CN112068033B - On-line identification method for open-circuit faults of inverter power tube based on 1/6 period current - Google Patents

On-line identification method for open-circuit faults of inverter power tube based on 1/6 period current Download PDF

Info

Publication number
CN112068033B
CN112068033B CN202010907258.5A CN202010907258A CN112068033B CN 112068033 B CN112068033 B CN 112068033B CN 202010907258 A CN202010907258 A CN 202010907258A CN 112068033 B CN112068033 B CN 112068033B
Authority
CN
China
Prior art keywords
elm
fault
open
wolves
power tube
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
CN202010907258.5A
Other languages
Chinese (zh)
Other versions
CN112068033A (en
Inventor
姚芳
王晓鹏
汤俊豪
董超群
赵靖英
唐圣学
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.)
Hebei University of Technology
Original Assignee
Hebei University of Technology
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 Hebei University of Technology filed Critical Hebei University of Technology
Priority to CN202010907258.5A priority Critical patent/CN112068033B/en
Publication of CN112068033A publication Critical patent/CN112068033A/en
Application granted granted Critical
Publication of CN112068033B publication Critical patent/CN112068033B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
    • G01R31/54—Testing for continuity
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Inverter Devices (AREA)

Abstract

The invention provides an IGWO-ELM-based three-phase two-level inverter open-circuit fault online identification method, which relates to the technical field of three-phase inverter fault detection, and comprises the steps of dividing a period T into 6 areas, constructing a multi-parameter fusion seven-dimensional open-circuit fault feature vector, constructing an Extreme Learning Machine (ELM) model of the power tube open-circuit fault by taking the feature vector as input and a fault mode number as output, and optimizing model parameters by using an improved gray wolf algorithm (IGWO); according to the method, the off-line training of the power tube ELM fault identification model is carried out by using the simulation data of the three-phase current during the open circuit fault of the power tube, and the optimal model parameters are determined; the method comprises the steps of embedding an optimal IGWO-ELM inverter power tube open-circuit fault identification model into a control link of the optimal IGWO-ELM inverter power tube open-circuit fault identification model, acquiring three-phase current in real time, extracting characteristic quantity as input of the optimal IGWO-ELM model, and taking fault mode numbers as output; the method converts the fault mode number into a binary number to realize real-time fault alarm; the method fault identification total time is about 3.36ms.

