Method for evaluating mechanical state of dry-type reactor based on equivalent impedance

Abstract

The invention discloses a method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance, which can accurately and quickly reflect faults, under normal conditions, the equivalent resistance of the reactor is basically unchanged, the impedance angle is basically a certain determined value, and the turn-to-turn short circuit fault will cause the change of the equivalent resistance and equivalent reactance of the reactor. Wherein the equivalent resistance is obviously increased, and the impedance angle is also changed, so that the monitoring and diagnosis of internal insulation faults and turn-to-turn short circuit faults can be realized according to the equivalent resistance of the dry reactor and the change characteristics of the power factor and the power angle, the faults can be found in time, the safe operation of the dry reactor is ensured, and the service is further provided for the diagnosis system of the internet of things. Signals are easy to extract, and slight turn-to-turn insulation faults occurring at different positions of the reactor can be rapidly detected.

Description

Technical Field

[0001] The invention relates to the field of evaluating the mechanical state of a dry-type reactor, and in particular to a method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance. Background Art

[0002] Dry-type reactors are important equipment in power systems. They are widely used in power systems because of their stable parameters, good linear characteristics, low noise, and low losses. With the increasing number of dry-type reactors in operation and the increase in their service life, early products have been subject to immature technology or inadequate process control, which has led to product quality defects. At the same time, since they use air cooling and have a relatively harsh operating environment, they are prone to insulation damage after long-term operation, further causing faults such as inter-turn short circuits, and even causing the reactor to burn out in severe cases, thus affecting the safe and stable operation of the power grid.

[0003] The causes of deformation of reactor windings can generally be summarized into four categories:

[0004] (1) There are defects or unreasonable factors in the manufacturing process. For example, if the radius of curvature at the edge of the copper wire used to wind the coil is small, the steep edge can easily cut off the insulation and cause direct contact between adjacent turns, thus causing a short circuit between adjacent turns of the wire segment.

[0005] (2) Reactors are inevitably subject to various short-circuit faults during operation. The current amplitude at this time is many times larger than the rated current, which will generate strong electromotive force on the winding. If the reactor structure itself has unreasonable design, such as insufficient local mechanical strength of the winding, or the reactor's self-resonance frequency is the same as the electromotive force frequency, it will cause resonance.

[0006] (3) Winding deformation has a cumulative effect. If the reactor winding is slightly deformed under the action of short-circuit current, it will lead to a decrease in insulation and mechanical properties. If it is not detected in time and continues to operate, the deformation of the reactor winding will increase again under the next short-circuit current, causing its mechanical properties to decrease again, resulting in a vicious cycle. Ultimately, the reactor will be damaged under subsequent short-circuit currents or overvoltages that are not too large, or even under normal operating conditions.

[0007] However, the existing monitoring methods have the problems of low sensitivity and poor reliability, and cannot detect faults in time, resulting in burning of the reactor. Therefore, it is particularly important to study more effective monitoring methods to detect faults in time. At present, the online monitoring methods for dry-type reactors at home and abroad mainly include temperature monitoring method, partial discharge monitoring method, and magnetic induction detection coil method, but none of them can accurately and reliably reflect the insulation and minor turn-to-turn short-circuit faults of the reactor. Summary of the invention

[0008] In view of the deficiencies in the prior art, the present invention provides a method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance to solve the above problems.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions.

[0010] A method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance, comprising:

[0011] Step S1 uses SolidWorks software to establish a reactor turn-to-turn short-circuit model, the model includes a normal part and a short-circuit part, the normal part is replaced by a cylindrical structure, the short-circuit part is replaced by a short-circuit ring, a short-circuit ring is set up in the short-circuit layer, the short-circuit layer is divided into three parts, the upper and lower parts are the normal current flowing parts of the short-circuit layer, and the middle part is the short-circuit turn;

[0012] Step S2: The inter-turn short-circuit magnetic field-circuit coupling calculation method adopts the finite element calculation method. The two ends of each layer of the reactor coil are connected with industrial frequency AC voltage. In addition to satisfying the Maxwell equations, the field-circuit coupling analysis method satisfies the Poisson equation in the magnetic field inside and outside the solution domain:

[0013]

[0014] Where: A θ is the circumferential component of the vector magnetic potential A, r, θ, z are the radius, angle and axial coordinates in the cylindrical coordinate system, μ is the magnetic permeability of the aluminum wire, J is the current density flowing through the coil and the short-circuit turn,

[0015] The circuit equations of each layer of coils and short-circuited turns satisfy:

[0016]

[0017] Where: Ψi, A i are the flux linkage and magnetic potential of the coil and the short-circuited turn, respectively.

