Isolation switch fault diagnosis method based on tFLO-VMD-SVM

Abstract

The invention discloses an isolation switch fault diagnosis method based on tFLO-VMD-SVM. The method comprises the following steps: step 1, obtaining a vibration signal of an isolation switch; 2, obtaining an optimal intrinsic mode function component of each layer by adopting a VMD algorithm in combination with a t distribution disturbance umbrella-sky optimization algorithm; 3, calculating the fuzzy dispersion entropy and the fine composite multi-scale fuzzy dispersion entropy of each layer of intrinsic mode function component, thereby constructing a multivariate feature fusion matrix; 4, obtaining a reconstruction space matrix based on the multivariate feature fusion matrix, and dividing the reconstruction space matrix into a training set and a test set; step 5, generating a support vector machine, training the support vector machine by using the training set, and performing optimization by using a whale optimization algorithm during training to obtain a fault diagnosis model; and step 6, inputting the test set into the fault diagnosis model to obtain a fault diagnosis result. According to the method, the iteration speed and convergence are improved, and the fault identification accuracy of the disconnecting switch is higher.

Description

Technical Field

[0001] The invention relates to the field of disconnector fault diagnosis methods, in particular to a disconnector fault diagnosis method based on tFLO-VMD-SVM. Background Art

[0002] At present, the power industry is ushering in an unprecedented period of development and prosperity. The rapid development of the core information technology of the intelligent Internet of Things is gradually reshaping the traditional face of the power industry. Therefore, the correct operation of electrical equipment in the power grid has received great attention. High-voltage electrical equipment such as disconnectors will be attacked by wind, rain and sun, which will lead to various mechanical failures such as mechanism jamming, mechanism loosening, and inadequate opening and closing, which will greatly affect the normal operation of the power system and the safety of maintenance personnel.

[0003] With the continuous deepening of the research on disconnectors, there are now detailed analyses and data on vibration signals during mechanical structure failures of disconnectors, but there is a common problem of low fault identification accuracy. Summary of the invention

[0004] The present invention provides an isolating switch fault diagnosis method based on tFLO-VMD-SVM to solve the problem of low accuracy in isolating switch fault diagnosis in the prior art.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for isolating switch fault diagnosis based on tFLO-VMD-SVM comprises the following steps:

[0007] Step 1, obtaining a vibration signal on the surface of the isolating switch;

[0008] Step 2: Decompose the vibration signal obtained in step 1 into multiple layers of intrinsic mode function components by using the VMD algorithm. In the decomposition process of the VMD algorithm, the t-distributed perturbation lizard optimization algorithm is used to find the optimal alpha parameter and k parameter in the VMD algorithm, thereby obtaining the intrinsic mode function components of each layer under the optimal alpha parameter and k parameter;

[0009] Step 3, calculating the fuzzy spread entropy of each layer of the intrinsic mode function component obtained in step 2, and calculating the corresponding fine composite multi-scale fuzzy spread entropy based on the fuzzy spread entropy of each layer of the intrinsic mode function component, and constructing a multivariate feature fusion matrix by the fine composite multi-scale fuzzy spread entropy of the intrinsic mode function components of each layer;

[0010] Step 4: obtain a reconstructed space matrix based on the multivariate feature fusion matrix described in step 3, and divide the obtained reconstructed space matrix into a training set and a test set;

[0011] Step 5: Generate a support vector machine, input the training set obtained in Step 4 into the support vector machine for training, and use the whale optimization algorithm to optimize the c parameter and g parameter of the support vector machine during training. Thus, a support vector machine with optimal c parameter and g parameter is obtained through training and used as the fault diagnosis model.

[0012] Step 6: Input the test set obtained in Step 4 into the fault diagnosis model obtained in Step 5, and the fault diagnosis result of the mechanical structure of the disconnector is output by the fault diagnosis model.

[0013] Further, in the VMD algorithm of Step 2, first perform IMF component decomposition, and construct a VMD constrained variational problem with the goal of minimizing the sum of the bandwidths of the center frequencies of each IMF; then transform the VMD constrained variational problem into an unconstrained problem and solve it by the alternating direction multiplier method to obtain the iterative expression of the IMF component in the frequency domain; finally, perform iterative solution based on the iterative expression to obtain the final IMF component.