Description

On-line identification method for open-circuit faults of inverter power tube based on 1/6 period current
Technical field:
the invention relates to the technical field of open-circuit fault detection of a three-phase two-level inverter power tube, in particular to an inverter power tube open-circuit fault online identification method based on multi-parameter fusion seven-dimensional fault feature quantity extraction and an IGWO-ELM fault identification model.
The background technology is as follows:
with the rapid development of power electronics technology and semiconductor technology, renewable energy technology has been greatly advanced, wherein inverter technology has been optimized. Inverters are widely used in various industries, but in comparison, the inverter has high failure rate, the reliability is lower than that of other parts of the whole inverter system, and once the inverter fails, the normal operation of the whole system is adversely affected. Wherein the power tube is one of the most inefficient devices in the inverter. In order to effectively prevent the inverter fault from further damaging the power generation system and realize the fault-tolerant operation of the inverter, the open-circuit fault identification research of the inverter is necessary.
At present, the diagnosis of the open-circuit fault of the inverter has achieved a certain result. The invention patent number 201020135717.9 discloses an on-line detection device and a detection method for open-circuit faults of power tubes of an inverter, wherein an on-line detection circuit for the open-circuit faults of each power tube in the method consists of an optocoupler circuit for detection, a logic circuit and a rising edge delay circuit; in particular, the system needs to be modified in the implementation process, a detection circuit is added, and the cost is increased. The invention patent with patent number 201510020290.0 proposes to weight the periodic average value of three-phase current and the periodic average value of absolute value, and obtain the detection variable d after normalization n And comparing the fault location with a threshold value to complete fault location.
In summary, at present, a certain research result is obtained for the open-circuit fault identification of the inverter, but there are some problems to be solved, and first, in practical application, the system is generally not allowed to add a detection circuit to the inverter so as not to interfere with the normal operation of the system. Secondly, when the variation of the applied load is large, the variation range of the current is also large, which may cause inaccurate measurement of the current, reduce reliability, and further affect the identification result. Third, how to implement online fault detection also requires the focus of research today.
The invention comprises the following steps:
the invention aims to provide an on-line diagnosis method for open-circuit faults of an inverter, which divides one period into 6 areas according to three-phase current i in a 1/6 power frequency period a 、i b And i c The measured value of (1) extracts the normalized kurtosis and the normalized pulse index, and combines the region numbers to form a multi-parameter fused seven-dimensional open circuit fault feature vector; structureEstablishing a power tube open circuit fault ELM identification model with a seven-dimensional open circuit fault feature vector as input and a fault mode number as output, and optimizing parameters of the open circuit fault identification ELM model by using an IGWO to determine an optimal model; embedding an optimal IGWO-ELM inverter power tube open-circuit fault identification model into a control link thereof, and according to a three-phase current i of 1/6 power frequency period a 、i b And i c Extracting characteristic quantity as input of an optimal IGWO-ELM fault identification model, and monitoring output of the identification model in real time; the fault mode number of 2-bit decimal output of the IGWO-ELM inverter power tube open circuit fault identification model is converted into a fault mode number of 5-bit binary, and seven-segment LEDs are driven by 2 seven-segment font display decoders, so that real-time alarm of the open circuit fault of the inverter power tube is realized; the method uses trained IGWO-ELM model to identify single failure sample with average time of 3.057X10 -5 s, adding three-phase current i a 、i b And i c The total fault identification time is about 3.36ms in the 1/6 power frequency period required by the measured value.
The technical scheme adopted by the invention is as follows:
(1) Construction of a multi-parameter fused seven-dimensional open-circuit fault feature vector
I. Fault zone partitioning
Considering that the waveform of the output current is different after faults occur at different moments in a period, the method can be based on i a The phasor angle of (a) divides a period into 6 regions, alpha a ＝1，2...6。
TABLE 1 area division
Normalized kurtosis-based open circuit fault feature extraction
