A method and device for identifying ground fault of stator winding of a hydro-generator

Abstract

This invention discloses a method and apparatus for identifying stator winding grounding faults in hydro-generators, relating to the field of hydro-generator fault identification technology. The method includes acquiring the zero-sequence current signal during a single-phase grounding fault in the hydro-generator; using the minimization of envelope entropy as the fitness function, the IDBO algorithm is used to adaptively optimize the number of modes and penalty factor in variational mode decomposition (VMD) to obtain the optimal parameter combination; wherein the IDBO algorithm is obtained by fusing Logistic-Tent chaotic mapping, exponential decay convergence factor, and random perturbation mechanism; based on the zero-sequence current signal and the optimal parameter combination, signal decomposition processing is performed to obtain multiple intrinsic mode functions (IMF) components; the three components with the highest correlation coefficients are selected as effective fault features, their sample entropy is calculated, and the sample entropies are combined into a feature vector representing the fault state; the feature vector is used as an input layer variable and imported into a support vector machine for fault identification.

Description

Technical Field

[0001] This invention relates to a method and device for identifying grounding faults in the stator winding of a hydro-generator, belonging to the field of hydro-generator fault identification technology. Background Technology

[0002] Due to their large unit capacity, complex structure, and harsh operating environment, hydro-generators are prone to single-phase grounding faults in their stator windings. Early-stage signals of this type of fault are weak and often masked by mechanical vibration, electromagnetic interference, and background noise, necessitating effective decomposition and feature extraction of the operating electrical signals.

[0003] To address the need for extracting this weak fault signal, Variational Modal Decomposition (VMD), as an adaptive signal decomposition method based on a variational framework, can effectively alleviate the mode aliasing and endpoint effects problems existing in traditional decomposition methods. However, the number of decomposition modes K and the penalty factor α have a significant impact on the signal decomposition results. In traditional decomposition methods, these two parameters often rely on expert experience or repeated experiments to obtain. For signals with different operating conditions and different fault degrees, there is no set of universal parameters that can adapt to all situations. Therefore, it is easy to lead to unstable decomposition results, poor repeatability, and difficulty in ensuring that the purest fault feature components can always be extracted. Summary of the Invention

[0004] The purpose of this invention is to provide a method and device for identifying grounding faults in the stator winding of a hydro-generator. By improving the dung beetle optimization algorithm to optimize the parameters of variational mode decomposition (VMD), adaptive selection of the mode number K and penalty factor α is achieved, which can effectively improve the convergence speed of the algorithm and avoid getting trapped in local optima.

[0005] To achieve the above objectives, the present invention is implemented using the following technical solution.

[0006] On one hand, the present invention provides a method for identifying ground faults in the stator winding of a hydro-generator, comprising:

[0007] Acquire the zero-sequence current signal when a single-phase ground fault occurs in a hydro-generator;

[0008] Using the minimization of envelope entropy as the fitness function, the IDBO algorithm is used to adaptively optimize the number of modes and the penalty factor of variational mode decomposition (VMD) to obtain the optimal parameter combination; wherein, the IDBO algorithm is obtained by integrating Logistic-Tent chaotic mapping, exponential decay convergence factor and random perturbation mechanism.

[0009] Based on the zero-sequence current signal and the optimal parameter combination, signal decomposition processing is performed to obtain multiple intrinsic mode function (IMF) components.

[0010] The three components with the highest correlation coefficients from the multiple intrinsic mode functions (IMF) are selected as valid fault features.

[0011] Calculate the sample entropy of the effective fault features, and combine the sample entropy into a feature vector characterizing the fault state;

[0012] The feature vectors are used as input layer variables and imported into a support vector machine for fault identification, outputting a category label indicating whether a single-phase ground fault exists in the stator winding of the hydro-generator.

[0013] Furthermore, before acquiring the zero-sequence current signal during a single-phase ground fault of the hydro-generator, a ground fault model of the hydro-generator is established, and a grounding point is set in the external circuit of the stator winding of the hydro-generator ground fault model to simulate the ground fault condition; the zero-sequence current signal is acquired under the simulated ground fault condition.

[0014] Furthermore, the Logistic-Tent chaotic mapping is used to generate a uniformly distributed initial population, and the specific calculation expression is as follows:

[0015] (1),

[0016] in, The value is the chaotic sequence value of the current iteration, with a range of [0,1]. r is a control parameter, and mod is the modulo operation used to map the result to the interval [0,1].

[0017] Furthermore, the exponentially decaying convergence factor is used to nonlinearly and dynamically adjust the balance between global exploration and local exploitation capabilities during the iteration process, thereby accelerating the algorithm's convergence speed. The specific calculation expression is as follows:

[0018] (2),

[0019] Where R is the exponential convergence factor, and t is the current iteration number. The maximum number of iterations, with a value ranging from 1 to t. .