[0018] Step S3 imports the established turn-to-turn short-circuit model into the Maxwell software, and sets the solver to a static magnetic field solver. In the static magnetic field solver, the excitation current is arbitrary, and a matrix is ​​added to set the number of turns of each layer of coils. For the turn-to-turn short-circuit fault layer, the upper and lower layers of coils are set in series, and after grid division, the calculation of the turn-to-turn short-circuit inductance matrix is ​​finally completed;

[0019] Step S4: The formula for calculating the inter-turn short-circuit current is:

[0020]

[0021] Where:

[0022]

[0023] U1=U2=…=U W-1 =U W =U N

[0024] U N is the rated phase voltage, M is the inductance matrix. I1…I W is the current of each layer, I W+1 …I W+x is the inter-turn short-circuit current, R1…R W+x are the resistance of each layer coil and short-circuit turn, M 1,2 …M1,W +1 …M W+x-1 , W+x are the mutual inductances between coils of each layer, between coils of each layer and short-circuited turns, and between short-circuited turns, L1…L W+x are the self-inductance of each layer of coil and short-circuit turn respectively. By calculating the inductance matrix M, substitute it into the formula The inter-turn short-circuit current can be obtained.

[0025] The final step S5 is to further investigate the diagnosis based on the above steps S1-S4.

[0026] Preferably, the steps of further studying and diagnosing in step S5 are as follows:

[0027] Step S51 establishes a reactor inter-turn short-circuit model, calculates the equivalent resistance R and equivalent reactance X between each turn coil, and constructs a characteristic vector H1.

[0028] Step S52 collects the vibration signal of the dry-type reactor in a normal state and in a fault state through an acceleration sensor, and performs wavelet packet decomposition on the vibration signal to obtain 8 frequency bands. The 8 frequency band signals are respectively represented in the form of energy and normalized to construct a feature vector H2.

[0029] Step S53 combines the feature vectors H1 and H2 constructed in step S51 and step S52 to construct a new feature vector H3 as the input of the SOM network.

[0030] Step S54 trains the SOM network by inputting the feature vector constructed in step S53 into the SOM network to train the SOM network; when the SOM network reaches the training error, the test data is input into the trained SOM network to determine the fault mode of the dry-type reactor.

[0031] Step S55 simulates the current, voltage and vibration changes of the winding turn-to-turn short circuit and insulation aging. The data acquisition card collects 20,000 points of vibration data at a rate of 20 kHz when the circuit breaker simulates action and sends them to the PC for processing;

[0032] The collected vibration data are divided into two groups, each group includes the feature vectors in the normal state and the feature vectors in the fault state. One group is used for training the SOM network, and the other group is used for testing the SOM network.

[0033] The dry-type reactor fault identification method based on equivalent impedance and SOM network described in step S51 is characterized in that the fault state is two conditions: short circuit between winding turns and insulation aging. In step S52, the collected vibration signal is subjected to a 2-layer wavelet packet transform using db10 wavelet to obtain 4 frequency bands.

[0034] The energy expression of the frequency band signal is:

[0035]

[0036] Among them, E3(i) is the energy of the i-th frequency band, W(3,i) is the signal of the i-th frequency band in the third layer after the two-layer decomposition of the wavelet packet, and x ik represents the value of the kth discrete point of the i-th frequency band signal W(2,i), i=0,1,···,3, represents 4 frequency bands, k=1,2,···,N, N represents the number of discrete points, the collected vibration signal,

[0037] The characteristic vector T constructed by the proportion of energy in each frequency band is:

[0038] T=[E3(0) / S3,…,E3(7) / S3]

[0039]

[0040] Among them, S3 is the total energy of the third layer after the three-layer decomposition of the wavelet packet, and then the SOM network is used for training.

[0041] Preferably, the training steps of the SOM network are as follows:

[0042] Competition, given the t-th input X(t), calculate the weight vector W of each neuron i (t) Relative distance D from the input vector i (t):||X(t)-W i (t)||=D i (t), the one with the smallest relative distance is selected as the winner, then excited, that is: i * (t) = {i:minD i (t)};

[0043] Learning, the winner and its neighboring neurons are allowed to learn the input pattern, and the adjustment of the weight coefficient is proportional to the difference between the weight coefficient and the input: ΔW i =η(t)(X(t)-W i(t)), η(t) is a variable learning rate that decays over time;

[0044] The competition and learning steps are repeated until the excited neurons correspond stably to the input samples, and the training is completed.

[0045] Preferably, the SOM network is trained 200 times using the training samples, and data is randomly selected from the training samples for training each time.