The kurtosis of the three-phase grid-connected current is normalized, and the normalized kurtosis I k-kur The definition is as follows:
where N is the number of sampling points, 1/6 t=n×sampling interval.
III, extraction of open circuit fault characteristic quantity based on normalized pulse index
Normalizing the pulse index of the three-phase grid-connected current, wherein the normalized pulse index I k-f The definition is as follows:
wherein I is a-max 、I b-max And I c-max Respectively the maximum value of the three-phase current.And->Three phases respectively average amplitude values.
Open-circuit fault feature vector construction for multi-parameter fusion
And fusing the normalized kurtosis, the normalized pulse index and the fault partition of the three-phase current to construct a seven-dimensional fault feature vector:
x＝[I a-kur ，I b-kur ，I c-kur ，I a-f ，I b-f ，I c-f ，α a ] (3)
the characteristic quantity is based on three-phase current i of 1/6 period a 、i b 、i c The seven-dimensional open-circuit fault feature vectors corresponding to any one of the 6 areas (1 area, 2 area, 3 area, 4 area, 5 area and 6 area) are obtained through calculation without being influenced by load change at the output side, and when the inverter is in 21 single-tube and double-tube open-circuit fault states, the two open-circuit fault feature vectors are different from each other, so that the inverter has clear identification.
(2) Inverter power tube open-circuit fault ELM identification model construction
According to three-phase current i a，b，c Sample data calculation normalized seven-dimensional fault feature quantity I a-kur 、I b-kur 、I c-kur 、I a-f 、I b-f 、I c-f And alpha a Seven-dimensional fault characteristic quantity is used as an input layer neuron x of an ELM identification model i (i=1.,), 7), the input column vector is x= [ x ] 1 ，x 2 ，x 3 ，x 4 ，x 5 ，x 6 ，x 7 ] T 。
Constructing an L-layer hidden layer of an ELM identification model by using an excitation function g l (x) (l=1,., L) input layer neuron x to ELM recognition model i (i=1,., 7) entitled ω li Adding threshold b l Summing to obtain hidden layer neuron g l (l=1.,.. L), implicit layer column vector g= [ g ] 1 ，g 2 ，...，g L ] T The calculation formula of (2) is as follows:
wherein omega is li (l=1..l; i=1..7) and b) l (l=1.,.. L) constitute the weight matrix ω and the threshold column vector b of the input layer neurons, respectively.
Numbering the fault pattern y i =j (j=1,.,. 21) as output layer neurons of the ELM recognition model, for hidden layer neurons g l (l=1,., L) weighted β jl (j=1..21; l=1..l.) summing to form an output layer neuron y j : output layer y j The calculation formula of (2) is as follows:
wherein beta is jl (j=1..21; l=1.. L) constitute the weighting matrix β of the hidden layer neurons.
The ELM identification model of an inverter power tube open circuit fault can be expressed as:
(3) Open circuit fault ELM model parameter optimization and training based on IGWO
To prefer the power tube open circuit fault ELM recognition model parameter matrix (w, b, β), an IGWO algorithm is implanted into the ELM model.
The wolf algorithm (Grey Wolf Optimization, IGWO) is a novel group intelligent optimization algorithm proposed by mirjallii et al in 2014 for simulating social grades and hunting behaviors of the wolves in the nature. However, aiming at the task of high-dimensional multi-polar value optimization, the IGWO algorithm is easy to fall into local optimum, and the convergence accuracy is not high enough. In response to this disadvantage, researchers have proposed the IGWO algorithm. The algorithm defines four gray wolves: alpha wolves, beta wolves, delta wolves and Omega wolves. Wherein Alpha wolves are the optimal gray wolves in the current wolf group; beta wolves are several gray wolves centered on Alpha wolves and randomly searched locally around them; delta wolves follow Alpha wolves to conduct global search; omega wolves are the worst fitness gray wolves.
The specific flow of the power tube open circuit fault IGWO-ELM identification is as follows:
1) Processing the sampled data. And extracting the open-circuit fault feature vector through sampling data, and carrying out normalization processing on the open-circuit fault feature vector.
2) And initializing parameters. The number M of the wolf population in the IGWO algorithm and the maximum iteration number N of the algorithm are initialized, and the positions of the wolves in the IGWO algorithm, namely the initial w and the initial b in the ELM algorithm, are initialized. Setting upper and lower bounds of related parameters, randomly generating population, and expressing the matrix as follows:
X＝[X 1 ，X 2 ，…，X i ，…，X M ] T (7)
wherein X is i Is the position of the ith wolf.
3) And calculating the fitness. The training sample outputs an identification value through an ELM algorithm according to the initial w and the initial b in the ELM, and the accuracy is used as a fitness value. The individuals with the highest fitness are selected as Alpha wolves respectively. The fitness function is:
F(X i )＝(k/N)×100 (8)