[0020] Furthermore, the implementation process of the random perturbation mechanism includes:

[0021] If the optimal fitness value of the population does not improve in M ​​consecutive iterations, a random perturbation is applied to the individual position X. The specific calculation expression is as follows:

[0022] (3),

[0023] in, The scaling factor has a value range of [1, 10]; {t1, t2, ..., t n} represents the current iteration number, n represents the iteration number, n≤M; X(t) represents the random perturbation applied to the individual position X in the t-th iteration; r(t) n ) is the t-th n The chaotic sequence value for the next iteration is calculated using the following expression:

[0024] (4),

[0025] The normalized mapping formula is used to constrain the position of the perturbed individuals. The specific calculation expression is as follows:

[0026] (5),

[0027] in, This represents the position of the i-th dimension random walk variable in the t-th iteration. and Let be the global minimum and maximum values ​​of the i-th dimension random walk variable, respectively. and Let represent the local minimum and maximum values ​​of the i-th dimension random walk variable in the t-th iteration, respectively.

[0028] Furthermore, the expression for calculating the envelope entropy is as follows:

[0029] (6),

[0030] in, Envelope entropy;

[0031] P k This is the normalized form of the decomposed signal a(k). K represents the number of intrinsic mode functions (IMFs) obtained from the decomposition; k represents the component of the k-th IMF.

[0032] H[x(k)] is the Hilbert transform of the original time-domain signal x(k).

[0033] Furthermore, obtaining the optimal parameter combination includes:

[0034] The zero-sequence current signal is decomposed into K intrinsic mode functions (IMF) components, and each IMF component is subjected to a Hilbert transform to obtain a one-sided spectrum. The specific calculation expression is as follows:

[0035] (7),

[0036] Where t is the current iteration number, Let be the modal component of the k-th intrinsic mode function (IMF) after t iterations. is the unit impulse function; k is the k-th intrinsic mode function (IMF) component; j is the imaginary unit;

[0037] modal components The spectrum is transferred to the corresponding fundamental frequency band, and the specific calculation expression is as follows:

[0038] (8),

[0039] in, for The center frequency; The symbol for convolution;

[0040] For the modal components transferred to the corresponding fundamental frequency band The spectrum is used to construct a constrained variational model, and the specific calculation expression of the constrained variational model is as follows:

[0041] (9),

[0042] in, This indicates taking the partial derivative with respect to t. for Norm, The zero-sequence current signal obtained after t iterations is the time-domain representation.

[0043] Introducing a secondary penalty term and Lagrange multipliers Construct the corresponding unconstrained augmented Lagrangian function, and calculate its expression as follows:

[0044] (10)

[0045] The alternating direction multiplier method is used to solve for the saddle point corresponding to the augmented Lagrangian function. The optimal solution of the variational model is gradually approximated through alternating iterations. The specific calculation expression is as follows:

[0046] (11),

[0047] (12),

[0048] (13)

[0049] in, Here, is the bandwidth parameter, and n is the number of iterations. For Lagrange multipliers, This refers to the k-th intrinsic mode function (IMF) component during the (n+1)-th iteration. This is the frequency domain representation of the zero-sequence current signal obtained after t iterations. , , They are , as well as Obtained through Fourier transform, Update the step size and repeat the above steps until the following convergence condition is met to calculate the expression:

[0050] (14)

[0051] in, This is the preset convergence accuracy value.

[0052] Furthermore, the expression for calculating the correlation coefficient is as follows:

[0053] (15)

[0054] Where i = 1, 2, ..., N, i is the index of the discrete sampling point, and N is the signal length; The correlation coefficient; u k (i) represents the k-th intrinsic mode function (IMF) component of the i-th sampling point; y is the average value of the k-th intrinsic mode function (IMF) component; y is the value of the zero-sequence current signal at the i-th sampling point. The mean value of the zero-sequence current signal;

[0055] The intrinsic mode function (IMF) components are sorted in descending order of their absolute correlation coefficients, and the top three IMF components are selected as effective fault feature components. The sample entropy of the effective fault feature components is combined to construct the following process:

[0056] The intrinsic mode function (IMF) components are represented as discrete sequences, and the specific expression of the discrete sequences is as follows:

[0057] (16)

[0058] Based on the discrete sequence, the phase space vector used to calculate the sample entropy is reconstructed, and the specific calculation expression is as follows:

[0059] (17)

[0060] Where m is the embedding dimension, and x(i) is the i-th sampling point of the intrinsic mode function (IMF) component;

[0061] Based on the phase space vectors, calculate the distance between any two vectors u(i) and u(l), using the following formula:

[0062] (18)

[0063] Where x(l) is the l-th sampling point of the intrinsic mode function (IMF) component; q is the element index within the interval [0, m-1].