[0046] Preferably, the step of further researching and diagnosing in step S5 may also be:

[0047] Step S511: collect statistics on the equivalent resistance of the reactor in a normal state and the equivalent resistance in a fault state as training data sets and test data sets.

[0048] Step S512: extract fault features based on the micro-increment wavelet singular entropy algorithm for the target training data and the test data respectively; use the obtained micro-increment wavelet singular entropy as the fault feature to construct a feature quantity;

[0049] Step S513: clustering the feature quantities using the PSO-Kmeans clustering algorithm.

[0050] Preferably, step S511 is as follows: simulate the impedance matrix (m=1, 2, ... 30) of the dry-type air-core reactor in normal and faulty states, and collect 10 groups of data for each fault mode, for a total of 30 groups of samples:

[0051] R m =diag(λ1,λ2,…,λ n )

[0052] Among them, λ1,λ2,…,λ n are the equivalent resistances between turns of the reactor, and n is the number of turns.

[0053] Preferably, the step S512 is as follows: constructing n-order micro-increment wavelet singular entropy of the test sample and the training sample, and the corresponding formula is:

[0054]

[0055] Among them, X i is i The i-th incremental wavelet singular entropy,

[0056] Construct the eigenvector X with n-order incremental wavelet singular entropy elements s1 , X s1 =[X1,X2,…,X n ].

[0057] Preferably, the specific PSO-Kmeans calculation steps in step S513 are as follows:

[0058] Step S5131: The position of the i-th particle in the particle swarm in the n-dimensional space is represented as x i =(x i1 ,x i2 ,···,x in ), speed v i =(v i1 ,v i2 ,···,v in );

[0059] Step S5132: the best position p experienced by the i-th particle in the particle swarm in the n-dimensional space i =(p i1 ,p i2 ,···,p in );

[0060] Step S5133: the best position experienced by the entire particle swarm g = (g1, g2, ···, g n );

[0061] Step S5134: The formula for the particle to update its own speed and position is:

[0062] v i (t+1)=ωv i (t)+c1×r1×(p i -x i )

[0063] +c2×r2×(gx i )

[0064] x i (t+1)=x i +v i (t+1)

[0065] Where t is a variable representing the number of iterations; ω is the inertia weight coefficient; constants c1 and c2 are learning factors; r1 and r2 are random numbers uniformly distributed in the range [0, 1].

[0066] Step S5135, the fitness function calculation formula is f(x)=1 / d max (x i ,x j ), where d max (x i ,x j ) is the point x i With cluster point x j The maximum distance between .

[0067] Preferably, the K-means clustering algorithm based on PSO is described as follows: input the data set to be clustered, the number of clusters k and the maximum number of iterations M; output k cluster divisions whose cluster centers of the clustered data set no longer change, and the steps of the algorithm are as follows:

[0068] Step S51351, divide the data set into k clusters, each cluster is a particle swarm, and randomly set the initial position x of each particle i and initial velocity v i ;

[0069] Step S51352, determining the updated particle speed and position according to the formula for updating the particle's own speed and position in step S5134;

[0070] Step S51353, calculating the fitness value of each particle;

[0071] Step S51354, compare the fitness value of each particle with the best position p it has ever experienced i If the fitness value is larger, the current particle position and fitness value are used to update p i and p i The fitness value of

[0072] Step S51355, compare the fitness value of each particle with the fitness value of the best position g experienced by the entire particle swarm. If the fitness value is greater, update the fitness values ​​of g and g with the current particle position and fitness value, and record the subscript of the position in the data set;

[0073] Step S51356, updating the speed and position of each particle according to the formula;

[0074] Step S51357, until the maximum number of iterations M is reached, the iteration of the particle swarm is terminated and the corresponding cluster center is used as the initial value of the K-means algorithm, otherwise, go to step S51352 to continue iterative execution;

[0075] Step S51358, starting from the k initial cluster centers, use the K-means algorithm to perform clustering.

[0076] Preferably, the fault status of step S511 includes winding turn-to-turn short circuit and insulation aging.