[0014] Further, the t-distribution perturbed frilled lizard optimization algorithm described in Step 2 introduces t-distribution perturbation in the global search stage of the frilled lizard optimization algorithm to interfere with and improve the update of the global search position.

[0015] Further, in Step 4, the PCA algorithm is used to obtain the reconstruction space matrix based on the multi-feature fusion matrix.

[0016] The present invention uses the frilled lizard optimization algorithm optimized by the t-distribution perturbation strategy to improve the variational mode decomposition method (VMD algorithm), constructs a feature matrix based on the fuzzy dispersion entropy and multi-scale entropy of the intrinsic mode function components decomposed by the VMD algorithm, trains a support vector machine optimized by the whale optimization algorithm using the feature matrix, and thus obtains a fault diagnosis model for fault diagnosis of vibration signals under different faults of the disconnector, which has important significance for solving the safety problem of the disconnector.

[0017] The method proposed by the present invention is applicable to the fault diagnosis of the mechanical structure of the disconnector, improves the iteration speed and convergence, and has a higher fault recognition accuracy for the disconnector. Description of the Drawings

[0018] Figure 1 It is a flowchart of an embodiment of the present invention.

[0019] Figure 2 It is a diagram of the original signal and the decomposed signal of the VMD decomposition of the specified signal in an embodiment of the present invention.

[0020] Figure 3 It is a comparison of the convergence of the PSO-VMD algorithm, the FLO-VMD algorithm, and the tFLO-VMD algorithm of an embodiment of the present invention.

[0021] Figure 4 This is a scatter plot of the PCA algorithm in the embodiment of the present invention for reducing the dimension of the multi - feature entropy fusion matrix to three dimensions.

[0022] Figure 5 This is the accuracy confusion matrix of the fault diagnosis of the disconnector fault data set using the WOA - SVM algorithm in the embodiment of the present invention. Detailed implementation manners

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] As Figure 1 shown, this embodiment discloses a disconnector fault diagnosis method based on tFLO - VMD - SVM, including the following steps:

[0025] Step 1: Obtain the vibration signal on the surface of the disconnector through the vibration signal test points arranged on the surface of the disconnector.

[0026] Step 2: Use the VMD algorithm to decompose the vibration signal obtained in Step 1 into multiple intrinsic mode function components. During the decomposition process of the VMD algorithm, the frilled lizard optimization algorithm (FLO) with t - distribution perturbation is used to find the optimal alpha parameter and k parameter in the VMD algorithm, thereby forming the tFLO - VMD algorithm to obtain each layer of intrinsic mode function components under the optimal alpha parameter and k parameter.

[0027] In this embodiment, the process of the VMD algorithm for decomposing the vibration signal into intrinsic mode function (IMF) components is as follows:

[0028] (A1) First, perform IMF component decomposition. The IMF component u k (t) decomposed by VMD can be expressed as shown in formula (1):

[0029] u k (t)=A k (t)cos[φ k (t)] (1)

[0030] In formula (1): A k (t) is the amplitude of this IMF signal; φ k (t) is the instantaneous phase; t is time.

[0031] (A2) The VMD constraint requires that the sum of the bandwidths of the center frequencies of each IMF is minimized. Therefore, the VMD constrained variational problem is constructed as shown in formulas (2) and (3):

[0032]

[0033] In formulas (2) and (3): {u k}\({\}\) represents the set of IMF component signals for each layer; \({\omega \}\) k}\({\}\) represents the set of center frequencies of the IMF components for each layer; \(*\) is the convolution operation; \(\nabla\) is the gradient operation; \(\delta(t)\) is the Dirac function; \(j\) represents the imaginary unit; \(f(t)\) represents expressing the IMF signals for each layer as a function of time.