wherein N is the total sample amount; k isThe correct number is classified. From this function, F (X i ) The closer the function value of (2) is to 100, the more satisfactory the individual is.
4) Beta wolves were randomly generated around Alpha wolves:
Beta i，j ＝Alpha j +D*(2r-1)，(i＝1，2，...，β) (9)
where r is a random number between 0 and 1. If the Beta wolf has a better fitness value than the Alpha wolf, the Beta wolf is used as a new Alpha wolf, namely:
5) Updating the Delta gray wolf. The Delta wolf is updated according to the Alpha wolf's position with the following formula:
A＝2ar 1 -a (12)
C＝2×r 2 (13)
D＝|C*Alpha-Delta i，j | (14)
wherein the convergence factor a decreases linearly with the number of iterations from 2 to 0, r 1 And r 2 Are all [0,1 ]]Is a random number of (a) in the memory.
6) Omega wolves were updated. Omega wolves were replaced with new wolves randomly generated between the upper and lower bounds.
7) And (5) iteratively updating. The position of the optimal wolf is updated according to equation (12) one (14).
8) And determining the optimal. And finishing iteration when the algorithm reaches the iteration times or the positions of the optimal wolves in the previous and subsequent iterations are the same, and outputting optimal initial w and b. And training an identification model.
9) Training the ELM system according to the w and the b obtained in the 8) to obtain an ELM optimal identification model parameter matrix (w, b, beta).
(4) IGWO-ELM-based open circuit fault online identification and fault alarm
The method willThe optimal IGWO-ELM inverter power tube open-circuit fault identification model is embedded into a control link thereof, and three-phase current i is obtained in real time from a built-in current sensor thereof a 、i b And i c Using the three-phase current i of 1/6 power frequency period a 、i b And i c And constructing the characteristic quantity as the input of an optimal IGWO-ELM inverter power tube open-circuit fault identification model, monitoring the fault mode number output by the optimal IGWO-ELM inverter power tube open-circuit fault identification model in real time, and realizing the on-line identification of the inverter power tube open-circuit fault.
In order to realize the on-line alarm of the open-circuit fault of the power tube of the inverter, the method converts the fault mode number output by the 2-bit decimal system of the open-circuit fault identification model of the power tube of the IGWO-ELM inverter into the fault mode number of the 5-bit binary system, and drives the seven-segment font LED through 2 seven-segment font display decoders to realize the real-time alarm of the open-circuit fault of the power tube of the inverter.
Description of the drawings:
the invention is further described below with reference to the drawings and the detailed description.
Fig. 1 is a graph of the output three-phase current after a fault at different times.
Fig. 2 is a fault occurrence area division.
Fig. 3 is a power tube open circuit fault ELM identification model.
FIG. 4 is a power tube open fault IGWO-ELM identification process.
FIG. 5 is a fitness curve of an IGWO algorithm optimizing ELM.
FIG. 6 is a single fault sample recognition time curve.
Fig. 7 is a time domain fault signature vector distribution diagram based on measured data.
FIG. 8 is an IGWO-ELM classification result based on measured data.
The specific embodiment is as follows:
the type of failure of the inverter is first analyzed. The probability of simultaneous occurrence of faults of a plurality of power tubes of the inverter is extremely low, and four fault modes can be provided if at most 2 power tubes simultaneously fail: I. single tube open circuit (e.g. T) 1 Or T 4 Open circuit, etc.); II. in-phase two-pipe open circuit (e.g. T 1 、T 4 Open circuit, etc.); III two tubes on the same side are open (e.g. T 1 、T 3 Open circuit, etc.); IV. different-side out-of-phase two-tube open circuit (e.g. T) 1 、T 6 Open circuit, etc.). The inverter failure-free operation mode is indicated by o.
Table 1 inverter power tube open fault numbering
The first step: seven-dimensional open circuit fault feature extraction for inverter power tube
After single-tube and double-tube open-circuit faults of inverter, three-phase grid-connected current i a 、i b 、i c The distortion is obvious, and the kurtosis, pulse index and other time domain characteristic parameters all contain fault information, so that fault characteristic quantities which are beneficial to improving the rapidity and the accuracy of fault identification can be extracted from the fault characteristic parameters.
(1) Fault zone partitioning
Considering that the waveform of the output current is different after faults occur at different moments in a period, the method can be based on i a The phasor angle of (a) divides a period into 6 regions, and the waveform of the output current is different in consideration of the failure at different times in a period (as shown in fig. 1, the solid line is a line where T occurs at time t= 0.0975s 1 Outputting three-phase current of the open-circuit fault; the dashed line is that T occurs at time t=0.1 s 1 Output three-phase current of open circuit fault), current i a A period of the phase angle of (a) is divided into 6 regions, alpha a =1, 2,3,4,5,6, as shown in fig. 2, the fault identification capability is enhanced by distinguishing the fault occurrence areas.