[0077] Compared with the prior art, the present invention discloses a method for evaluating the mechanical state of a dry-type inductor based on equivalent impedance, which can accurately and quickly reflect faults. Under normal circumstances, the equivalent resistance of the inductor remains basically unchanged, and the impedance angle is basically a certain value. A turn-to-turn short-circuit fault will cause changes in the equivalent resistance and equivalent reactance of the inductor, among which the equivalent resistance increases significantly and the impedance angle also changes. Therefore, according to the change characteristics of the equivalent resistance and power factor power angle of the dry reactor, internal insulation faults and turn-to-turn short-circuit faults can be monitored and diagnosed, faults can be discovered in time, and the safe operation of the dry-type inductor can be ensured. It also serves the diagnostic system of the Internet of Things, and the signal is easy to extract, and slight turn-to-turn insulation faults occurring at different positions of the reactor can be quickly detected. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 This is a flow chart of a method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance according to the present invention;

[0079] Figure 2 This is an experimental data diagram of the method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance of the present invention;

[0080] Figure 3 This is an experimental data diagram of the method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance of the present invention;

[0081] Figure 4 This is an experimental data diagram of the method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance of the present invention;

[0082] Figure 5 This is an experimental data diagram of the method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance of the present invention;

[0083] Figure 6 This is an experimental data diagram of the method for evaluating the mechanical state of a dry-type inductor based on equivalent impedance according to the present invention. DETAILED DESCRIPTION

[0084] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0085] A method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance, comprising:

[0086] Step S1 uses SolidWorks software to establish a reactor turn-to-turn short-circuit model, the model includes a normal part and a short-circuit part, the normal part is replaced by a cylindrical structure, the short-circuit part is replaced by a short-circuit ring, a short-circuit ring is set up in the short-circuit layer, the short-circuit layer is divided into three parts, the upper and lower parts are the normal current flowing parts of the short-circuit layer, and the middle part is the short-circuit turn;

[0087] Step S2: The inter-turn short-circuit magnetic field-circuit coupling calculation method adopts the finite element calculation method. The two ends of each layer of the reactor coil are connected with industrial frequency AC voltage. In addition to satisfying the Maxwell equations, the field-circuit coupling analysis method satisfies the Poisson equation in the magnetic field inside and outside the solution domain:

[0088]

[0089] Where: A θ is the circumferential component of the vector magnetic potential A, r, θ, z are the radius, angle and axial coordinates in the cylindrical coordinate system, μ is the magnetic permeability of the aluminum wire, J is the current density flowing through the coil and the short-circuit turn,

[0090] The circuit equations of each layer of coils and short-circuited turns satisfy:

[0091]

[0092] Where: i , A i are the flux linkage and magnetic potential of the coil and the short-circuited turn, respectively.

[0093] Step S3 imports the established turn-to-turn short-circuit model into the Maxwell software, and sets the solver to a static magnetic field solver. In the static magnetic field solver, the excitation current is arbitrary, and a matrix is ​​added to set the number of turns of each layer of coils. For the turn-to-turn short-circuit fault layer, the upper and lower layers of coils are set in series, and after grid division, the calculation of the turn-to-turn short-circuit inductance matrix is ​​finally completed;

[0094] Step S4: The formula for calculating the inter-turn short-circuit current is:

[0095]

[0096] Where:

[0097]

[0098] U1=U2=…=U W-1 =U W =U N

[0099] U N is the rated phase voltage, M is the inductance matrix. I1…I W is the current of each layer, I W+1 …I W+x is the inter-turn short-circuit current, R1…R W+x are the resistance of each layer coil and short-circuit turn, M 1,2 …M1,W +1 …M W+x-1, W+x are the mutual inductances between coils of each layer, between coils of each layer and short-circuited turns, and between short-circuited turns, L1…L W+x are the self-inductance of each layer of coil and short-circuit turn respectively. By calculating the inductance matrix M, substitute it into the formula The inter-turn short-circuit current can be obtained.

[0100] The final step S5 is to further investigate the diagnosis based on the above steps S1-S4.

[0101] The steps of further studying the diagnosis in step S5 are as follows:

[0102] Step S51 establishes a reactor inter-turn short-circuit model, calculates the equivalent resistance R and equivalent reactance X between each turn coil, and constructs a characteristic vector H1.

[0103] Step S52 collects the vibration signal of the dry-type reactor in a normal state and in a fault state through an acceleration sensor, and performs wavelet packet decomposition on the vibration signal to obtain 8 frequency bands. The 8 frequency band signals are respectively represented in the form of energy and normalized to construct a feature vector H2.

[0104] Step S53 combines the feature vectors H1 and H2 constructed in step S51 and step S52 to construct a new feature vector H3 as the input of the SOM network.

[0105] Step S54 trains the SOM network by inputting the feature vector constructed in step S53 into the SOM network to train the SOM network; when the SOM network reaches the training error, the test data is input into the trained SOM network to determine the fault mode of the dry-type reactor.