[0034] (A3) Introduce the augmented Lagrangian function to transform equations (2) and (3) into an unconstrained problem, as shown in formula (4):

[0035]

[0036] In formula (4): \(\alpha\) is the penalty factor; \(L(\cdot)\) is the Lagrangian operator; \(\langle \cdot, \cdot \rangle\) is the inner product operation; \(\lambda\) is the Lagrange multiplier; \(\lambda(t)\) is the function of the Lagrangian with respect to time \(t\).

[0037] (A4) Solve equation (4) by the alternating direction method of multipliers to obtain the IMF component \(u\) k (t) in the frequency domain \(u\) k (\(\omega\)) The iterative expression of \(u\) k,n+1 (\(\omega\)) is as shown in formula (5):

[0038]

[0039] In formula (5): \(\omega\) is the frequency; \(\omega\) k,n is the frequency after the \(n\)th iteration for \(k\) modes; \(u\) k,n+1 (\(\omega\)) is the mode component after the \(n\)th iteration for \(k\) modes; \(u\) i (\(\omega\)) is the sum of the mode components; \(f(\omega)\) is the function of the mode component with respect to the frequency; \(\lambda\) n (\(\omega\)) is the function of the Lagrangian function with respect to the frequency.

[0040] (A5) Perform iterative solution based on the iterative formula shown in formula (5) until the iterative termination condition of equation (6) is satisfied, thereby obtaining the final IMF component. Formula (6) is as follows:

[0041]

[0042] In formula (6): \(\varepsilon\) is a manually set threshold, usually \(10\) -6 .

[0043] Thus, in this embodiment, according to steps (A1)-(A5), the VMD algorithm is implemented to decompose the vibration signal into multi-layer IMF components. The obtained signal decomposition diagram is as Figure 2 shown. It can be seen from Figure 2 that the VMD algorithm can effectively decompose the vibration signal according to the set parameters and there is no mode mixing phenomenon.

[0044] In this embodiment, when using the VMD algorithm for IMF component decomposition, the umbrella lizard optimization algorithm (FLO) with t-distribution perturbation is used to optimize the alpha parameter and k parameter in the decomposition process of the VMD algorithm to find the optimal alpha parameter and k parameter. Among them, the alpha parameter refers to the key bandwidth constraint parameter in the VMD algorithm, and the k parameter is the key parameter that determines the number of modal components after decomposition in the VMD algorithm. The process of using the umbrella lizard optimization algorithm with t-distribution perturbation is as follows:

[0045] (B1) Development stage, i.e., global search stage:

[0046] In the first stage of the umbrella lizard optimization algorithm (FLO), i.e., the global search stage, the positions of population individuals in the problem solution space are updated according to the hunting strategy of the umbrella lizard. In the design of the FLO algorithm, for each umbrella lizard, the positions of other population members with better objective function values are regarded as prey positions. Accordingly, the candidate prey position set of each umbrella lizard is determined using formula (7), and formula (7) is as follows:

[0047] CP i ={X k :F k <F i and k≠i},where i=1,2,…,N and k∈{1,2,…,N}(7)

[0048] In formula (7): CP i is the candidate prey set of the i-th umbrella lizard; X k is the population member with a better objective function value than the i-th umbrella lizard; Fk and Fi are the objective functions of the k-th and i-th umbrella lizard algorithms respectively; N represents the total population of the umbrella lizard algorithm.

[0049] In the design of the FLO algorithm, it is assumed that the umbrella lizard randomly selects one of the candidate preys and attacks it. Based on the modeling of the umbrella lizard moving towards the selected prey, the new position of each individual in the population is calculated using formula (7). Then, if the objective function value is better, the previous position of the corresponding individual is replaced with this new position using formulas (8), (9), and formulas (10), (11) are as follows:

[0050]

[0051] In formulas (8), (9): represents the new proposed position for the i-th umbrella lizard based on the first stage of the FLO algorithm; represents the d-th dimension of the umbrella lizard individual; this algorithm simulates the characteristics of the umbrella lizard preying at different positions and needs to update the position to determine the optimal preying position of the umbrella lizard, i.e., the optimal fitness function of the algorithm. represents the objective function value of the algorithm at position 1 (i.e., the first position Place1 where the algorithm starts to iterate); r is a random number with a normal distribution in the interval [0, 1]; SP i,d represents the d-th dimension of the prey selected by the i-th frill-necked lizard; I is a number randomly drawn from the set {1, 2}; N represents the total number of frill-necked lizards; m is the number of decision variables given; x i,d represents the individual position of the i-th frill-necked lizard; X i represents the original position before update.

[0052] For the frill-necked lizard optimization algorithm, this algorithm can perform a global search for prey within a corresponding range in the global search stage, but it may also lose the best individual information due to excessive jumping. In order to allow as many other frill-necked lizards to pass as possible, in the first stage of the FLO algorithm, namely the global search stage, a t-distribution perturbation alarm operator is used to interfere with and improve the random global search position update formula (8) of the frill-necked lizard, and the global search position update formula using t-distribution perturbation is obtained as shown in formula (10):

[0053]

[0054] In formula (10): is the new position after mutation, x i,d is the individual position of the i-th frill-necked lizard, and t-distrub(·) is the adaptive t-distribution perturbation operator.

[0055] Thus, in this embodiment, formula (10) is finally adopted as the global search position update formula in the global search stage of the FLO algorithm.

[0056] (B2) Exploration stage, i.e., local search stage:

[0057] In the second stage of the FLO algorithm, according to the strategy that the frill-necked lizard retreats to the top of the tree after eating, the positions of the population individuals in the solution space are updated. Based on the modeling of the movement of the frill-necked lizard climbing to the top of the nearby tree, formula (11) is used to calculate the new position of each population individual. Then, if this new position improves the objective function value, formula (14) is used to replace the previous position of the corresponding individual. Formulas (11) and (12) are as follows:

[0058]

[0059] In formulas (11) and (12): represents the new proposed position for the i-th frill-necked lizard based on the second stage of FLO; represents the d-th dimension of the frill-necked lizard individual; Give the objective function value of the algorithm at position 2 (i.e., the second position Place2 where the algorithm starts to iterate); t represents the iteration counter of the algorithm; T describes the maximum number of iterations of the algorithm; ub d represents the upper limit of the specified update parameter; lb d represents the lower limit of the specified update parameter.

[0060] In this embodiment, when performing IMF component decomposition using the VMD algorithm according to steps (A1)-(A5), the frilled lizard optimization algorithm (FLO) with t-distribution perturbation is used according to steps (B1)-(B2) to find the optimal alpha parameter and k parameter during the decomposition process of the VMD algorithm, and then the intrinsic mode function components of each layer under the optimal alpha parameter and k parameter are obtained. As Figure 3 shown, it is the performance graph of the frilled lizard optimization algorithm with t-distribution perturbation adopted in this embodiment. By Figure 3 it can be seen that the frilled lizard optimization algorithm with t-distribution perturbation adopted in this embodiment has superiority in terms of convergence and iteration speed.

[0061] Step 3: Calculate the fuzzy dispersion entropy of the intrinsic mode function components of each layer obtained in Step 2, and calculate the corresponding refined composite multi-scale fuzzy dispersion entropy based on the fuzzy dispersion entropy of the intrinsic mode function components of each layer. Construct a multivariate feature fusion matrix from the refined composite multi-scale fuzzy dispersion entropy of the intrinsic mode function components of each layer.

[0062] In this embodiment, the calculation process of the fuzzy dispersion entropy is as follows:

[0063] (C1) Divide the IMF data h = {h1, h2,..., h L} with length L obtained in Step 2 into equal distances. The number of divisions is defined as τ, and the starting point of each small segment of data is defined. The relationship between its k-th coarse-grained sequence and the original data h is shown in formula (13):

[0064]

[0065] In formula (13): is the k-th coarse-grained sequence; h b is the b-th data in the data set h.

[0066] (C2) Calculate the probability of each coarse-grained dispersion pattern π and its average value, specifically as follows:

[0067] C2.1) Map f k,j to a new sequence x = {x j , j = 1, 2,..., N}. Among them, f k,j represents the coarse-grained sequence; x j ∈(0, 1), x jThe calculation formula is as shown in formula (14):

[0068]

[0069] In formula (14): σ represents the standard deviation; μ represents the mean; t represents the simulated time step.

[0070] C2.2) Map x j to the range of [1, 2, …, c] to obtain a new sequence as shown in formula (15):

[0071]

[0072] In formula (15): c represents the number of classes in the time series.

[0073] C2.3) Use the fuzzy membership function to calculate the membership degree relative to the q-th class. The fuzzy functions are as shown in formulas (16), (17), and (18):

[0074]

[0075] In formulas (16), (17), and (18): q is an integer in [1, 2, …, c]; μ(·) is the membership function with respect to ; α is the independent variable of each membership function, that is, α is used as the independent variable in each membership function to represent

[0076] C2.4) Construct an embedded vector sequence from the two parameters m and d in formula (10) as shown in formula (19):

[0077]

[0078] In formula (19): i = {1, 2, …, N - (m - 1)d}, and N is the total population size.

[0079] C2.5) Calculate the scattering pattern denotes applying this function to v0, v1 …… v m-1 where v0, v1 …… v m-1 represent classes respectively; if } represent the embedded vector sequences constructed from the two parameters m and d in formula (10) respectively, then the corresponding scattering pattern is as shown in formula (20):

[0080]

[0081] In formula (20): is the membership degree of the pattern equal to the product of the membership degrees of the classes v0v1…v m-1 .

[0082] C2.6) Calculate the probability of the spreading pattern under all coarsened sequences as shown in formula (21):

[0083]

[0084] In formula (21): represents the membership degree of the spreading pattern of all sequences divided by the total number of embedded signals.

[0085] In this embodiment, based on the calculated fuzzy spreading entropy, calculate the refined composite multi-scale fuzzy spreading entropy RCMFDE, as shown in formula (22):

[0086]

[0087] In formula (22): is the average value of the probabilities of the spreading pattern π of the coarsened sequence ; is the probability of the k-th spreading pattern.

[0088] In this embodiment, finally, construct a multi-feature fusion matrix based on the refined composite multi-scale fuzzy spreading entropy of each layer of intrinsic mode function components.

[0089] Step 4: Obtain a reconstructed space matrix based on the multi-feature fusion matrix described in Step 3, and divide the obtained reconstructed space matrix into a training set and a test set.

[0090] In this embodiment, use the PCA algorithm to obtain a reconstructed space matrix based on the multi-feature fusion matrix. Specifically, use the PCA algorithm to reduce the dimension of each layer of IMF feature matrix, and select appropriate basic components without losing the initial information. Input a data set containing τ samples and n dimensions (types of faults), and reduce the dimension of τ. Write the sample set (a feature matrix of τ×n) as matrix X, as shown in formula (23):

[0091]

[0092] Reduce the matrix X to q dimensions (q < τ) to obtain a sample set

[0093] In this embodiment, the three-dimensional scatter plot of the multi-feature fusion matrix after PCA dimensionality reduction is as Figure 4 shown. From Figure 4 it can be seen that after dimensionality reduction, the features of the signal can be represented by a matrix with a smaller dimension, saving the computational cost and improving the diagnostic efficiency.

[0094] Step 5: Generate a support vector machine (SVM), and input the training set obtained in Step 4 into the support vector machine for training.

[0095] During training, the whale optimization algorithm (WOA algorithm) is used to optimize the c parameter and g parameter of the support vector machine (SVM). The WOA algorithm is a new swarm intelligence optimization algorithm, which is inspired by two unique hunting behaviors of whales: surrounding prey and separating bubble nets to drive away prey. By simulating the above two behaviors, the WOA algorithm realizes global search in the early stage and local search in the later stage to obtain the optimal solution. Therefore, in this embodiment, the above WOA algorithm and SVM are combined to form the WOA-SVM algorithm to optimize the c and g parameters in the SVM to improve the classification accuracy. Among them, the c parameter is an important regularization parameter in the support vector machine (SVM), and the g parameter is the gamma parameter of the RBF (radial basis function) kernel function used in the support vector machine (SVM).

[0096] Thus, in this embodiment, the support vector machine SVM is optimized by the whale optimization algorithm WOA during training to obtain the support vector machine SVM with the optimal c parameter and g parameter as the fault diagnosis model.

[0097] Step 6: Input the test set obtained in Step 4 into the fault diagnosis model obtained in Step 5, and the fault diagnosis result of the mechanical structure of the disconnector is output by the fault diagnosis model. As Figure 5 shown, it is the confusion matrix diagram after the method of this embodiment classifies and diagnoses faults. From Figure 5 it can be seen that the diagnostic method proposed in this embodiment has good applicability to the disconnector.

[0098] The preferred embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. The embodiments described in the present invention are only descriptions of the preferred embodiments of the present invention, and do not limit the concept and scope of the present invention. Among the various specific technical features described in the above specific embodiments, they can be combined in any suitable way without contradiction. As long as such a combination does not violate the idea of the present invention, it should also be regarded as the content disclosed in this disclosure. To avoid unnecessary repetition, the present invention does not separately describe various possible combination methods.

[0099] The present invention is not limited to the specific details in the above embodiments. Without departing from the technical concept of the present invention and within the premise of not deviating from the design idea of the present invention, various modifications and improvements made by those skilled in the art to the technical solution of the present invention shall fall within the protection scope of the present invention. The technical content claimed by the present invention has been fully recorded in the claims.

Claims

1. A disconnector fault diagnosis method based on tFLO-VMD-SVM, characterized in that, Including the following steps: Step 1: Obtain the vibration signal on the surface of the disconnector; Step 2: Use the VMD algorithm to decompose the vibration signal obtained in Step 1 into multiple intrinsic mode function components. During the decomposition process of the VMD algorithm, the frilled lizard optimization algorithm with t-distribution perturbation is used to find the optimal alpha parameter and k parameter in the VMD algorithm, and thus the intrinsic mode function components at the optimal alpha parameter and k parameter are obtained; Step 3: Calculate the fuzzy dispersion entropy of each layer of intrinsic mode function components obtained in Step 2, and calculate the corresponding refined composite multi-scale fuzzy dispersion entropy based on the fuzzy dispersion entropy of each layer of intrinsic mode function components. The multi-element feature fusion matrix is constructed from the refined composite multi-scale fuzzy dispersion entropy of each layer of intrinsic mode function components; Step 4: Obtain the reconstructed space matrix based on the multi-element feature fusion matrix described in Step 3, and divide the obtained reconstructed space matrix into a training set and a test set; Step 5: Generate a support vector machine, input the training set obtained in Step 4 into the support vector machine for training, and use the whale optimization algorithm to optimize the c parameter and g parameter of the support vector machine during training. Thus, the support vector machine with the optimal c parameter and g parameter is obtained through training and used as the fault diagnosis model; Step 6: Input the test set obtained in Step 4 into the fault diagnosis model obtained in Step 5, and the fault diagnosis result of the mechanical structure of the disconnector is output by the fault diagnosis model.

2. The method for isolating switch fault diagnosis based on tFLO-VMD-SVM according to claim 1, characterized in that, In the VMD algorithm of Step 2, first, the IMF component decomposition is performed, and the VMD constrained variational problem is constructed with the goal of minimizing the sum of the bandwidths of the center frequencies of each IMF; then the VMD constrained variational problem is transformed into an unconstrained problem and solved by the alternating direction multiplier method to obtain the iterative expression of the IMF component in the frequency domain; finally, iterative solution is performed based on the iterative expression to obtain the final IMF component.

3. The method for isolating switch fault diagnosis based on tFLO-VMD-SVM according to claim 1, wherein, The frilled lizard optimization algorithm with t-distribution perturbation described in Step 2 is to introduce t-distribution perturbation in the global search stage of the frilled lizard optimization algorithm to interfere with and improve the update of the global search position.

4. A disconnector fault diagnosis method based on tFLO-VMD-SVM according to claim 1, characterized in that, In Step 4, the PCA algorithm is used to obtain the reconstructed space matrix based on the multi-element feature fusion matrix.

Images