TABLE 2 area division
(2) Open circuit fault feature extraction based on normalized kurtosis
Considering universality of a fault identification method on inversion working conditions, normalizing kurtosis of three-phase grid-connected currentProcessing, normalized kurtosis I k-kur The definition is as follows: :
where N is the number of sampling points, 1/6 t=n×sampling interval.
(3) Open circuit fault feature extraction based on normalized pulse index
Considering universality of a fault identification method on inversion working conditions, carrying out normalization processing on pulse indexes of three-phase grid-connected current, and normalizing the pulse indexes I k-f The definition is as follows:
wherein I is a-max 、I b-max And I c-max Respectively the maximum value of the three-phase current.And->The average amplitudes of the three-phase currents are respectively.
(4) Multi-parameter fused open-circuit fault feature vector construction
And fusing the normalized kurtosis, the normalized pulse index and the fault partition of the three-phase current to construct a seven-dimensional fault feature vector:
x＝[I a-kur ，I b-kur ，I c-kur ，I a-f ，I b-f ，I c-f ，α a ] (3)
the characteristic quantity is based on three-phase current i of 1/6 period a 、i b 、i c The seven-dimensional open circuit fault feature vector corresponding to any one of the 6 areas (1 area, 2 area, 3 area, 4 area, 5 area and 6 area) is calculated without being influenced by the load change of the output side, when the inverter is in 21 single-tube and double-tube open circuit fault states,the two images are different from each other, and the image has clear identification.
And a second step of: inverter power tube open-circuit fault ELM identification model construction
According to three-phase current i a，b，c Sample data calculation normalized seven-dimensional fault feature quantity I a-kur 、I b-kur 、I c-kur 、I a-f 、I b-f 、I c-f And alpha a Seven-dimensional fault characteristic quantity is used as an input layer neuron x of an ELM identification model i (i=1.,), 7), the input column vector is x= [ x ] 1 ，x 2 ，x 3 ，x 4 ，x 5 ，x 6 ，x 7 ] T 。
Constructing an L-layer hidden layer of an ELM identification model by using an excitation function g l (x) (l=1,., L) input layer neuron x to ELM recognition model i (i=1,., 7) entitled ω li Adding threshold b l Summing to obtain hidden layer neuron g l (l=1.,.. L), implicit layer column vector g= [ g ] 1 ，g 2 ，...，g L ] T The calculation formula of (2) is as follows:
wherein omega is li (l=1..l; i=1..7) and b) l (l=1.,.. L) constitute the weight matrix ω and the threshold column vector b of the input layer neurons, respectively.
Numbering the fault pattern y i =j (j=1,.,. 21) as output layer neurons of the ELM recognition model, for hidden layer neurons g l (l=1,., L) weighted β jl (j=1..21; l=1..l.) summing to form an output layer neuron y j : output layer y j The calculation formula of (2) is as follows:
wherein beta is jl (j＝1，...，21；l＝L.) constitute a weighting matrix β for the hidden layer neurons.
The ELM identification model of an inverter power tube open circuit fault can be expressed as:
and a third step of: open circuit fault ELM model parameter optimization and training based on IGWO
To prefer the power tube open circuit fault ELM recognition model parameter matrix (w, b, β), an IGWO algorithm is implanted into the ELM model.
The wolf algorithm (Grey Wolf Optimization, IGWO) is a novel group intelligent optimization algorithm proposed by mirjallii et al in 2014 for simulating social grades and hunting behaviors of the wolves in the nature. However, aiming at the task of high-dimensional multi-polar value optimization, the IGWO algorithm is easy to fall into local optimum, and the convergence accuracy is not high enough. In response to this disadvantage, researchers have proposed the IGWO algorithm. The algorithm defines four gray wolves: alpha wolves, beta wolves, delta wolves and Omega wolves. Wherein Alpha wolves are the optimal gray wolves in the current wolf group; beta wolves are several gray wolves centered on Alpha wolves and randomly searched locally around them; delta wolves follow Alpha wolves to conduct global search; omega wolves are the worst fitness gray wolves.
The specific flow of the power tube open circuit fault IGWO-ELM identification is as follows:
1) Processing the sampled data. And extracting the open-circuit fault feature vector through sampling data, and carrying out normalization processing on the open-circuit fault feature vector.
2) And initializing parameters. The number M of the wolf population in the IGWO algorithm and the maximum iteration number N of the algorithm are initialized, and the positions of the wolves in the IGWO algorithm, namely the initial w and the initial b in the ELM algorithm, are initialized. Setting upper and lower bounds of related parameters, randomly generating population, and expressing the matrix as follows:
X＝[X 1 ，X 2 ，…，X i ，…，X M ] T (7)
wherein X is i Is the position of the ith wolf.
3) And calculating the fitness. The training sample outputs an identification value through an ELM algorithm according to the initial w and the initial b in the ELM, and the accuracy is used as a fitness value. The individuals with the highest fitness are selected as Alpha wolves respectively. The fitness function is:
F(X i )＝(k/N)×100 (8)
wherein N is the total sample amount; k is the correct number of classification. From this function, F (X i ) The closer the function value of (2) is to 100, the more satisfactory the individual is.
4) Beta wolves were randomly generated around Alpha wolves:
Beta i，j ＝Alpha j +D*(2r-1)，(i＝1，2，...，β) (9)
where r is a random number between 0 and 1. If the Beta wolf has a better fitness value than the Alpha wolf, the Beta wolf is used as a new Alpha wolf, namely:
5) Updating the Delta gray wolf. The Delta wolf is updated according to the Alpha wolf's position with the following formula:
A＝2ar 1 -a (12)
C＝2×r 2 (13)
D＝|C*Alpha-Delta i，j | (14)
wherein the convergence factor a decreases linearly with the number of iterations from 2 to 0, r 1 And r 2 Are all [0,1 ]]Is a random number of (a) in the memory.
6) Omega wolves were updated. Omega wolves were replaced with new wolves randomly generated between the upper and lower bounds.
7) And (5) iteratively updating. The position of the optimal wolf is updated according to equations (12) - (14).
8) And determining the optimal. And finishing iteration when the algorithm reaches the iteration times or the positions of the optimal wolves in the previous and subsequent iterations are the same, and outputting optimal initial w and b. And training an identification model.
9) Training the ELM system according to the w and the b obtained in the 8) to obtain an ELM optimal identification model parameter matrix (w, b, beta).
IGWO algorithm parameter settings: the initial population quantity M=10, the maximum iteration number N=100, the upper and lower limits of the number c of ELM hidden layer nodes to be optimized are [1, 500], and c is an integer. And randomly selecting 800 groups of training samples and 80 groups of test samples, inputting the training samples and the 80 groups of test samples into the IGWO-ELM, and outputting the training samples and the 80 groups of test samples as fault serial numbers of 0-21. To reduce the influence of random factors, the algorithm is repeatedly executed 100 times, and the average recognition accuracy of the training sample and the test sample is calculated to be 99.95% and 99.81%, respectively.
TABLE 3 simulation results of different open circuit fault identification methods
As can be seen from table 3, the training accuracy and the test accuracy of the open circuit fault recognition using IGWO-ELM are higher than those of the fault recognition using ELM.
When seven-dimensional fault feature vector training is adopted, an IGWO algorithm optimizes a certain fitness curve of the ELM and a single fault sample identification time curve of 100 times of algorithm execution are respectively shown in fig. 5 and 6. As can be seen from fig. 5, the fitness function value increases with the number of iterations, and the optimal fitness value stabilizes after about 10 iterations. As can be seen from FIG. 6, the average recognition time of the trained IGWO-ELM model for a single failure sample is 3.057X10 -5 s, adding three-phase current i a 、i b And i c The total fault identification time is about 3.36ms in the 1/6 power frequency period required by the measured value.
Fourth step: online identification and fault alarm of open circuit fault identification model based on IGWO-ELM
In order to verify the effectiveness of the IGWO-ELM open circuit fault identification model, an inversion experiment platform based on the DSP28335 is built. The experiment is based on the measured data of three-phase current after fault (random fault time), a normalized kurtosis value, a normalized pulse index value and a fault interval are extracted, and a trained IGWO-ELM model is used for fault identification. The seven-dimensional feature vector distributions and partial test results for 21 open faults are shown in fig. 7 and table 4, respectively. The recognition result is shown in fig. 8. As can be seen from fig. 8, a total of 2×21 (2 groups of 21 failure modes) data samples were collected for identification. The result shows that the method has a good open circuit fault identification result, and the fault identification accuracy is 97.62%.
TABLE 4 partial input data and output results for IGWO-ELM comparison
And finally, designing an open-circuit fault alarm device of the inverter power tube. The invention provides a method for converting a fault mode number output by a 2-bit decimal system of an IGWO-ELM inverter power tube open circuit fault identification model into a 5-bit binary fault mode number, and driving seven-segment font LEDs through 2 seven-segment font display decoders to realize real-time alarm of the open circuit fault of the inverter power tube.