[0064] The expression for calculating the sample entropy is:

[0065] (19)

[0066] Where ɛ is the similarity tolerance, SampEn(m, ) represents the sample entropy. This is the preset convergence accuracy value;

[0067] B i (m) Let be the matching probability in m dimensions. N is the IMF signal length, N ɛ The number of vectors that satisfy d[u(i),u(j)]<ɛ;

[0068] B m Let be the average matching probability in m dimensions. .

[0069] Furthermore, the support vector machine achieves fault identification through an optimal hyperplane classification function, the calculation expression of which is:

[0070] (20)

[0071] in, For the classification output results of the support vector machine, It is a Lagrange multiplier, and C is the penalty factor; For kernel functions; The classification threshold is determined for the training samples; sgn is the sign function used to output the category label of the classification result; the category label is used to indicate whether there is a single-phase grounding fault in the stator winding of the hydro-generator.

[0072] On the other hand, the present invention also provides a device for identifying ground faults in the stator winding of a hydro-generator, comprising:

[0073] The signal acquisition module is configured to acquire the zero-sequence current signal when a single-phase ground fault occurs in the hydro-generator.

[0074] The optimal module is configured to: use the minimization of envelope entropy as the fitness function, and use the IDBO algorithm to adaptively optimize the number of modes and the penalty factor of variational mode decomposition (VMD) to obtain the optimal parameter combination; wherein, the IDBO algorithm is obtained by integrating Logistic-Tent chaotic mapping, exponential decay convergence factor and random perturbation mechanism.

[0075] The decomposition processing module is configured to: perform signal decomposition processing based on the zero-sequence current signal and the optimal parameter combination to obtain multiple intrinsic mode function (IMF) components;

[0076] The feature selection module is configured to select the three most correlated components from the plurality of intrinsic mode functions (IMF) as valid fault features.

[0077] The vector construction module is configured to: calculate the sample entropy of the effective fault features and combine the sample entropy into a feature vector characterizing the fault state;

[0078] The fault identification module is configured to: import the feature vector as an input layer variable into a support vector machine for fault identification, and output a category label indicating whether there is a single-phase grounding fault in the stator winding of the hydro-generator.

[0079] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0080] This invention proposes an improved dung beetle optimization algorithm, IDBO, which enhances the uniformity of the initial population distribution by utilizing a Logistic-Tent composite chaotic mapping. The linear convergence factor is improved to an exponentially decaying form, dynamically balancing global exploration and local exploitation in a nonlinear manner. A random perturbation mechanism is introduced to effectively escape local optima when the population stagnates. The IDBO algorithm is used to adaptively optimize the parameters of Variational Modal Decomposition (VMD). Applying the optimized parameter combination to VMD decomposition of the fault signal yields Intrinsic Mode Function (IMF) components with a clearer center frequency distribution and significantly reduced mode aliasing. Furthermore, a threshold is set based on the correlation coefficient between the IMF components and the original signal to eliminate low-correlation noise components, thereby reducing interference. Finally, the feature vector constructed from the selected effective IMF components is input into a Support Vector Machine (SVM) model to achieve accurate diagnosis of stator single-phase grounding faults. Attached Figure Description

[0082] Figure 1The diagram shows a flowchart of the method for identifying grounding faults in the stator winding of a hydro-generator provided by the present invention.

[0083] Figure 2 The diagram shows a flowchart of the IDBO algorithm provided by this invention.

[0084] Figure 3 The diagram shown is a schematic of the support vector machine provided by this invention. Detailed Implementation

[0086] Example 1

[0087] This embodiment uses a large hydro-generator to identify a phase winding grounding fault. The basic parameters of this generator are shown in Table 1: Table 1: Basic Parameters of a Large Hydro-Generator

[0088]

[0089] See Figure 1 This embodiment provides a method for identifying grounding faults in the stator winding of a hydro-generator, the method including the following steps:

[0090] Step S1: Obtain the zero-sequence current signal when a single-phase ground fault occurs in a large hydro-generator;

[0091] In this embodiment, a ground fault model of a large hydro-generator is established and the corresponding circuit structure diagram is drawn. A grounding point is set in the external circuit of the stator winding of the model to simulate the ground fault condition. The zero-sequence current signal is collected during the ground fault condition. The sampling frequency is 5 kHz and the sampling period is 0.0002 s to ensure the time domain continuity and frequency domain integrity of the signal, providing basic data for subsequent feature extraction and fault diagnosis.

[0092] Step S2: Using the minimization of envelope entropy as the fitness function, the IDBO algorithm is used to adaptively optimize the number of modes and the penalty factor of variational mode decomposition (VMD) to obtain the optimal parameter combination; wherein, the IDBO algorithm is obtained by integrating Logistic-Tent chaotic mapping, exponential decay convergence factor and random perturbation mechanism.

[0093] In this embodiment, the DBO algorithm is improved by integrating three synergistic strategies: Logistic-Tent composite chaotic mapping, exponentially decaying convergence factor, and random perturbation mechanism, forming the IDBO algorithm. The DBO algorithm is an intelligent optimization tool that simulates dung beetle behavior. It can efficiently search the parameter space for the set of parameters that makes the fitness function reach its optimal value. In this embodiment, this is the required set of parameters: the number of modes K and the penalty factor α.

[0094] The Logistic-Tent composite chaotic mapping combines the Logistic and Tent mappings to generate a more evenly distributed and stable initial population. It increases the diversity of the global search, reduces the risk of all individuals prematurely clustering in suboptimal regions, and lowers the probability of getting trapped in local optima from the outset. The computational expression for the Logistic-Tent composite chaotic mapping is shown below:

[0095] (1),

[0096] in, The value is the chaotic sequence value of the current iteration, with a range of [0,1]. r is a control parameter, and mod is the modulo operation used to map the result to the interval [0,1].

[0097] The exponentially decaying convergence factor is used to enhance the global exploration and local exploitation capabilities of dynamic equilibrium algorithms. In the early stages of iteration, a larger convergence factor and a faster rate of decay encourage the algorithm to perform a rapid, large-scale global search. In the later stages of iteration, a smaller convergence factor and a more gradual rate of decay allow the algorithm to perform smaller-scale local exploitation. This significantly reduces invalid searches, thus accelerating the algorithm's convergence speed. The expression for calculating the exponentially decaying convergence factor is shown below:

[0098] (2),

[0099] Where t is the current iteration number, The maximum number of iterations, with a value ranging from 1 to t. . It is used to control the convergence factor to gradually decay as the iteration progresses, and to normalize the iteration progress, so that the exponential decay smoothly decreases from 1 to e-1.

[0100] The random perturbation mechanism is used to perform random walks when the algorithm stagnates, avoiding premature convergence and improving optimization stability. Specifically, when the optimal fitness value of the population has not improved in M ​​consecutive iterations, a random perturbation is applied to the individual's position X. The specific calculation expression is as follows:

[0101] (3),

[0102] in, is a scaling factor with a value range of [1, 10], used to adjust the position range of the initial population, mapping the standardized sequence generated by the Logistic-Tent chaotic mapping to the search interval allowed by the algorithm's decision variables. {t1, t2, ..., t n} represents the current iteration number, n represents the iteration number, n≤k; X(t) represents the random perturbation applied to the individual position X in the t-th iteration; r(t) n ) is the t-th n The chaotic sequence value for the next iteration is calculated using the following expression:

[0103] (4),

[0104] To ensure that the random perturbation is searched precisely within a local range, a normalized mapping formula is used to constrain the position of the individual after the perturbation. The specific calculation expression is shown in formula (5):

[0105] (5),

[0106] in, This represents the position of the i-th dimension random walk variable in the t-th iteration. and Let be the global minimum and maximum values ​​of the i-th dimension random walk variable, respectively. and Let represent the local minimum and maximum values ​​of the i-th dimension random walk variable in the t-th iteration, respectively.

[0107] See Figure 2 In this embodiment, the purpose of VMD decomposition is to extract fault feature components with obvious impact or periodicity from a strong noise background, and envelope entropy can directly measure the significance and concentration of the extracted features. Therefore, this embodiment chooses envelope entropy as the fitness function. The calculation expression is as follows:

[0108] (6),

[0109] Among them, P k This is the normalized form of the decomposed signal a(k). K represents the number of intrinsic mode functions (IMFs) obtained from the decomposition; k represents the component of the k-th IMF.

[0110] H[x(k)] is the Hilbert transform of the original time-domain signal x(k).

[0111] In this embodiment, the specific process of performing variational mode decomposition (VMD) is as follows:

[0112] Define the number of modes K and the penalty factor. This parameter space; using the Logistic-Tent chaotic mapping, multiple initial individuals are uniformly generated within the parameter space to avoid population clustering in local regions, with each individual forming a group (K, ).

[0113] Input raw zero-sequence current signal The intrinsic mode function (IMF) is decomposed into K IMF components, and each IMF component is subjected to a Hilbert transform to obtain a one-sided spectrum. The specific calculation expression is as follows:

[0114] (7),

[0115] Where t is the current iteration number, Let be the modal component of the k-th intrinsic mode function (IMF) after t iterations. is the unit impulse function; k is the k-th intrinsic mode function (IMF) component; j is the imaginary unit;

[0116] modal components The spectrum is transferred to the corresponding fundamental frequency band, and the specific calculation expression is as follows:

[0117] (8),

[0118] in, for The center frequency; This is the convolution symbol.

[0119] Next, the modal components transferred to the corresponding fundamental frequency band are... A constrained variational model is constructed using the spectrum of the signal. The bandwidth is estimated by smoothing the demodulated signal using Gaussian smoothing. The objective is to minimize the sum of the estimated bandwidths of each mode function, while constraining the sum to equal the original signal. The variational model calculation expression is as follows:

[0120] (9),

[0121] in, This indicates taking the partial derivative with respect to t. for Norm, The zero-sequence current signal obtained after t iterations is the time-domain representation.

[0122] To solve the variational problem, a quadratic penalty term is introduced. With Lagrange multipliers The constrained problem is transformed into an unconstrained problem. The augmented Lagrangian function includes bandwidth terms, constraint terms, and a quadratic penalty term. This represents the bandwidth parameter, used to adjust the smoothness of the bandwidth for each mode. The specific calculation expression is as follows:

[0123] (10)

[0124] in, This is a secondary penalty item. It is a Lagrange multiplier.

[0125] The mode function is updated alternately using the alternating direction multiplier method. Center frequency and Lagrange multipliers The saddle point corresponding to the augmented Lagrangian function is found, and the optimal solution to the variational problem is gradually approximated through alternating iterations. The specific calculation expression is as follows:

[0126] (11),

[0127] (12),

[0128] (13)

[0129] in, Here, is the bandwidth parameter, and n is the number of iterations. For Lagrange multipliers, This refers to the k-th intrinsic mode function (IMF) component during the (n+1)-th iteration. This is the frequency domain representation of the zero-sequence current signal obtained after t iterations. , , They are , as well as Obtained through Fourier transform, Update the step size and repeat the above steps until the following convergence condition is met to calculate the expression:

[0130] (14)

[0131] in, This is the preset convergence accuracy value.

[0132] Finally, K updated intrinsic mode function (IMF) components are output, and their envelope entropy is calculated.

[0133] Optimize the parameters of Variational Mode Decomposition (VMD) using the IDBO algorithm:

[0134] Step S2a1: Calculate the exponential decay convergence factor using formula (2) and dynamically balance exploration and development: the exponential decay convergence factor is larger in the early stage of iteration, which encourages global exploration; the exponential decay convergence factor is smaller in the later stage of iteration, which encourages local development.

[0135] Step S2a2: Use the IDBO algorithm to simulate dung beetle behavior and update the position, i.e., adjust (K, ):The rolling behavior simulates the dung beetle rolling a dung ball and is responsible for long-distance global exploration to find new potential advantageous areas; the reproduction behavior simulates the dung beetle burying the dung ball and laying eggs, and is responsible for fine local development near the current position; the foraging behavior simulates the foraging of juvenile dung beetles and provides an auxiliary search strategy to enhance local search ability; the stealing behavior simulates some dung beetles stealing other people's dung balls, introducing randomness to increase population diversity and help jump out of local optima;

[0136] Step S2a3: After each group (K, ) executes the above four behaviors, recalculate the fitness of the new position, that is, the envelope entropy, and compare it with the envelope entropy of the old position for determination: if the new position is better, update the original position; if the old position is better, retain the original position and abandon this move;

[0137] Step S2a4: After determining the position, determine whether the search enters a stagnation state: if it enters a stagnation state, apply a small random perturbation to some individuals using formula (3), and use normalization mapping to ensure that the perturbed K and are still within the preset reasonable range; if it does not enter a stagnation state, determine whether the current iteration number t has reached the preset maximum iteration number T.

[0138] Step S2a5: If t < T, return to Step S1 to dynamically adjust the exponential decay convergence factor; if t ≥ T, output the (K, ) with the highest fitness.

[0139] The IDBO algorithm can integrate three major improvement mechanisms and optimize the VMD parameters with the envelope entropy as an index.

[0140] Step S3: According to the zero-sequence current signal, based on the optimal parameter combination, perform signal decomposition processing to obtain multiple intrinsic mode function IMF components;

[0141] Using the combination (K, ) of the optimal VMD parameter intrinsic mode component number K and bandwidth parameter optimized by the IDBO algorithm in Step S2, perform variational mode decomposition on the preprocessed zero-sequence current signal to obtain K intrinsic mode function IMF components.

[0142] Step S4: Select three IMF components with the highest correlation coefficients from multiple intrinsic mode function IMF components as effective fault features;

[0143] Calculate the correlation coefficient between each intrinsic mode function IMF component and the original undecomposed signal. The calculation expression of the correlation coefficient is as follows:

[0144] (15),

[0145] Where i = 1, 2, ..., N, i is the index of the discrete sampling point, and N is the signal length; The correlation coefficient; u k (i) represents the k-th intrinsic mode function (IMF) component of the i-th sampling point; y is the average value of the k-th intrinsic mode function (IMF) component; y is the value of the zero-sequence current signal at the i-th sampling point. The mean value of the zero-sequence current signal;

[0146] The intrinsic mode function (IMF) components are sorted in descending order of their absolute correlation coefficients, and the top three IMF components are selected as valid fault feature components. Meanwhile, the remaining components are defined as noise components and removed.

[0147] Step S5: Calculate the sample entropy of the effective fault features and combine the sample entropy into a feature vector representing the fault state;

[0148] The discretized expression for the intrinsic mode function (IMF) components is:

[0149] (16)

[0150] The expression for calculating the vectors reconstructing the phase space is:

[0151] (17)

[0152] Where m is the embedding dimension, This is the i-th sampling point of the intrinsic mode function (IMF) component.

[0153] The expression for calculating the distance in the intrinsic mode function (IMF) components is as follows:

[0154] (18)

[0155] Where x(l) is the l-th sampling point of the intrinsic mode function (IMF) component; q is the element index within the interval [0, m-1].

[0156] The expression for calculating sample entropy is:

[0157] (19)

[0158] Where ɛ is the similarity tolerance, SampEn(m, ) represents the sample entropy. This is the preset convergence accuracy value;

[0159] B i (m) Let be the matching probability in m dimensions. N is the IMF signal length, N ɛ The number of vectors that satisfy d[u(i),u(j)]<ɛ;

[0160] B m Let be the average matching probability in m dimensions. ;

[0161] The top three IMF components are selected, and their sample entropy values ​​are calculated for each. They are then combined strictly according to the ranking order of their corresponding IMF components' correlation coefficients, and the sample entropy values ​​are arranged sequentially to form a three-dimensional feature vector.

[0162] Step S6: Input the feature vectors as input layer variables into the support vector machine for fault identification;

[0163] This embodiment uses a support vector machine approach to achieve fault identification, primarily through an optimal hyperplane function. The specific calculation expression is shown below:

[0164] (20)

[0165] in, For the classification output results of the support vector machine, It is a Lagrange multiplier, and C is the penalty factor; For kernel functions; The classification threshold is determined for the training samples; sgn is the sign function, used to output the category label of the classification result. The category label output by the sign function is used to indicate whether a single-phase ground fault exists in the stator winding of the hydro-generator. In this embodiment, a category label of "1" indicates that a single-phase ground fault has occurred; a category label of "0" indicates that no single-phase ground fault has occurred. By outputting category labels through a support vector machine, fault identification can be achieved quickly and conveniently.

[0166] See Figure 3 H is a hyperplane serving as the decision boundary for fault identification. H1 and H2 are classification surfaces parallel to H and equidistant from each other. H1 is set as the boundary for the fault class, containing the support vectors of fault samples; H2 is set as the boundary for the healthy class, containing the support vectors of healthy samples. The maximum margin is the distance between H1 and H2. When the margin is maximized, H is the optimal hyperplane. H has the strongest separation ability between fault samples and healthy samples, the strongest generalization ability, and can effectively avoid misjudgments caused by noise interference.

[0167] For further details, please refer to [link / reference]. Figure 1In this embodiment, the SMA slime mold algorithm is used to construct the SVM optimal hyperplane function. The optimal parameters are found directly through population iteration optimization of SMA; or the optimal hyperplane is indirectly derived by optimizing the Lagrange multipliers.

[0168] Example 2

[0169] Based on the same inventive concept as Embodiment 1, this embodiment introduces a device for identifying ground faults in the stator winding of a hydro-generator, comprising:

[0170] The signal acquisition module is configured to acquire the zero-sequence current signal when a single-phase ground fault occurs in a large hydro-generator.

[0171] The optimal module is configured to: use the minimization of envelope entropy as the fitness function, and use the IDBO algorithm to adaptively optimize the number of modes and the penalty factor of variational mode decomposition (VMD) to obtain the optimal parameter combination; wherein, the IDBO algorithm is obtained by integrating Logistic-Tent chaotic mapping, exponential decay convergence factor and random perturbation mechanism.

[0172] The decomposition processing module is configured to: perform signal decomposition processing based on the zero-sequence current signal and the optimized VMD parameters to obtain multiple intrinsic mode function (IMF) components;

[0173] The feature selection module is configured to select the three most correlated coefficients from multiple intrinsic mode function (IMF) components as valid fault features.

[0174] The vector construction module is configured to: calculate the sample entropy of effective fault features and combine the sample entropy into a feature vector representing the fault state;

[0175] The fault identification module is configured to: input the feature vector as an input layer variable into the support vector machine for fault identification, and output a category label indicating whether there is a single-phase grounding fault in the stator winding of the hydro generator.

[0176] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0177] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0178] This invention is described with reference to flowchart illustrations of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each combination of processes in the flowcharts can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the processes. Figure 1 A device for a function specified in one or more processes.

[0179] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 The function specified in one or more processes.

[0180] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 Steps of a specified function in one or more processes.

[0181] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for identifying grounding faults in the stator winding of a hydro-generator, characterized in that, include: Acquire the zero-sequence current signal when a single-phase ground fault occurs in a hydro-generator; Using the minimization of envelope entropy as the fitness function, the IDBO algorithm is used to adaptively optimize the number of modes and the penalty factor of variational mode decomposition (VMD) to obtain the optimal parameter combination; wherein, the IDBO algorithm is obtained by integrating Logistic-Tent chaotic mapping, exponential decay convergence factor and random perturbation mechanism. Based on the zero-sequence current signal and the optimal parameter combination, signal decomposition processing is performed to obtain multiple intrinsic mode function (IMF) components. The three components with the highest correlation coefficients from the multiple intrinsic mode functions (IMF) are selected as valid fault features. Calculate the sample entropy of the effective fault features, and combine the sample entropy into a feature vector characterizing the fault state; The feature vectors are used as input layer variables and imported into a support vector machine for fault identification, outputting a category label indicating whether a single-phase ground fault exists in the stator winding of the hydro-generator.

2. The method for identifying grounding faults in the stator winding of a hydro-generator according to claim 1, characterized in that, Before acquiring the zero-sequence current signal during a single-phase ground fault in a hydro-generator, a ground fault model of the hydro-generator is established, and a grounding point is set in the external circuit of the stator winding of the hydro-generator ground fault model to simulate the ground fault condition; the zero-sequence current signal is acquired under the simulated ground fault condition.

3. The method for identifying grounding faults in the stator winding of a hydro-generator according to claim 1, characterized in that, The Logistic-Tent chaotic mapping is used to generate a uniformly distributed initial population, and the specific calculation expression is as follows: （1）， in, The value is the chaotic sequence value of the current iteration, with a range of [0,1]. r is a control parameter, and mod is the modulo operation used to map the result to the interval [0,1].

4. The method for identifying grounding faults in the stator winding of a hydro-generator according to claim 1, characterized in that, The exponentially decaying convergence factor is used to nonlinearly and dynamically adjust the balance between global exploration and local exploitation capabilities during the iteration process, thereby accelerating the algorithm's convergence speed. The specific calculation expression is as follows: （2）， Where R is the exponential convergence factor, and t is the current iteration number. The maximum number of iterations, with a value ranging from 1 to t. .

5. The method for identifying grounding faults in the stator winding of a hydro-generator according to claim 1, characterized in that, The implementation process of the random perturbation mechanism includes: If the optimal fitness value of the population does not improve in M ​​consecutive iterations, a random perturbation is applied to the individual position X. The specific calculation expression is as follows: （3）， in, The scaling factor has a value range of [1, 10]; {t1, t2, ..., t n } represents the current iteration number, n represents the iteration number, n≤M; X(t) represents the random perturbation applied to the individual position X in the t-th iteration; r(t) n ) is the t-th n The chaotic sequence value for the next iteration is calculated using the following expression: （4）， The normalized mapping formula is used to constrain the position of the perturbed individuals. The specific calculation expression is as follows: （5）， in, This represents the position of the i-th dimension random walk variable in the t-th iteration. and Let be the global minimum and maximum values ​​of the i-th dimension random walk variable, respectively. and Let represent the local minimum and maximum values ​​of the i-th dimension random walk variable in the t-th iteration, respectively.

6. The method for identifying grounding faults in the stator winding of a hydro-generator according to claim 1, characterized in that, The expression for calculating the envelope entropy is as follows: （6）， in, Envelope entropy; P k This is the normalized form of the decomposed signal a(k). K represents the number of intrinsic mode functions (IMFs) obtained from the decomposition; k represents the component of the k-th IMF. H[x(k)] is the Hilbert transform of the original time-domain signal x(k).

7. The method for identifying grounding faults in the stator winding of a hydro-generator according to claim 1, characterized in that, Obtaining the optimal parameter combination includes: The zero-sequence current signal is decomposed into K intrinsic mode functions (IMF) components, and each IMF component is subjected to a Hilbert transform to obtain a one-sided spectrum. The specific calculation expression is as follows: （7）， Where t is the current iteration number, Let be the modal component of the k-th intrinsic mode function (IMF) after t iterations. is the unit impulse function; k is the k-th intrinsic mode function (IMF) component; j is the imaginary unit; modal components The spectrum is transferred to the corresponding fundamental frequency band, and the specific calculation expression is as follows: （8）， in, for The center frequency; The symbol for convolution; For the modal components transferred to the corresponding fundamental frequency band The spectrum is used to construct a constrained variational model, and the specific calculation expression of the constrained variational model is as follows: （9）， in, This indicates taking the partial derivative with respect to t. for Norm, The zero-sequence current signal obtained after t iterations is the time-domain representation. Introducing a secondary penalty term and Lagrange multipliers Construct the corresponding unconstrained augmented Lagrangian function, and calculate its expression as follows: （10）， The alternating direction multiplier method is used to solve for the saddle point corresponding to the augmented Lagrangian function. The optimal solution of the variational model is gradually approximated through alternating iterations. The specific calculation expression is as follows: （11）， （12）， （13）， in, Here, is the bandwidth parameter, and n is the number of iterations. For Lagrange multipliers, This refers to the k-th intrinsic mode function (IMF) component during the (n+1)-th iteration. This is the frequency domain representation of the zero-sequence current signal obtained after t iterations. , , They are , as well as Obtained through Fourier transform, Update the step size and repeat the above steps until the following convergence condition is met to calculate the expression: （14）， in, This is the preset convergence accuracy value.

8. The method for identifying grounding faults in the stator winding of a hydro-generator according to claim 1, characterized in that, The formula for calculating the correlation coefficient is as follows: （15）， Where i = 1, 2, ..., N, i is the index of the discrete sampling point, and N is the signal length; The correlation coefficient; u k (i) represents the k-th intrinsic mode function (IMF) component of the i-th sampling point; y is the average value of the k-th intrinsic mode function (IMF) component; y is the value of the zero-sequence current signal at the i-th sampling point. The mean value of the zero-sequence current signal; The intrinsic mode function (IMF) components are sorted in descending order of their absolute correlation coefficients, and the top three IMF components are selected as effective fault feature components. The sample entropy of the effective fault feature components is combined to construct the following process: The intrinsic mode function (IMF) components are represented as discrete sequences, and the specific expression of the discrete sequences is as follows: （16）， Based on the discrete sequence, the phase space vector used to calculate the sample entropy is reconstructed, and the specific calculation expression is as follows: （17）， Where m is the embedding dimension, and x(i) is the i-th sampling point of the intrinsic mode function (IMF) component; Based on the phase space vectors, calculate the distance between any two vectors u(i) and u(l), using the following formula: （18）， Where x(l) is the l-th sampling point of the intrinsic mode function (IMF) component; q is the element index within the interval [0, m-1]. The expression for calculating the sample entropy is: （19）， Where ɛ is the similarity tolerance, SampEn(m, ) represents the sample entropy. This is the preset convergence accuracy value; B i (m) Let be the matching probability in m dimensions. N is the IMF signal length, N ɛ The number of vectors that satisfy d[u(i),u(j)]<ɛ; B m Let be the average matching probability in m dimensions. .

9. The method for identifying grounding faults in the stator winding of a hydro-generator according to claim 1, characterized in that, The support vector machine identifies faults using an optimal hyperplane classification function, the calculation expression of which is: （20）， in, For the classification output results of the support vector machine, It is a Lagrange multiplier, and C is the penalty factor; For kernel functions; The classification threshold is determined for the training samples; sgn is the sign function used to output the category label of the classification result; the category label is used to indicate whether there is a single-phase grounding fault in the stator winding of the hydro-generator.

10. A device for identifying grounding faults in the stator winding of a hydro-generator, characterized in that, include: The signal acquisition module is configured to acquire the zero-sequence current signal when a single-phase ground fault occurs in the hydro-generator. The optimal module is configured to: use the minimization of envelope entropy as the fitness function, and use the IDBO algorithm to adaptively optimize the number of modes and the penalty factor of variational mode decomposition (VMD) to obtain the optimal parameter combination; wherein, the IDBO algorithm is obtained by integrating Logistic-Tent chaotic mapping, exponential decay convergence factor and random perturbation mechanism. The decomposition processing module is configured to: perform signal decomposition processing based on the zero-sequence current signal and the optimal parameter combination to obtain multiple intrinsic mode function (IMF) components; The feature selection module is configured to select the three most correlated components from the plurality of intrinsic mode functions (IMF) as valid fault features. The vector construction module is configured to: calculate the sample entropy of the effective fault features and combine the sample entropy into a feature vector characterizing the fault state; The fault identification module is configured to: import the feature vector as an input layer variable into a support vector machine for fault identification, and output a category label indicating whether there is a single-phase grounding fault in the stator winding of the hydro-generator.

Images