[0106] Step S55 simulates the current, voltage and vibration changes of the winding turn-to-turn short circuit and insulation aging. The data acquisition card collects 20,000 points of vibration data at a rate of 20 kHz when the circuit breaker simulates action and sends them to the PC for processing;

[0107] The collected vibration data are divided into two groups, each group includes the feature vectors in the normal state and the feature vectors in the fault state. One group is used for training the SOM network, and the other group is used for testing the SOM network.

[0108] The dry-type reactor fault identification method based on equivalent impedance and SOM network described in step S51 is characterized in that the fault state is two conditions: short circuit between winding turns and insulation aging. In step S52, the collected vibration signal is subjected to a 2-layer wavelet packet transform using db10 wavelet to obtain 4 frequency bands.

[0109] The energy expression of the frequency band signal is:

[0110]

[0111] Among them, E3(i) is the energy of the i-th frequency band, W(3,i) is the signal of the i-th frequency band in the third layer after the two-layer decomposition of the wavelet packet, and x ik represents the value of the kth discrete point of the i-th frequency band signal W(2,i), i=0,1,···,3, represents 4 frequency bands, k=1,2,···,N, N represents the number of discrete points, the collected vibration signal,

[0112] The characteristic vector T constructed by the proportion of energy in each frequency band is:

[0113] T=[E3(0) / S3,…,E3(7) / S3]

[0114]

[0115] Among them, S3 is the total energy of the third layer after the three-layer decomposition of the wavelet packet, and then the SOM network is used for training.

[0116] The training steps of the SOM network are as follows:

[0117] Competition, given the t-th input X(t), calculate the weight vector W of each neuron i (t) Relative distance D from the input vector i (t):||X(t)-W i (t)||=D i (t), the one with the smallest relative distance is selected as the winner, then excited, that is: i * (t) = {i:minD i (t)};

[0118] Learning, the winner and its neighboring neurons are allowed to learn the input pattern, and the adjustment of the weight coefficient is proportional to the difference between the weight coefficient and the input: ΔW i =η(t)(X(t)-W i (t)), η(t) is a variable learning rate that decays over time;

[0119] The competition and learning steps are repeated until the excited neurons correspond stably to the input samples, and the training is completed.

[0120] The SOM network was trained 200 times using the training samples, and data was randomly selected from the training samples for training each time.

[0121] The further research and diagnosis steps in step S5 may also include:

[0122] Step S511: collect statistics on the equivalent resistance of the reactor in a normal state and the equivalent resistance in a fault state as training data sets and test data sets.

[0123] Step S512: extract fault features based on the micro-increment wavelet singular entropy algorithm for the target training data and the test data respectively; use the obtained micro-increment wavelet singular entropy as the fault feature to construct a feature quantity;

[0124] Step S513: clustering the feature quantities using the PSO-Kmeans clustering algorithm.

[0125] The step S511 is as follows: simulate the impedance matrix (m=1, 2, ... 30) of the dry-type air-core reactor in normal and faulty states, and collect 10 sets of data for each fault mode, for a total of 30 sets of samples:

[0126] R m =diag(λ1,λ2,…,λ n )

[0127] Among them, λ1,λ2,…,λ n are the equivalent resistances between turns of the reactor, and n is the number of turns.

[0128] The step S512 is as follows: construct the n-order micro-increment wavelet singular entropy of the test sample and the training sample, and the corresponding formula is:

[0129]

[0130] Among them, X i is i The i-th incremental wavelet singular entropy,

[0131] Construct the eigenvector X with n-order incremental wavelet singular entropy elements s1 , X s1 =[X1,X2,…,X n ].

[0132] The specific calculation steps of PSO-Kmeans in step S513 are as follows:

[0133] Step S5131: The position of the i-th particle in the particle swarm in the n-dimensional space is represented as x i =(x i1 ,x i2 ,···,x in ), speed v i =(v i1 ,v i2 ,···,v in );

[0134] Step S5132: the best position p experienced by the i-th particle in the particle swarm in the n-dimensional space i =(p i1 ,p i2,···,p in );

[0135] Step S5133: the best position experienced by the entire particle swarm g = (g1, g2, ···, g n );

[0136] Step S5134: The formula for the particle to update its own speed and position is:

[0137] v i (t+1)=ωv i (t)+c1×r1×(p i -x i )

[0138] +c2×r2×(gx i )

[0139] x i (t+1)=x i +v i (t+1)

[0140] Where t is a variable representing the number of iterations; ω is the inertia weight coefficient; constants c1 and c2 are learning factors; r1 and r2 are random numbers uniformly distributed in the range [0, 1].

[0141] Step S5135, the fitness function calculation formula is f(x)=1 / d max (x i ,x j ), where d max (x i ,x j ) is the point x i With cluster point x j The maximum distance between .

[0142] The K-means clustering algorithm based on PSO is described as follows: input the data set to be clustered, the number of clusters k, and the maximum number of iterations M; output k cluster divisions whose cluster centers of the clustered data set no longer change. The steps of the algorithm are as follows:

[0143] Step S51351, divide the data set into k clusters, each cluster is a particle swarm, and randomly set the initial position x of each particle i and initial velocity v i ;

[0144] Step S51352, determining the updated particle speed and position according to the formula for updating the particle's own speed and position in step S5134;

[0145] Step S51353, calculating the fitness value of each particle;

[0146] Step S51354, compare the fitness value of each particle with the best position p it has ever experienced i If the fitness value is larger, the current particle position and fitness value are used to update p i and p i The fitness value of

[0147] Step S51355, compare the fitness value of each particle with the fitness value of the best position g experienced by the entire particle swarm. If the fitness value is greater, update the fitness values ​​of g and g with the current particle position and fitness value, and record the subscript of the position in the data set;

[0148] Step S51356, updating the speed and position of each particle according to the formula;

[0149] Step S51357, until the maximum number of iterations M is reached, the iteration of the particle swarm is terminated and the corresponding cluster center is used as the initial value of the K-means algorithm, otherwise, go to step S51352 to continue iterative execution;

[0150] Step S51358, starting from the k initial cluster centers, use the K-means algorithm to perform clustering.

[0151] The fault status in step S511 includes winding turn-to-turn short circuit and insulation aging.

[0152] The following is a calculation object of the BKGKL-2000 / 35 dry-type air-core shunt reactor produced by a certain reactor manufacturer. Its specific parameters are shown in Table 1:

[0153] Table 1 Main parameters of dry-type air-core shunt reactor

[0154]

[0155] Reactor turn-to-turn short-circuit equivalent circuit model Figure 2 As shown, I1…I W is the current of each layer, I W+1 …I W+x is the inter-turn short-circuit current, R1…R W+x are the resistance of each layer coil and short-circuit turn, M 1,2 …M 1,W+1 …M W+x-1,W+x are the mutual inductances between coils at each layer, between coils at each layer and short-circuited turns, and between short-circuited turns, respectively. L1…LW+x are the self-inductances of coils at each layer and short-circuited turns. The reactor turn-to-turn short-circuit model and fault setting are as follows: Figure 3As shown. A short-circuit ring is set up in the short-circuit layer. The short-circuit layer is divided into three parts. The upper and lower parts are the parts where the normal current flows through the short-circuit layer, and the middle part is the short-circuit turn. This reactor is composed of 5 encapsulated 20 layers of coils. In order to analyze the inter-turn short-circuit conditions at different positions, each layer is divided into six equal parts and set into 7 types of faults. From the inner layer to the outer layer, from the bottom to the top, the inter-turn short-circuit faults are named 1-1, 1-2...20-6, 20-7. Due to the various inter-turn short-circuit conditions, in order to simplify the calculation, the quarter model of the reactor will be used as the simulation object.

[0156] 140 different types of turn-to-turn short-circuit models were established in SolidWorks software and then imported into Maxwell software respectively. The inductance matrix was solved through solver setting, excitation addition and meshing.

[0157] Depend on Figure 4 It can be seen that when the reactor has a turn-to-turn short circuit, the current at the short circuit point increases sharply and is symmetrical about the center position. The maximum short circuit current appears at the 15-4 fault point, with a value of 1617.982A, which is 295.3 times its layer current and 16.3 times the rated current. Figure 5 It can be seen that the short-circuit current at different positions in the axial direction is quite different. From the end to the center, the short-circuit current increases rapidly. The closer to the center, the greater the short-circuit current. The short-circuit current is the largest at half the height of the reactor, and decreases towards both ends. Figure 6 It can be seen that in the radial direction, the short-circuit currents in the same package are similar, and the short-circuit currents of different packages are different. The closer to the outer layer, the larger the short-circuit current. The faults of 1 and 7, 2 and 6, 3 and 5 are symmetrical, and the short-circuit currents are close. The short-circuit currents at the same position of the 3rd, 4th and 5th packages are almost the same, the short-circuit current of the 4th package is larger than that of the 5th package, and the short-circuit current of the 1st package is relatively smaller than that of other packages.

[0158] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0159] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0160] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance, characterized in that: include Step S1 uses SolidWorks software to establish a reactor turn-to-turn short-circuit model, the model includes a normal part and a short-circuit part, the normal part is replaced by a cylindrical structure, the short-circuit part is replaced by a short-circuit ring, a short-circuit ring is set up in the short-circuit layer, the short-circuit layer is divided into three parts, the upper and lower parts are the normal current flowing parts of the short-circuit layer, and the middle part is the short-circuit turn; Step S2: The inter-turn short-circuit magnetic field-circuit coupling calculation method adopts the finite element calculation method. The two ends of each layer of the reactor coil are connected with industrial frequency AC voltage. In addition to satisfying the Maxwell equations, the field-circuit coupling analysis method satisfies the Poisson equation in the magnetic field inside and outside the solution domain: Where: A θ is the circumferential component of the vector magnetic potential A, r, θ, z are the radius, angle and axial coordinates in the cylindrical coordinate system, μ is the magnetic permeability of the aluminum wire, J is the current density flowing through the coil and the short-circuit turn, The circuit equations of each layer of coils and short-circuited turns satisfy: Where: i , A i are the flux linkage and magnetic potential of the coil and the short-circuited turn, respectively. Step S3 imports the established turn-to-turn short-circuit model into Maxwell software, and sets the solver to a static magnetic field solver. In the static magnetic field solver, the excitation current is arbitrary, and a matrix is ​​added to set the number of turns of each layer of coils. For the turn-to-turn short-circuit fault layer, the upper and lower layers of coils are set in series, and after grid division, the calculation of the turn-to-turn short-circuit inductance matrix is ​​finally completed; Step S4: The formula for calculating the inter-turn short-circuit current is: Where: U1=U2=…=U W-1 =U W =U N U N is the rated phase voltage, M is the inductance matrix, I1…I W is the current of each layer, I W+1 …I W+x is the inter-turn short-circuit current, R1…R W+x are the resistance of each layer coil and short-circuit turn, M 1,2 …M1,W +1 …M W+x-1 , W+x are the mutual inductances between coils of each layer, between coils of each layer and short-circuited turns, and between short-circuited turns, L1…L W+x are the self-inductance of each layer of coil and short-circuit turn respectively. By calculating the inductance matrix M, substitute it into the formula The inter-turn short-circuit current can be obtained. The final step S5 is to further investigate the diagnosis based on the above steps S1-S4.

2. The method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance according to claim 1, characterized in that: The steps of further studying the diagnosis in step S5 are as follows: Step S51 establishes a reactor inter-turn short-circuit model, calculates the equivalent resistance R and equivalent reactance X between each turn coil, and constructs a characteristic vector H1. Step S52 collects the vibration signal of the dry-type reactor in a normal state and in a fault state through an acceleration sensor, and performs wavelet packet decomposition on the vibration signal to obtain 8 frequency bands. The 8 frequency band signals are respectively represented in the form of energy and normalized to construct a feature vector H2. Step S53 combines the feature vectors H1 and H2 constructed in step S51 and step S52 to construct a new feature vector H3 as the input of the SOM network. Step S54 trains the SOM network by inputting the feature vector constructed in step S53 into the SOM network to train the SOM network; When the SOM network reaches the training error, the test data is input into the trained SOM network to determine the fault mode of the dry-type reactor. Step S55 simulates the current, voltage and vibration changes of the winding turn-to-turn short circuit and insulation aging. The data acquisition card collects 20,000 points of vibration data at a rate of 20 kHz when the circuit breaker simulates action and sends them to the PC for processing; The collected vibration data are divided into two groups, each group includes the feature vectors in the normal state and the feature vectors in the fault state. One group is used for training the SOM network, and the other group is used for testing the SOM network. The dry-type reactor fault identification method based on equivalent impedance and SOM network described in step S51 is characterized in that the fault state is two conditions: short circuit between winding turns and insulation aging. In step S52, the collected vibration signal is subjected to a 2-layer wavelet packet transform using db10 wavelet to obtain 4 frequency bands. The energy expression of the frequency band signal is: Among them, E3(i) is the energy of the i-th frequency band, W(3,i) is the signal of the i-th frequency band in the third layer after the two-layer decomposition of the wavelet packet, and x ik represents the value of the kth discrete point of the i-th frequency band signal W(2,i), i=0,1,···,3, represents 4 frequency bands, k=1,2,···,N, N represents the number of discrete points, the collected vibration signal, The characteristic vector T constructed by the proportion of energy in each frequency band is: T=[E3(0) / S3,…,E3(7) / S3] Among them, S3 is the total energy of the third layer after the three-layer decomposition of the wavelet packet, and then the SOM network is used for training.

3. The method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance according to claim 2, characterized in that: The training steps of the SOM network are as follows: Competition, given the t-th input X(t), calculate the weight vector W of each neuron i (t) Relative distance D from the input vector i (t): ‖X(t)-W i (t)‖=D i (t), the one with the smallest relative distance is selected as the winner, then excited, that is: i * (t) = {i:min D i (t)}; study , the winner and its neighboring neurons are allowed to learn the input pattern, and the adjustment of the weight coefficient is proportional to the difference between the weight coefficient and the input: ΔW i =η(t)(X(t)-W i (t)), η(t) is a variable learning rate that decays over time; The competition and learning steps are repeated until the excited neurons correspond stably to the input samples, and the training is completed.

4. The method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance according to claim 3, characterized in that: The SOM network was trained 200 times using the training samples, and data was randomly selected from the training samples for training each time.

5. The method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance according to claim 1, characterized in that: The further research and diagnosis steps in step S5 may also include: Step S511: collect statistics on the equivalent resistance of the reactor in a normal state and the equivalent resistance in a fault state as training data sets and test data sets. Step S512: performing fault feature extraction based on the micro-increment wavelet singular entropy algorithm on the target training data and the test data respectively; The obtained micro-increment wavelet singular entropy is used as the fault feature to construct the feature quantity; Step S513: clustering the feature quantities using the PSO-Kmeans clustering algorithm.

6. The method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance according to claim 5, characterized in that: The step S511 is as follows: simulate the impedance matrix (m=1, 2, ... 30) of the dry-type air-core reactor in normal and faulty states, and collect 10 sets of data for each fault mode, for a total of 30 sets of samples: R m =diag(λ1,λ2,…,λ n ) Among them, λ1,λ2,…,λ n are the equivalent resistances between turns of the reactor, and n is the number of turns.

7. The method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance according to claim 5, characterized in that: The step S512 is as follows: construct the n-order micro-increment wavelet singular entropy of the test sample and the training sample, and the corresponding formula is: Among them, X i is i The i-th incremental wavelet singular entropy, Construct the eigenvector X with n-order incremental wavelet singular entropy elements s1 , X s1 =[X1,X2,…,X n ].

8. The method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance according to claim 5, characterized in that: The specific calculation steps of PSO-Kmeans in step S513 are as follows: Step S5131: The position of the i-th particle in the particle swarm in the n-dimensional space is represented as x i =(x i1 ,x i2 ,···,x in ), speed v i =(v i1 ,v i2 ,···,v in ); Step S5132: the best position p experienced by the i-th particle in the particle swarm in the n-dimensional space i =(p i1 ,p i2 ,···,p in ); Step S5133: the best position experienced by the entire particle swarm g = (g1, g2, ···, g n ); Step S5134: The formula for the particle to update its own speed and position is: v i (t+1)=ωv i (t)+c1×r1×(p i -x i )+c2×r2×(g-x i ) x i (t+1)=x i +v i (t+1) Where t is a variable representing the number of iterations; ω is the inertia weight coefficient; constants c1 and c2 are learning factors; r1 and r2 are random numbers uniformly distributed in the range [0, 1]. Step S5135, the fitness function calculation formula is f(x)=1 / d max (x i ,x j ), where d max (x i ,x j ) is the point x i With cluster point x j The maximum distance between .

9. The method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance according to claim 8, characterized in that: The K-means clustering algorithm based on PSO is described as follows: input the data set to be clustered, the number of clusters k, and the maximum number of iterations M; output k cluster divisions whose cluster centers of the clustered data set no longer change. The steps of the algorithm are as follows: Step S51351, divide the data set into k clusters, each cluster is a particle swarm, and randomly set the initial position x of each particle i and initial velocity v i ; Step S51352, determining the updated particle speed and position according to the formula for updating the particle's own speed and position in step S5134; Step S51353, calculating the fitness value of each particle; Step S51354, compare the fitness value of each particle with the best position p it has ever experienced i If the fitness value is larger, the current particle position and fitness value are used to update p i and p i The fitness value of Step S51355, compare the fitness value of each particle with the fitness value of the best position g experienced by the entire particle swarm. If the fitness value is greater, update the fitness values ​​of g and g with the current particle position and fitness value, and record the subscript of the position in the data set; Step S51356, updating the speed and position of each particle according to the formula; Step S51357, until the maximum number of iterations M is reached, the iteration of the particle swarm is terminated and the corresponding cluster center is used as the initial value of the K-means algorithm, otherwise, go to step S51352 to continue iterative execution; Step S51358, starting from the k initial cluster centers, use the K-means algorithm to perform clustering.

10. The method for evaluating the mechanical state of a dry-type reactor based on equivalent impedance according to claim 6, characterized in that: The fault status in step S511 includes winding turn-to-turn short circuit and insulation aging.

Images