Claims (4)

1. An on-line identification method for open-circuit faults of an inverter power tube based on 1/6 period current is characterized in that a power frequency period T is divided into 6 areas, namely, an area 1 is 330-30 degrees, an area 2 is 30-90 degrees, an area 3 is 90-150 degrees, an area 4 is 150-210 degrees, an area 5 is 210-270 degrees, an area 6 is 270-330 degrees, and three-phase grid-connected current i of each area is calculated according to the three-phase grid-connected current i of each area a 、i b And i c The method comprises the steps of (1) extracting normalized kurtosis and normalized pulse indexes, constructing a multi-parameter fused seven-dimensional open-circuit fault feature vector by combining an area number alpha, taking the constructed seven-dimensional open-circuit fault feature vector as input and a fault mode number as output, designing an inverter power tube open-circuit fault IGWO-ELM identification model, and designing a fault alarm module, wherein the total fault identification duration is about 3.36ms;
(1) Fault zone partitioning
Considering that three-phase grid-connected current waveforms of a power frequency period T are different after faults occur at different moments in the period, the power frequency period T can be divided into 6 areas according to phase angles of the three-phase grid-connected current, and α=1, 2..6;
TABLE 1 area division
(2) Open circuit fault feature based on normalized kurtosis
Carrying out normalization processing on kurtosis of three-phase grid-connected current in a single area, wherein the normalized kurtosis I k-kur The definition is as follows:
where N is the number of sampling points, (1/6) t=n×sampling interval;
(3) Open circuit fault characterization based on normalized pulse indicators
Normalizing pulse indexes of three-phase grid-connected current in a single area, wherein the normalized pulse indexes I k-f The definition is as follows:
wherein i is a-max 、i b-max And i c-max The maximum value of the three-phase grid-connected current is respectively set;and->The average amplitude of the three-phase grid-connected currents is respectively;
(4) Multi-parameter fusion seven-dimensional open circuit fault feature vector
The seven-dimensional open circuit fault feature vector for constructing the multi-parameter fusion is as follows:
x＝[I a-kur ，I b-kur ，I c-kur ，I a-f ，I b-f ，I c-f ，α] (3)
the fault characteristic vector is based on three-phase grid-connected current i of 1/6 power frequency period T a 、i b 、i c The seven-dimensional open-circuit fault feature vectors corresponding to any one of the 6 areas are obtained through calculation without being influenced by load change of an output side, and when the inverter is in 1 non-fault state and 21 single-double-tube open-circuit fault states, the inverter is different from each other, and has clear identification.
2. The on-line identification method of open-circuit faults of the inverter power tube according to claim 1, which is characterized by constructing an ELM identification model of open-circuit faults of the inverter power tube;
according to the three-phase grid-connected current i a 、i b And i c Sample data calculation normalized seven-dimensional fault feature quantity I a-kur 、I b-kur 、I c-kur 、I a-f 、I b-f 、I c-f And alpha, using seven-dimensional fault feature quantity as input layer neuron x of ELM identification model i I=1,..7, the input column vector is x= [ x 1 ，x 2 ，x 3 ，x 4 ，x 5 ，x 6 ，x 7 ] T ；
Constructing an L-layer hidden layer of an ELM identification model by using an excitation function g l (x) L=1,.. input layer neuron x for ELM recognition model i I=1,..7 weight ω li Add threshold b l Obtaining hidden layer neuron g l L=1,.. implicit layer column vector g= [ g ] 1 ，g 2 ，...，g L ] T The calculation formula of (2) is as follows:
wherein omega is li L=1,. -%, L; i=1,..7, and b. l ，l＝1,., L, respectively forming a weight matrix ω and a threshold column vector b of the input layer neurons;
numbering the fault pattern y j =j, j=1,..21, as output layer neurons of ELM recognition model, for hidden layer neurons g l L=1,.. L, weighted beta jl J=1,..21, l=1,.. L, forming output layer neuron y j Output layer y j The calculation formula of (2) is as follows:
wherein beta is jl The term "j=1, 21, l=1, L, constitutes the weighting matrix β of the hidden layer neurons;
the ELM identification model of an inverter power tube open circuit fault can be expressed as:
。
3. the on-line identification method of open circuit faults of the inverter power tube according to claim 1, which is characterized in that parameters of an open circuit fault ELM identification model of the inverter power tube are optimized by utilizing an IGWO algorithm;
for optimizing the parameter matrix w, b and beta of the ELM identification model of the open-circuit fault of the power tube, an IGWO algorithm is implanted into the ELM model; the IGWO algorithm defines four gray wolves: alpha wolves, beta wolves, delta wolves, omega wolves; wherein Alpha wolves are the optimal gray wolves in the current wolf group; beta wolves are several gray wolves centered on Alpha wolves and randomly searched locally around them; delta wolves follow Alpha wolves to conduct global search; omega wolves are the worst-fitness gray wolves;
the specific process of optimizing the parameters of the power tube open circuit fault IGWO-ELM identification model is as follows:
1) Processing the sampled data; extracting the sampling data to obtain an open-circuit fault feature vector, and carrying out normalization processing on the open-circuit fault feature vector;
2) Initializing parameters; initializing the number M of the gray wolf population in the IGWO algorithm and the maximum iteration number N of the algorithm, and initializing the gray wolf positions in the IGWO algorithm, namely the initial w and the initial b in the ELM algorithm; setting upper and lower bounds of related parameters, randomly generating population, and expressing the matrix as follows:
X＝[X 1 ，X 2 ，…，X i ，…，X M ] T (7)
wherein X is i Is the position of the ith gray wolf;
3) Calculating the fitness; the training sample outputs an identification value through an ELM algorithm according to the initial w and the initial b in the ELM, and the accuracy is used as a fitness value; selecting individuals with highest fitness as Alpha wolves respectively; the fitness function is:
F(X i )＝(k/N)×100 (8)
wherein N is the total sample amount; k is the number of correctly classified products; from this function, F (X i ) The closer the function value of (2) is to 100, the more satisfactory the individual is;
4) Beta wolves were randomly generated around Alpha wolves:
Beta i，j ＝Alpha j +D*(2r-1)，i＝1，2，...，β (9)
wherein r is a random number between 0 and 1; if the Beta wolf has a better fitness value than the Alpha wolf, the Beta wolf is used as a new Alpha wolf, namely:
5) Updating Delta gray wolves; the Delta wolf is updated according to the Alpha wolf's position with the following formula:
A＝2ar 1 -a (12)
C＝2×r 2 (13)
D＝|C*Alpha-Delta i，j | (14)
in the convergence factorSub-a decreases linearly from 2 to 0 with iteration number, r 1 And r 2 Are all [0,1 ]]Random numbers of (a);
6) Updating Omega wolves; replacement of Omega wolves with new wolves randomly generated between upper and lower bounds;
7) Iterative updating; updating the position of the optimal wolf according to formulas (11) - (14);
8) Determining an optimal value; finishing iteration when the algorithm reaches the iteration times or the positions of the optimal wolves in the front iteration and the back iteration are the same, and outputting optimal initial w and b; training an identification model;
9) Training the ELM system according to the w and the b obtained in the 8) to obtain an ELM optimal identification model parameter matrix w, b and beta.
4. The on-line identification method for the open-circuit faults of the inverter power tube is characterized in that the method converts the fault mode number of 2-bit decimal output of an identification model of the open-circuit faults of the inverter power tube IGWO-ELM into a fault mode number of 5-bit binary, and drives seven-segment font LEDs through 2 seven-segment display decoders to realize real-time alarm of the open-circuit faults of the inverter power tube.
CN202010907258.5A 2020-09-02 2020-09-02 On-line identification method for open-circuit faults of inverter power tube based on 1/6 period current Active CN112068033B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010907258.5A CN112068033B (en) 2020-09-02 2020-09-02 On-line identification method for open-circuit faults of inverter power tube based on 1/6 period current

