers the influence of different positions of the fault arc on the current in the circuit.
3. The direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM as claimed in claim 1, wherein the method comprises the following steps:
the construction is based on a complete set empirical mode decomposition ICEEMDAN decomposition method for improving self-adaptive noise, and the data decomposition and filtering of the arc signals are carried out, so that an intrinsic mode component IMF containing fault characteristic information is obtained, and the method comprises the following steps:
defining the modal component as a difference value between the residual quantity and the local mean value, and adding white noise to a kth IMF (inertial measurement unit) function containing Gaussian white noise EMD (empirical mode decomposition) so as to greatly reduce the residual noise;
respectively carrying out EEMD, CEEMDAN and ICEEMDAN decomposition treatment on the original signals in 6 states of normal operation, resistance load fault arc, resistance inductance load fault arc, motor load fault arc and branch fault arc; setting the noise amplitude, the total integration number and the overall iteration number of three parameters of the algorithm to 0.2, 100 and 500 respectively; the IMF components obtained through decomposition represent frequency components in the original signal respectively, are sequentially arranged according to the sequence from high frequency to low frequency, and reflect signal characteristics in the original arc fault from different angles;
the ICEEMDAN carries out EMD decomposition on Gaussian white noise, and then adds an IMF component obtained after the decomposition into an original signal x (t); then a specific residual error is obtained through calculation, the average value of the specific residual error is obtained, and IMF is defined as the difference between the existing residual error signal and the local average value of the specific residual error; define A (·) as solving for EMDResidual component after, E k (. Cndot.) is the kth IMF, ζ using EMD output i (i=1, 2, …, I) is a set of gaussian white noise;
the decomposition process of ICEEMDAN is as follows:
(1) Adding I noise components into the original harmonic signal x (t) to obtain
X i (t)＝x(t)+β 0 E 1 (ξ i (t)),i＝1,2,...,I
Wherein: beta 0 -original signal to noise ratio;
(2) Solving a first order residual component of the signal to obtain a first residual:
(3) By r 1 (t) obtaining a first intrinsic mode function IMF1, specifically:
c 1 (t)＝x(t)-r 1 (t)
(4) At r 1 Adding I Gaussian white noise to (t) to obtain a second residual error r of the signal 2 (t); and according to c 2 (t)＝r 1 (t)-r 2 (t) obtaining a second intrinsic mode function IMF2, specifically:
wherein: beta 1 -signal-to-noise ratio of the first stage;
(5) Similarly, according to formula c k (t)＝r k-1 (t)-r k (t) determining the value of the remaining IMF component, wherein
Wherein: beta k-1 -signal to noise ratio at stage k-1, k=3, 4,5, …;
(6) Repeating (5) until the residual error becomes a monotonic function or the standard deviation of the adjacent IMF is less than 0.2, and stopping iteration;
(7) Finally, IMF components of each order are obtained, and the original signal can be expressed as
4. The direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM as claimed in claim 1, wherein the method comprises the following steps: the calculating of the correlation coefficient and variance contribution rate between the IMF and the original signal respectively, and the correlation analysis of the IMF and the original signal comprises the following steps:
the correlation coefficient between each order of modal components obtained by decomposition and the original signal is used as an index for measuring the correlation degree between data, and real and meaningful components can be screened and judged to be used as sensitive IMF components; the correlation coefficient is the ratio of covariance between two time sequences and the standard deviation of each time sequence, so that misjudgment caused by interference to a final result due to the difference of two signal amplitude magnitudes is eliminated; let the original signal be X and the components of each order be Y i The correlation coefficient between each order component and the original signal is:
the absolute value range of the correlation coefficient is [0,1], and the larger the correlation coefficient is, the stronger the correlation between the two signals is, which means that the IMF component obtained by decomposition contains more original characteristic values; conversely, the smaller the correlation coefficient, the less false the IMF component can be determined to be;
the variance contribution rate is the ratio of the IMF variance to the original sequence variance, and the variance contribution rate is larger, which means that the influence of the component on the original data is larger than that of other components.
5. The direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM as claimed in claim 1, wherein the method comprises the following steps: the denoising method using the wavelet soft threshold value includes:
wavelet threshold denoising is divided into three steps:
the first step is wavelet transformation decomposition, namely selecting a wavelet to carry out n layers of wavelet decomposition on signals;
the second part is threshold processing, namely, threshold processing is carried out on each decomposed coefficient of the layers to obtain an estimated wavelet coefficient;
thirdly, carrying out wavelet inverse transformation, namely carrying out wavelet reconstruction on the effective signals to obtain denoised signals;
selecting a threshold value suitable for processing the target signal, reserving characteristic quantity in the target signal, and filtering wavelet coefficients containing noise by a threshold function; therefore, an unbiased risk estimation threshold value and a soft threshold function are selected; the formula is:
wherein,as a soft threshold function, w represents a wavelet decomposed signal, and σ represents a wavelet threshold.
6. The direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM as claimed in claim 1, wherein the method comprises the following steps: the fault feature extraction, respectively calculating a time domain statistical feature and a frequency domain statistical feature, simultaneously calculating the power entropy of each IMF component, and constructing a multi-domain high-dimensional fault feature set comprises the following steps:
calculating the multi-domain fault feature indexes of each decomposed IMF component by using an ICEEDAN method, and finally calculating the principal component contribution rate of the multi-domain features by using a KPCA algorithm, and screening out the fused nuclear principal component fault features; the formula is as follows:
variance:
average value of：
Kurtosis:
peak-to-peak value: x is X p-p ＝max(x i )-min(x i )
Skewness factor:
kurtosis factor:
maximum value: x is X max ＝max{|x i |}
Absolute average:
root mean square value:
peak factor:
waveform factor:
minimum value: x is X min ＝min{x i }
Skew degree:
margin factor:
pulse factor:
power entropy of each IMF component:
7. the direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM as claimed in claim 1, wherein the method comprises the following steps: the step of mapping the fault sample data to a high-dimensional space based on KPCA, eliminating the space correlation and redundant data generated by irradiance temperature on different fault characteristics, further extracting nonlinear characteristic principal elements of the data, and dividing the acquired data into a training data set and a test data set comprises the following steps:
(1) KPCA maps the original fault sample data to a high dimensional space phi to form new data phi (e i )＝{φ(e 1 ),φ(e 2 ),...,φ(e n ) I=1, 2,..n; assuming that the samples have been centered in the high dimensional space, the covariance matrix is:
(2) Introducing a kernel function of K * ＝φ T And phi, carrying out principal component analysis and solving on the data in S:
K * η＝λη
wherein: lambda is a characteristic value; η is a feature vector;
(3) Setting the cumulative contribution rate as 90%, and taking the first s eigenvalues lambda in descending order j (j=1, 2,., s) and its corresponding feature vector η j (j＝1,2,...,s)：
(4) When the accumulated contribution rate reaches the set requirement, calculating a nonlinear sample G after dimension reduction mapping:
8. the direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM as claimed in claim 1, wherein the method comprises the following steps: the detection model for constructing the KELM core extreme learning machine of the direct current fault arc comprises the following steps:
in ELM algorithm:
wherein: x is the given training sample input; f (x) is the actual output of the network; h (x) is a sample matrix; h is a network hidden layer output matrix; ρ is the connection weight vector of the hidden layer and the output layer; t is a training sample class vector matrix; i is a diagonal matrix; c is regularization coefficient;
defining a KELM kernel matrix Ω ELM Replacing HH T The formula is as follows:
wherein: x is x i And x j For an input sample;
K(x i ,x j ) Selected as an RBF kernel function:
wherein: gamma is a nuclear parameter;
the output of the KELM model is therefore:
9. the direct current fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM as claimed in claim 1, wherein the method comprises the following steps:
the training of the KELM detection model includes:
(1) Firstly, determining the number of hidden layer neurons, and randomly determining the connection weight omega between an input layer and a hidden layer and the value of a threshold value b of the hidden layer neurons;
(2) Selecting an activation function of hidden layer neurons, wherein the function is infinitely differentiable, and solving H;
(3) Calculating the weight of the output layer
Wherein,a least squares solution for the minimum norm of the set of hβ=t equations; h+ is Moore-Penrose generalized inverse of hidden layer output matrix H;
(4) Updating parameters, sequentially iterating until convergence conditions are met, and finishing training of the KELM model; inputting the test set data with the labels removed into a trained model for identification;
the optimizing the KELM detection model parameters through the SOA seagull optimization algorithm comprises the following steps:
in order to avoid collision among different seagulls in the migration process and approach of all the seagulls to the direction of the optimal seagull position, updating of the seagull position is realized through the following formula;
in the formula D S (t) is a new position updated by the seagull; p (P) S (t) is the current position of the sea gullPlacing; c (C) S (t) is a new location where no collision occurs; p (P) best (t) is the optimal gull position; m is M S (t) is the direction in which the optimal position of the seagull is located; t is the iteration number; a is an additional variable used for simulating the motion of seagulls in a search space, and the size of the additional variable is controlled by the following formula; b is a random parameter for local and global searching of the balance algorithm, and the size of the random parameter is controlled by the following formula;
B＝2×A 2 ×rd
wherein: t is t MAX The maximum iteration number; f (f) c A variable which is changed according to the iteration times is linearly reduced from 2 to 0; the seagull produces an attack behavior during foraging, the angle and the flight spiral radius can be continuously changed, and the position of the seagull in the three-dimensional space is updated through the following steps;
wherein: k is [0,2 pi ]]Random angle values within the range; r is the gull spiral radius, r=u×e kv U and v are correlation constants of spiral shape, and e is a base number of natural logarithm;
the KELM direct current fault arc detection model based on the seagull optimization algorithm comprises the following steps:
(1) Setting upper and lower boundaries of an SOA model; the upper boundary and the lower boundary of the SOA model are set according to the number of the weight and the threshold in the KELM model, so that the optimal initial weight and the optimal initial threshold can be obtained when the KELM model is calculated each time; carrying out ICEEMDAN signal decomposition on the original data to extract time domain, frequency domain and IMF components;
(2) Performing standard normalization processing on the feature data, performing dimension reduction on the feature data set by using KPCA, extracting sample feature main components, and dividing a training set and a testing set;
(3) Initializing parameters of a KELM model and an SOA algorithm, and setting the number of gull population, the maximum iteration times, the upper and lower limits of independent variables and the dimension;
(4) Calculating the fitness value of each seagull, continuously updating the seagull position, calculating the fitness value of a new position, comparing the fitness value with the fitness value before updating, and storing smaller fitness value as the current optimal;
(5) If the maximum iteration number of the seagull algorithm is reached, the optimal parameters are given to the KELM and training is carried out, and the predicted value of each component is output; otherwise, jumping to the step (4);
(6) Updating the fault identification model by using the KELM optimal super parameters;
setting the iteration times, population number, initial positions and initial search rate of a model, and assuming that the number of neurons of an hidden layer is l and the number of elements contained in each trained feature vector is R during KELM training and learning, d in the SOA model is:
d＝l×R+l
setting a fitness function; because the SOA-KELM model is used for detecting and identifying the fault arc and the detection and identification accuracy of the model is improved by optimizing parameters, the fitness function is designed as the error rate P of the SOA-KELM model when training the fault arc feature vector T And error rate P generated by SOA-KELM model during test C Sum up; the fitness function is expressed as:
F＝argmin(P T +P C )
setting upper and lower boundaries of an SOA model; setting the upper and lower boundaries of the SOA model according to the number of the weight and the threshold in the KELM model;
the evaluating the fault model includes:
the definition of the evaluation standard Accuracy Accuracy, precision, recall rate Recall and comprehensive evaluation index is as follows:
introduction of F 1 Score concept to characterize anti-interference ability of diagnostic model, F 1 Score is a comprehensive characterization of model precision and recall; comprehensive evaluation index F 1 The formula of Score is:
p in the formula c For the precision, R z Is the recall rate; precision indicates the specific gravity of the true arc of fault predicted to occur in the series fault, and Recall indicates the specific gravity of the true arc of fault predicted to occur in the series fault.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311442981.0A CN117471252A (en) | 2023-10-31 | 2023-10-31 | DC fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311442981.0A CN117471252A (en) | 2023-10-31 | 2023-10-31 | DC fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117471252A true CN117471252A (en) | 2024-01-30 |
Family
ID=89630689
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311442981.0A Pending CN117471252A (en) | 2023-10-31 | 2023-10-31 | DC fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117471252A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118013375A (en) * | 2024-04-08 | 2024-05-10 | 中国机械总院集团江苏分院有限公司 | Knitting machine fault detection method based on gsinSOA-ELM model |
-
2023
- 2023-10-31 CN CN202311442981.0A patent/CN117471252A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118013375A (en) * | 2024-04-08 | 2024-05-10 | 中国机械总院集团江苏分院有限公司 | Knitting machine fault detection method based on gsinSOA-ELM model |
| CN118013375B (en) * | 2024-04-08 | 2024-06-04 | 中国机械总院集团江苏分院有限公司 | Knitting machine fault detection method based on gsinSOA-ELM model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109272156B (en) | Ultra-short-term wind power probability prediction method | |
| CN112685961B (en) | Method and system for predicting remaining service life of analog circuit | |
| CN112149373B (en) | Complex analog circuit fault identification and estimation method and system | |
| CN117471252A (en) | DC fault arc diagnosis method based on ICEEMDAN-KPCA feature extraction and SOA-KELM | |
| CN110672905A (en) | CNN-based self-supervision voltage sag source identification method | |
| CN111553495A (en) | Small circuit breaker fault analysis method based on probabilistic neural network | |
| CN115081316A (en) | DC/DC converter fault diagnosis method and system based on improved sparrow search algorithm | |
| CN116401532B (en) | Method and system for recognizing frequency instability of power system after disturbance | |
| CN113125095A (en) | Universal circuit breaker contact system residual mechanical life prediction method based on deep learning | |
| Omar et al. | Fault classification on transmission line using LSTM network | |
| CN111931983A (en) | Precipitation prediction method and system | |
| CN112434743A (en) | Fault identification method based on GIL metal particle partial discharge time domain waveform image | |
| CN111814284A (en) | On-line voltage stability evaluation method based on correlation detection and improved random forest | |
| CN113536662B (en) | Electronic transformer error state prediction method based on firefly optimized LightGBM algorithm | |
| CN114330486A (en) | Power system bad data identification method based on improved Wasserstein GAN | |
| CN116087692B (en) | Distribution network tree line discharge fault identification method, system, terminal and medium | |
| CN115795360A (en) | Cable fault detection method based on artificial neural network | |
| CN115712871A (en) | Power electronic system fault diagnosis method combining resampling and integrated learning | |
| CN114239653A (en) | KPCA-MIV-LSTM-based fault arc multi-domain identification method | |
| CN114966224A (en) | Improved neural network-based modular multilevel converter capacitance monitoring method | |
| CN114511078A (en) | BP neural network prediction method and device based on multi-strategy sparrow search algorithm | |
| CN112183745A (en) | High-voltage cable partial discharge mode identification method based on particle swarm algorithm and DBN | |
| CN112329535A (en) | CNN-based quick identification method for low-frequency oscillation modal characteristics of power system | |
| Chen et al. | Research on Cable Terminal Partial Discharge Identification Method Based on Genetic Algorithm Optimized CNN Model | |
| CN112288153B (en) | Automatic optimization method for initial value weight of cerebellum model neural network fault diagnoser |
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 |