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010907258.5A CN112068033B (en) 2020-09-02 2020-09-02 On-line identification method for open-circuit faults of inverter power tube based on 1/6 period current

Publications (2)

Publication Number Publication Date
CN112068033A CN112068033A (en) 2020-12-11
CN112068033B true CN112068033B (en) 2024-03-26

Family

ID=73665867

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010907258.5A Active CN112068033B (en) 2020-09-02 2020-09-02 On-line identification method for open-circuit faults of inverter power tube based on 1/6 period current

Country Status (1)

Country Link
CN (1) CN112068033B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115047305A (en) * 2022-05-12 2022-09-13 河北工业大学 Inverter open-circuit fault identification method based on signal processing reconstruction
CN115114964B (en) * 2022-07-21 2023-04-07 西南交通大学 Sensor intermittent fault diagnosis method based on data driving

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104122479A (en) * 2014-07-29 2014-10-29 华中科技大学 Online detection method for open-circuit faults of power tubes of induction-motor vector control system
CN107153735A (en) * 2017-04-28 2017-09-12 无锡开放大学 Motor driven systems PWM inverter method for diagnosing faults
CN107703415A (en) * 2016-12-20 2018-02-16 中国矿业大学 A kind of two level three-phase four-line D STATCOM major loop switching device open fault localization methods
CN108418459A (en) * 2018-03-20 2018-08-17 中南大学 A kind of the inverter open fault fault tolerant control method and device of current hysteresis-band control
CN108646560A (en) * 2018-05-11 2018-10-12 河北工业大学 Power of fan model parameter optimization method based on grey wolf algorithm improvement
CN110084106A (en) * 2019-03-19 2019-08-02 中国地质大学（武汉） Microgrid inverter method for diagnosing faults based on wavelet transformation and probabilistic neural network

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104122479A (en) * 2014-07-29 2014-10-29 华中科技大学 Online detection method for open-circuit faults of power tubes of induction-motor vector control system
CN107703415A (en) * 2016-12-20 2018-02-16 中国矿业大学 A kind of two level three-phase four-line D STATCOM major loop switching device open fault localization methods
CN107153735A (en) * 2017-04-28 2017-09-12 无锡开放大学 Motor driven systems PWM inverter method for diagnosing faults
CN108418459A (en) * 2018-03-20 2018-08-17 中南大学 A kind of the inverter open fault fault tolerant control method and device of current hysteresis-band control
CN108646560A (en) * 2018-05-11 2018-10-12 河北工业大学 Power of fan model parameter optimization method based on grey wolf algorithm improvement
CN110084106A (en) * 2019-03-19 2019-08-02 中国地质大学（武汉） Microgrid inverter method for diagnosing faults based on wavelet transformation and probabilistic neural network

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
《基于BP神经网络的三电平逆变器开路故障诊断研究》;王丽华 等;《机床与液压》;20200515;全文 *
《基于GWO-ELM的逆变器开路故障诊断》;姚芳 等;《电源学报》;20200115;第45-53页 *
《基于数学形态学广义分形维数逆变器故障诊断》;宋平岗;《华东交通大学学报》;20161215;全文 *

Also Published As

Publication number Publication date
CN112068033A (en) 2020-12-11

Similar Documents

Publication Publication Date Title
US10657447B1 (en) Automated model building search space reduction
CN108596327B (en) Seismic velocity spectrum artificial intelligence picking method based on deep learning
CN110766137A (en) Power electronic circuit fault diagnosis method based on longicorn whisker optimized deep confidence network algorithm
CN112068033B (en) On-line identification method for open-circuit faults of inverter power tube based on 1/6 period current
CN111860658A (en) Transformer fault diagnosis method based on cost sensitivity and integrated learning
CN109145948A (en) A kind of injection molding machine putty method for detecting abnormality based on integrated study
CN111292195A (en) Risk account identification method and device
CN116448419A (en) Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization
CN115712873A (en) Photovoltaic grid-connected operation abnormity detection system and method based on data analysis and infrared image information fusion
CN111310722A (en) Power equipment image fault identification method based on improved neural network
CN115659254A (en) Power quality disturbance analysis method for power distribution network with bimodal feature fusion
Dhibi et al. Interval-valued reduced ensemble learning based fault detection and diagnosis techniques for uncertain grid-connected PV systems
CN116345555A (en) CNN-ISCA-LSTM model-based short-term photovoltaic power generation power prediction method
Sebbane et al. Diagnosis and classification of photovoltaic panel defects based on a hybrid intelligent method
CN113222067B (en) Island intelligent detection method based on SVM-Adaboost algorithm
CN108830407B (en) Sensor distribution optimization method in structure health monitoring under multi-working condition
Batic et al. Improving knowledge distillation for non-intrusive load monitoring through explainability guided learning
CN114595788B (en) Transformer fault diagnosis method, device and equipment
CN115456106A (en) High-voltage circuit breaker fault diagnosis model optimization method
CN110852597B (en) Electricity consumption peak period resident load ratio calculation method based on generation of countermeasure network
Weng et al. Boosted backtracking search optimization with information exchange for photovoltaic system evaluation
Li et al. Fault diagnosis of analog circuit based on wavelet neural networks and chaos differential evolution algorithm
Jianli et al. Wind power forecasting by using artificial neural networks and Grubbs criterion
Jin et al. Research on multi-sensor information fusion algorithm of fan detection robot based on improved BP neural network
Zain Al-Thalabi et al. Modeling and prediction using an artificial neural network to study the impact of foreign direct investment on the growth rate/a case study of the State of Qatar

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant