A transmission line lightning stroke fault feature extraction method based on wild dog optimization algorithm

Abstract

This invention discloses a method for extracting features of lightning strike faults in transmission lines based on the Wild Dog optimization algorithm, which relates to the field of transmission line fault diagnosis technology. This invention utilizes the symplectic geometric mode decomposition method to decompose the fault traveling wave signal of the transmission line and calculate the symplectic geometric entropy of the effective symplectic geometric components. The Wild Dog optimization algorithm is introduced to optimize the initial parameters of the multi-scale permutation entropy to obtain the optimal multi-scale permutation entropy. Finally, the multi-scale permutation entropy and the symplectic geometric entropy are combined to construct a feature vector, which is then fed into a random forest classifier to obtain good classification results.

Description

Technical Field

[0001] This invention relates to the field of power system lightning strike transmission line detection technology, specifically to a method for extracting features of transmission line lightning strike faults based on the Wild Dog optimization algorithm. Background Technology

[0002] Transmission lines are a crucial component of the power system. Located in open fields with complex natural and geographical conditions, they have a high probability of being struck by lightning. When transmission lines are struck by lightning, it can easily cause widespread power outages, resulting in economic losses. Identifying various lightning- and non-lightning-struck faults in transmission lines is the prerequisite and foundation for targeted and rapid on-site troubleshooting. To improve the accuracy of transmission line fault identification, the most important aspect is the extraction of fault characteristics.

[0003] When a transmission line fault occurs, voltage and current traveling waves propagate at near the speed of light to both sides of the line at the fault point. These traveling waves are non-stationary signals containing rich fault information. The proper use of this information can effectively improve the accuracy of fault identification. Commonly used feature extraction methods include wavelet transform (WT), Hilbert-Huang transform (HHT), and variational mode decomposition (VMD). The effectiveness of wavelet transform in extracting features from fault traveling waves is often limited by the appropriate selection of wavelet basis functions and decomposition scale. While HHT performs better than wavelet transform, its key step, Empirical Mode Decomposition (EMD), suffers from severe over-envelope and under-envelope issues, mode aliasing, and severe endpoint effects. VMD offers better feature extraction than the former two methods, effectively avoiding mode aliasing; however, the selection of VMD parameters is often based on experience. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention employs Symptotic Geometry Mode Decomposition (SGMD). SGMD exhibits strong robustness in processing nonlinear signals, largely avoiding noise interference with fault traveling wave signals. It also offers advantages such as preserving the essential characteristics of the time series and suppressing mode aliasing. Most importantly, SGMD does not require the definition of any subjective parameters. By combining SGMD with a Wild Dog optimization algorithm-improved Multiscale Permutation Entropy (MPE) and solving for the symptotic geometric entropy, a low-dimensional feature vector is constructed. This vector is then classified using a random forest classifier, achieving high accuracy in identifying transmission line fault types.

[0005] The present invention specifically adopts the following technical solution:

[0006] A method for extracting features of lightning strike faults in transmission lines based on the Wild Dog optimization algorithm includes the following steps:

[0007] Step 1: Acquire the original three-phase voltage signal U of the transmission line fault to be identified. A U B U C .

[0008] Step 2: To eliminate the complex coupling relationship between the three-phase conductors of the transmission line, the three-phase voltage signals of the transmission line fault collected in Step 1 are decoupled using phase mode conversion technology.

[0009] The formula for decoupling by phase mode transformation is:

[0010]

[0011] In the formula, U0, U1, and U2 are the decoupled 0-mode component, 1-mode component, and 2-mode component, respectively.

[0012] Step 3: Since transmission lines generate phase-to-phase line-mode components (1-mode and 2-mode components are called line-mode components) regardless of the type of fault, the line-mode traveling wave components are selected to extract features of the fault traveling wave. Here, we take the 1-mode component U1 as an example to extract the fault traveling wave features. U1 is decomposed using symplectic geometric mode decomposition to obtain symplectic geometric components SGC1, SGC2, SGC3...SGC n The specific steps of symplectic geometric mode decomposition are as follows:

[0013] Step 3.1 Let the time sequence of the original fault traveling wave signal U1 be x = x1, x2, x3, ..., x r, where r is the data length. According to Takens' embedding theorem, a multidimensional signal can be reconstructed from a one-dimensional signal using a time-series delay topological equivalence method to obtain the trajectory matrix X.

[0014]

[0015] Where d is the embedding dimension, τ is the delay time and τ = 1, and m = r - (d - 1)τ, by choosing an appropriate embedding dimension d and delay time τ, the corresponding reconstruction matrix X can be obtained. First, the power spectral density (PSD) of the initial time series x is calculated; then, the frequency f corresponding to the maximum peak value in the PSD is estimated. max If the normalized frequency f max / F S Less than a given threshold of 10 -3 In this case, d is set to r / 3, where r is the data length. Otherwise, the embedding dimension is set to d = 1.2 × (F S / f max ).

[0016] Step 3.2 Construct the Hamiltonian matrix M based on the trajectory matrix X obtained in Step 3.1. A = X T X.

[0017] Step 3.3 Calculate the eigenvalues ​​λ of matrix A i (i = 1, 2, ..., d) and the eigenvectors Q corresponding to the eigenvalues ​​of matrix A. i (i = 1, 2, ..., d).

[0018] Step 3.4 Calculate the transformation coefficient matrix S i , Through formula Z i =Q i S i (i = 1, 2, ..., d), to obtain the corresponding reconstruction matrix. Then the trajectory matrix X = X1 + X2 + ... + X d .

[0019] Step 3.5 Transform X using diagonal mean. i Convert to a time series Y of length r i =y1, y2, ...y r (i = 1, 2, ..., d), thus obtaining d time series of length r, i.e., d initial components, whose sum is a one-dimensional time series x = x1, x2, ..., xd. r The specific steps of the above process are as follows:

[0020] For matrix Xi element x in ij (1≤i≤m,1≤j≤d), let d * =min(m,d), m * =max(m,d), r = m + (d-1)τ, then the corresponding time series Y i element y in k (k = 1, 2, ..., r) is calculated as follows:

[0021]

[0022] Where, if m < d, let

[0023] Step 4: Calculate the center frequency of the symplectic geometric component and its correlation coefficient with the original fault signal. Select the effective component based on the correlation coefficient and discard the noisy component.

[0024] The correlation coefficient is a statistic used to measure the degree of correlation between variables. The larger the correlation coefficient between two signals, the stronger the correlation, and vice versa. Therefore, the modal components with larger and smaller correlation coefficients to the original signal are respectively called the effective component and the noisy component.

[0025] The formula for calculating the correlation coefficient is as follows:

[0026]

[0027] In the formula, SGC e This represents the e-th component of U1 after symplectic geometric mode decomposition, where e = 1, 2, 3, ..., n.

[0028] The effective component and the noisy component are determined based on the ratio of the correlation coefficients between adjacent components and the original fault traveling wave signal, and the judgment conditions are as follows:

[0029]

[0030] Step 5: Calculate the multiscale permutation entropy (MPE) of the effective symplectic geometric components of the fault traveling wave obtained in Step 4. Here, we still assume that SGC1 and SGC2 are the two effective components with the largest correlation coefficients obtained in Step 4, and calculate the multiscale permutation entropies o1 and o2 of SGC1 and SGC2. The specific steps of the above process are as follows:

[0031] Step 5.1 Combine the two time series SGC1(X={x 11 ,x 12 ,x 13 ,…,x 1r}) and SGC2(X={x21 ,x 22 ,x 23 ,…,x 2r (Assuming SGC1 and SGC2 are the two most effective components with the largest correlation coefficients obtained in Step 4) After coarsening, we get:

[0032]

[0033] Where j = 1, 2, 3, ..., [r / s], [r / s] represents rounding down r / s, and s represents the scaling factor of DOA optimized by the Wild Dog optimization algorithm.

[0034] Step 5.2 Reconstruct the new sequence obtained after the coarsening in Step 5.1:

[0035]

[0036] Where b represents the embedding dimension, λ represents the delay time, and both b and λ are optimized using the Wild Dog optimization algorithm, and l represents the reconstructed component and

[0037] l = 1, 2, ..., r - (b - 1)λ

[0038] Step 5.3 Arrange the formulas for reconstructing the new sequence in ascending order. Each coarsened sequence yields a new sequence s(v) = (l1, l2, ..., l...). b ), where v = 1, 2, ..., V. V ≤ b!, and the number of s(v) is the same as the number of reconstructed sequences b!.

[0039] Step 5.4 Calculate the permutation entropy at different scales:

[0040]

[0041] Among them, P v Let H be the probability of the v-th symbol sequence appearing. From the above formula, we can obtain: H P (b) When taking the maximum value lnb!, For H P (b) After normalization, we get:

[0042]

[0043] Step 6: Calculate the symplectic geometric entropy of the effective symplectic geometric components of the fault traveling wave obtained in Step 4. Here, we still assume that SGC1 and SGC2 are the two effective components with the largest correlation coefficients obtained in Step 4, and calculate the symplectic geometric entropies q1 and q2 of SGC1 and SGC2. The specific steps of the above process are as follows:

[0044] In Step 3, the eigenvalues ​​λ of the real symmetric matrix A constructed from the trajectory matrix X are... i (i = 1, 2, ..., d), λ i If the distribution represents the spectral distribution of matrix A, then the probability of energy distribution in different directions can be defined as P1, P2, P3, ..., P d The calculation formula is:

[0045]

[0046] Where d represents the embedding dimension, 0≤p i ≤1, p i This describes the uncertainty of entropy in different directions. Therefore, the definition of symplectic geometric entropy can be obtained as:

[0047]

[0048] Step 7: Combine the multi-scale permutation entropies o1 and o2 obtained in Step 5 and the symplectic geometric entropies q1 and q2 obtained in Step 6 into a new four-dimensional feature vector eigenvector = [o1, o2, q1, q2].

[0049] Step 8: Input the feature vector eigenvector obtained in Step 7 into the random forest classifier to perform fault classification and identification.

[0050] Furthermore, the specific steps in Step 5 for optimizing the scale factor s, embedding dimension b, and delay time λ using the Wild Dog optimization algorithm are as follows:

[0051] First, the time series SGC1(X={x 11 ,x 12 ,x 13 ,…,x 1r (Assuming SGC1 is one of the effective components with the largest correlation coefficient obtained in Step 4) Substitute this into the unoptimized multi-scale permutation entropy model, select the square function of the kurtosis of the multi-scale permutation entropy as the fitness function, and then continuously update the new [s, b, λ] combination according to the Wild Dog optimization algorithm. Calculate the fitness value after each update, compare the sizes of all fitness values, and take the [s, b, λ] combination corresponding to the smallest fitness value as the optimal parameter combination for the multi-scale permutation entropy algorithm, thus constructing a new multi-scale permutation entropy algorithm optimized by the Wild Dog optimization algorithm. The process of generating the fitness function in the above parameter optimization process is as follows:

[0052] Let the time series X = {x i The permutation entropy of the sequence H at all scales of {i = 1, 2, ..., r} constitutes the sequence. p (X)={H p(1),H p (2),…,H p The formula for calculating the kurtosis of the multi-scale permutation entropy is:

[0053]

[0054] In the above formula, Represents sequence H p The mean of (X); Represents sequence H p The standard deviation of (X) is represented by E(·), which represents the expected value.

[0055] The square function of the entropy kurtosis of a multi-scale arrangement is:

[0056] F(X) = Kurt 2

[0057] Furthermore, the specific steps of the Wild Dog optimization algorithm are as follows:

[0058] (1) Population initialization

[0059] The wild dog population is randomly initialized within the search boundary:

[0060]

[0061] Among them, lb i and ub i rand represents the upper and lower bounds of an individual, respectively. i It is a random number between [0,1], where i represents the i-th stray dog.

[0062] (2) Strategy 1: Mass Attack

[0063] Predators typically employ highly intelligent hunting techniques. Wild dogs usually hunt small prey, such as rabbits, alone, but when hunting larger prey, such as kangaroos, they hunt in packs. Wild dogs can locate their prey and surround it, behaving as follows:

[0064]

[0065] Where t represents the current iteration number. na is the new location of the wild dogs; na is a random integer generated in reverse order of [2, SizePop / 2], where SizePop is the size of the wild dog population. It is a subset of the wild dogs that attack, where k represents the k-th wild dog population; where x i (t) is a randomly generated wild dog population; It is the best wild dog found in the previous iteration; β1 is a uniformly generated random number in [-2,2], which is a scaling factor that can change the size of the wild dog's trajectory.

[0066] (3) Strategy 2: Persecution and Attack

[0067] Wild dogs typically hunt small prey until they catch it alone. The following formula simulates this behavior:

[0068]

[0069] in, This is the new location for stray dogs. This is the best wild dog found in the previous iteration. The value of β1 is the same as the value in the first policy formula. β2 is a random number uniformly generated in the interval [-1, 1]. r1 is a random number generated in the interval from 1 to the size of the maximum search agent (wild dog). It is the r1th wild dog randomly selected, where i ≠ r1.

[0070] (4) Strategy 3: Cleaning

[0071] Scavenging behavior is defined as the act of wild dogs finding and eating carrion while wandering freely in their habitat. The following formula is used to simulate this behavior:

[0072]

[0073] in, This is the new location of the wild dog. The value of β2 is the same as the value in Strategy 2. r1 is a random number generated in the range from 1 to the maximum search agent (wild dog) size. σ is the r1th wild dog randomly selected, where i ≠ r1, and σ is a binary number randomly generated by the algorithm, where σ ∈ [0, 1].

[0074] (5) Strategy 4: Survival rate of wild dogs

[0075] In DOA, the survival rate of wild dogs is given by the following formula:

[0076]

[0077] Among them, fitness max and fitness min These are the worst and best fitness values ​​in the current generation, respectively, while fitness(i) is the current fitness value of the i-th wild dog. The survival vector in the above formula contains the normalized fitness in the interval [0,1]. The following formula is applied to low survival rates using strategy three, for example, when the survival rate is equal to or less than 0.3.

[0078]

[0079] in, These are the wild dogs with lower survival rates that will be updated. r1 and r2 are random numbers generated within a range from 1 to the maximum size of the search agent (wild dog population), where r1 ≠ r2. and It is the r1th and r2th wild dogs that are randomly selected. It is the best wild dog found in the previous iteration; σ is a binary number randomly generated by the algorithm, σ∈[0,1].

[0080] The beneficial effects of this invention are:

[0081] The symplectic geometric mode decomposition method introduced in this invention is a time series decomposition method with strong decomposition capability for traveling wave signals of lightning faults in transmission lines. The obtained single components are more accurate and have better noise robustness compared to other decomposition methods. Symplectic geometric mode decomposition is performed on the traveling wave of the lightning fault to obtain symplectic geometric components. The useful components are obtained using the correlation coefficient method, and their symplectic geometric entropy is calculated. The initial parameters of the multi-scale permutation entropy are tuned using the Wild Dog optimization algorithm. The Wild Dog optimization algorithm has advantages such as strong optimization ability and fast convergence speed. The multi-scale permutation entropy is calculated for the useful symplectic geometric components. The multi-scale permutation entropy and the symplectic geometric entropy are combined to construct a feature vector, which is then fed into a random forest classifier, achieving good classification and recognition results. Attached Figure Description

[0082] Figure 1 The flowchart shows a method for extracting features of lightning-induced faults in transmission lines based on the Wild Dog optimization algorithm.

[0083] Figure 2 This is a simulation circuit diagram for lightning strike faults on power transmission lines.

[0084] Figure 3 The iterative process for obtaining the optimal multi-scale permutation entropy algorithm parameters;

[0085] Figure 4 For different permutation entropies under different scale factors;

[0086] Figure 5 This is a graph showing the actual and predicted classifications for the test set. Detailed Implementation

[0087] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings.

[0088] like Figure 1 As shown, an embodiment of the present invention discloses a method for extracting features of lightning strike faults in transmission lines based on the Wild Dog optimization algorithm, comprising the following steps:

[0089] Step 1: Noise is added to the detected original fault signal of the transmission line. It should be noted that this step of adding noise is only to more closely resemble a real lightning strike fault signal, thereby better reflecting the effect of the invention, and is not an actual step of the invention. The three-phase voltage signal U of the original transmission line fault to be identified is collected. A U B U C .

[0090] Step 2: To eliminate the complex coupling relationship between the three-phase conductors of the transmission line, the three-phase voltage signals of the transmission line fault collected in Step 1 are decoupled using phase mode conversion technology.

[0091] The formula for decoupling by phase mode transformation is:

[0092]

[0093] In the formula, U0, U1, and U2 are the decoupled 0-mode component, 1-mode component, and 2-mode component, respectively.

[0094] Step 3: Since transmission lines generate phase-to-phase line-mode components (1-mode and 2-mode components are called line-mode components) regardless of the type of fault, the line-mode traveling wave components are selected for feature extraction of the fault traveling wave. Here, the 1-mode component U1 is selected for fault traveling wave feature extraction. U1 is decomposed using symplectic geometric mode decomposition to obtain symplectic geometric components SGC1, SGC2, SGC3...SGC n The specific steps of symplectic geometric mode decomposition are as follows:

[0095] Step 3.1 Let the time sequence of the original fault traveling wave signal U1 be x = x1, x2, x3, ..., x r , where r is the data length. According to Takens' embedding theorem, a multidimensional signal can be reconstructed from a one-dimensional signal using a time-series delay topological equivalence method to obtain the trajectory matrix X.

[0096]

[0097] Where d is the embedding dimension, τ is the delay time and τ = 1, and m = r - (d - 1)τ, by choosing an appropriate embedding dimension d and delay time τ, the corresponding reconstruction matrix X can be obtained. First, the power spectral density (PSD) of the initial time series x is calculated; then, the frequency f corresponding to the maximum peak value in the PSD is estimated. max If the normalized frequency f max / F S Less than a given threshold of 10 -3In this case, d is set to r / 3, where r is the data length. Otherwise, the embedding dimension is set to d = 1.2 × (F S / f max ).

[0098] Step 3.2 Construct the Hamiltonian matrix M based on the trajectory matrix X obtained in Step 3.1. A = X T X.

[0099] Step 3.3 Calculate the eigenvalues ​​λ of matrix A i (i = 1, 2, ..., d) and the eigenvectors Q corresponding to the eigenvalues ​​of matrix A. i (i = 1, 2, ..., d);

[0100] Step 3.4 Calculate the transformation coefficient matrix S i , Through formula Z i =Q i S i (i = 1, 2, ..., d), to obtain the corresponding reconstruction matrix. Then the trajectory matrix X = X1 + X2 + ... + X d ;

[0101] Step 3.5 Transform X using diagonal mean. i Convert to a time series Y of length r i =u1, y2, ... y r (i = 1, 2, ..., d), thus obtaining d time series of length r, i.e., d initial components, whose sum is a one-dimensional time series x = x1, x2, ..., xd. r The specific steps of the above process are as follows:

[0102] For matrix X i element x in ij (1≤i≤m,1≤j≤d), let d * =min(m,d), m * =max(m,d), r = m + (d-1)τ, then the corresponding time series Y i element y in k (k = 1, 2, ..., r) is calculated as follows:

[0103]

[0104] Where, if m < d, let

[0105] Step 4: Calculate the center frequency of the symplectic geometric component and its correlation coefficient with the original fault signal. Select the effective component based on the correlation coefficient and discard the noisy component.

[0106] The correlation coefficient is a statistic used to measure the degree of correlation between variables. The larger the correlation coefficient between two signals, the stronger the correlation, and vice versa. Therefore, the modal components with larger and smaller correlation coefficients to the original signal are respectively called the effective component and the noisy component.

[0107] The formula for calculating the correlation coefficient is as follows:

[0108]

[0109] In the formula, SGC e This represents the e-th component of U1 after symplectic geometric mode decomposition, where e = 1, 2, 3, ..., n.

[0110] The effective component and the noisy component are determined based on the ratio of the correlation coefficients between adjacent components and the original fault traveling wave signal, and the judgment conditions are as follows:

[0111]

[0112] Step 5: Calculate the multiscale permutation entropy (MPE) of the effective symplectic geometric components of the fault traveling wave obtained in Step 4. Here, we still assume that SGC1 and SGC2 are the two effective components with the largest correlation coefficients obtained in Step 4, and calculate the multiscale permutation entropies o1 and o2 of SGC1 and SGC2. The specific steps of the above process are as follows:

[0113] Step 5.1 Combine the two time series SGC1(X={x 11 ,x 12 ,x 13 ,…,x 1r}) and SGC2(X={x 21 ,x 22 ,x 23 ,…,x 2r (Assuming SGC1 and SGC2 are the two most effective components with the largest correlation coefficients obtained in Step 4) After coarsening, we get:

[0114]

[0115] Where j = 1, 2, 3, ..., [r / s], [r / s] represents rounding down r / s, and s represents the scaling factor of DOA optimized by the Wild Dog optimization algorithm.

[0116] Step 5.2 Reconstruct the new sequence obtained after the coarsening in Step 5.1:

[0117]

[0118] Where b represents the embedding dimension, λ represents the delay time, and both b and λ are optimized using the Wild Dog optimization algorithm, and l represents the reconstructed component and

[0119] l = 1, 2, ..., r - (b - 1)λ

[0120] Step 5.3 Arrange the formulas for reconstructing the new sequence in ascending order. Each coarsened sequence yields a new sequence s(v) = (l1, l2, ..., l...). b ), where v = 1, 2, ..., V. V ≤ b!, and the number of s(v) is the same as the number of reconstructed sequences b!.

[0121] Step 5.4 Calculate the permutation entropy at different scales:

[0122]

[0123] Among them, P v Let H be the probability of the v-th symbol sequence appearing. From the above formula, we can obtain: H P (b) When taking the maximum value lnb!, For H P (b) After normalization, we get:

[0124]

[0125] Step 6: Calculate the symplectic geometric entropy of the effective symplectic geometric components of the fault traveling wave obtained in Step 4. Here, we still assume that SGC1 and SGC2 are the two effective components with the largest correlation coefficients obtained in Step 4, and calculate the symplectic geometric entropies q1 and q2 of SGC1 and SGC2. The specific steps of the above process are as follows:

[0126] In Step 3, the eigenvalues ​​λ of the real symmetric matrix A constructed from the trajectory matrix X are... i (i = 1, 2, ..., d), λ i If the distribution represents the spectral distribution of matrix A, then the probability of energy distribution in different directions can be defined as P1, P2, P3, ..., P d The calculation formula is:

[0127]

[0128] Where d represents the embedding dimension, 0≤p i ≤1, p iThis describes the uncertainty of entropy in different directions. Therefore, the definition of symplectic geometric entropy can be obtained as:

[0129]

[0130] Step 7: Combine the multi-scale permutation entropies o1 and o2 obtained in Step 5 and the symplectic geometric entropies q1 and q2 obtained in Step 6 into a new four-dimensional feature vector eigenvector = [o1, o2, q1, q2].

[0131] Step 8: Input the feature vector eigenvector obtained in Step 7 into the random forest classifier to perform fault classification and identification.

[0132] Furthermore, the specific steps in Step 5 for optimizing the scale factor s, embedding dimension b, and delay time λ using the Wild Dog optimization algorithm are as follows:

[0133] First, the time series SGC1(X={x 11 ,x 12 ,x 13 ,…,x 1r (Assuming SGC1 is one of the effective components with the largest correlation coefficient obtained in Step 4) Substitute this into the unoptimized multi-scale permutation entropy model, select the square function of the kurtosis of the multi-scale permutation entropy as the fitness function, and then continuously update the new [s, b, λ] combination according to the Wild Dog optimization algorithm. Calculate the fitness value after each update, compare the sizes of all fitness values, and take the [s, b, λ] combination corresponding to the smallest fitness value as the optimal parameter combination for the multi-scale permutation entropy algorithm, thus constructing a new multi-scale permutation entropy algorithm optimized by the Wild Dog optimization algorithm. The process of generating the fitness function in the above parameter optimization process is as follows:

[0134] Let the time series X = {x i The permutation entropy sequence H at all scales of {i = 1, 2, ..., r} is composed of... p (X)={H p (1),H p (2),…,H p The formula for calculating the kurtosis of the multi-scale permutation entropy is:

[0135]

[0136] In the above formula, Represents sequence H p The mean of (X); Represents sequence H p The standard deviation of (X) is represented by E(·), which represents the expected value.

[0137] The square function of the entropy kurtosis of a multi-scale arrangement is:

[0138] F(X) = Kurt 2

[0139] Furthermore, the specific steps of the Wild Dog optimization algorithm are as follows:

[0140] (1) Population initialization

[0141] The wild dog population is randomly initialized within the search boundary:

[0142]

[0143] Among them, lb i and ub i rand represents the upper and lower bounds of an individual, respectively. i It is a random number between [0,1], where i represents the i-th stray dog.

[0144] (2) Strategy 1: Mass Attack

[0145] Predators typically employ highly intelligent hunting techniques. Wild dogs usually hunt small prey, such as rabbits, alone, but when hunting larger prey, such as kangaroos, they hunt in packs. Wild dogs can locate their prey and surround it, behaving as follows:

[0146]

[0147] Where t represents the current iteration number. na is the new location of the wild dogs; na is a random integer generated in reverse order of [2, SizePop / 2], where SizePop is the size of the wild dog population. It is a subset of the wild dogs that attack, where k represents the k-th wild dog population, where x i (t) is a randomly generated wild dog population; It is the best wild dog found in the previous iteration; β1 is a uniformly generated random number in [-2,2], which is a scaling factor that can change the size of the wild dog's trajectory.

[0148] (3) Strategy 2: Persecution and Attack

[0149] Wild dogs typically hunt small prey until they catch it alone. The following formula simulates this behavior:

[0150]

[0151] in, This is the new location for stray dogs. This is the best wild dog found in the previous iteration. The value of β1 is the same as the value in the first policy formula. β2 is a random number uniformly generated in the interval [-1, 1]. r1 is a random number generated in the interval from 1 to the size of the maximum search agent (wild dog). It is the r1th wild dog randomly selected, where i ≠ r1.

[0152] (4) Strategy 3: Cleaning

[0153] Scavenging behavior is defined as the act of wild dogs finding and eating carrion while wandering freely in their habitat. The following formula is used to simulate this behavior:

[0154]

[0155] in, This is the new location of the wild dog. The value of β2 is the same as the value in Strategy 2. r1 is a random number generated in the range from 1 to the maximum search agent (wild dog) size. σ is the r1th wild dog randomly selected, where i ≠ r1, and σ is a binary number randomly generated by the algorithm, where σ ∈ [0, 1].

[0156] (5) Strategy 4: Survival rate of wild dogs

[0157] In DOA, the survival rate of wild dogs is given by the following formula:

[0158]

[0159] Among them, fitness max and fitness min These are the worst and best fitness values ​​in the current generation, respectively, while fitness(i) is the current fitness value of the i-th wild dog. The survival vector in the above formula contains the normalized fitness in the interval [0,1]. The following formula is applied to low survival rates using strategy three, for example, when the survival rate is equal to or less than 0.3.

[0160]

[0161] in, These are the wild dogs with lower survival rates that will be updated. r1 and r2 are random numbers generated within a range from 1 to the maximum size of the search agent (wild dog population), where r1 ≠ r2. and It is the r1th and r2th wild dogs that are randomly selected. It is the best wild dog found in the previous iteration; σ is a binary number randomly generated by the algorithm, σ∈[0,1].

[0162] This example uses PSCAD simulation software to simulate a 220kV transmission line. The simulation circuit diagram is as follows. Figure 2 As shown; Figure 3 It is based on Figure 2 The iterative process of obtaining the optimal parameter combination of the multi-scale permutation entropy algorithm by collecting fault traveling wave data; Figure 4 These are the different permutation entropies obtained under different scale factors; Figure 5 The classification and identification results are obtained by inputting the transmission line fault traveling wave dataset obtained through the above process into a random forest classifier.

[0163] Finally, it should be noted that the above description is only for specific embodiments of the present invention. However, the present invention is not limited to the specific embodiments described above. Equivalent modifications and substitutions made to the present invention by those skilled in the art are also within the scope of the present invention. Therefore, all equivalent changes and modifications made without departing from the spirit and scope of the present invention are covered within the scope of the present invention.

Claims

1. A method for extracting features of lightning strike faults in transmission lines based on the Wild Dog optimization algorithm, characterized in that: Includes the following steps: Step 1: collect the original power transmission line fault three-phase voltage signal U A , U B , U C ; Step 2: Decouple the three-phase voltage signals of the transmission line fault collected in Step 1 using phase mode conversion technology; The formula for decoupling by phase mode transformation is: In the formula, U0, U1, and U2 are the decoupled 0-mode component, 1-mode component, and 2-mode component, respectively; Step 3: The 1-mode component and the 2-mode component are called linear mode components, and the linear mode wave component is selected to extract the fault traveling wave feature. Taking the 1-mode component U1 as an example, the fault traveling wave feature is extracted, and the U1 is decomposed by using the symplectic geometric modal decomposition to obtain symplectic geometric components SGC1, SGC2, SGC3,..., SGCn n ; n represents the number of symplectic geometric components obtained by symplectic geometric decomposition; Step 4: Calculate the center frequency of the symplectic geometric component and its correlation coefficient with the original fault signal. Select the effective component based on the correlation coefficient and discard the noisy component. Step 5: Calculate the multi-scale arrangement entropy of the effective symplectic geometric components of the fault traveling wave obtained in Step 4. Assuming that SGC1 and SGC2 are the two effective components with the largest correlation coefficients obtained in Step 4, calculate the multi-scale arrangement entropy o1 and o2 of SGC1 and SGC2. Step 6: Calculate the symplectic geometric entropy of the effective symplectic geometric components of the fault traveling wave obtained in Step 4. Here, we still assume that SGC1 and SGC2 are the two effective components with the largest correlation coefficients obtained in Step 4. Calculate the symplectic geometric entropy q1 and q2 of SGC1 and SGC2. Step 7: Combine the multi-scale permutation entropies o1 and o2 obtained in Step 5 and the symplectic geometric entropies q1 and q2 obtained in Step 6 into a new four-dimensional feature vector eigenvector = [o1, o2, q1, q2]; Step 8: Input the feature vector eigenvector obtained in Step 7 into the random forest classifier to perform fault classification and identification.

2. The method for extracting features of lightning strike faults in transmission lines based on the Wild Dog optimization algorithm according to claim 1, characterized in that, The steps of the symplectic geometric mode decomposition described in Step 3 are as follows: Step3.1: Set the original fault traveling wave signal U1 time series as x=x1, x2, x3, …, x r where r is the data length; according to the Takens embedding theorem, a multi-dimensional signal can be reconstructed by using time series delay topology equivalence method on a one-dimensional signal to obtain the trajectory matrix X; Where d is the embedding dimension, τ is the delay time and τ = 1, and m = r - (d - 1)τ; Step 3.2: Construct the Hamiltonian matrix M based on the trajectory matrix X obtained in Step 3.

1. A = X T X; Step 3.3: Calculate the eigenvalues λ of the matrix A i and the eigenvectors Q corresponding to the eigenvalues of the matrix A i i = 1, 2,..., d; Step 3.4: Calculate the transformation coefficient matrix S i , Through formula Z i =Q i S i For i = 1, 2, ..., d, the corresponding reconstruction matrix is ​​obtained. Then the trajectory matrix X = X1 + X2 + ... + X d ; Step 3.5: Transform X using diagonal mean. i Convert to a time series Y of length r i =y1, y2, ...y r Let i = 1, 2, ..., d. This yields d time series of length r, i.e., d initial components, whose sum is a one-dimensional time series x = x1, x2, ..., xd. r The specific steps of the above process are as follows: For matrix X i element x in ij ; 1≤i≤m, 1≤j≤d, let d * =min(m,d), m * =max(m,d), r = m + (d-1)τ, then the corresponding time series Y i element y in k The calculation is as follows: Where, if m < d, let 3. The method for extracting features of lightning strike faults in transmission lines based on the Wild Dog optimization algorithm according to claim 1, characterized in that, Step 4 includes the following steps: The modal components with larger and smaller correlation coefficients with the original signal are respectively referred to as effective components and noisy components; The formula for calculating the correlation coefficient is as follows: In the formula, SGC e This represents the e-th component of U1 after symplectic geometric mode decomposition, where e = 1, 2, 3, ..., n; The effective component and the noisy component are determined based on the ratio of the correlation coefficients between adjacent components and the original fault traveling wave signal, and the judgment conditions are as follows:

4. The method for extracting features of lightning strike faults in transmission lines based on the Wild Dog optimization algorithm according to claim 1, characterized in that, Step 5 includes the following steps: Step 5.1 Combine the two time series SGC1(X={x 11 ,x 12 ,x 13 ,…,x 1r }) and SGC2(X={x 21 ,x 22 ,x 23 ,…,x 2r After coarsening, the following is obtained: Where j = 1, 2, 3, ..., [r / s], [r / s] represents the floor function of r / s, and s represents the scale factor optimized by the Wild Dog Optimization Algorithm (DOA). Step 5.2 Reconstruct the new sequence obtained after the coarsening in Step 5.1: Where b represents the embedding dimension, λ represents the delay time, both parameters b and λ are optimized by the Wild Dog optimization algorithm, and l represents the reconstructed component; l = 1, 2, ..., r - (b - 1)λ Step 5.3 Arrange the formulas for reconstructing the new sequence in ascending order. Each coarsened sequence yields a new sequence s(v) = (l1, l2, ..., l...). b ), where v = 1, 2, ..., V; V ≤ b!, and the number of s(v) is the same as the number of reconstructed sequences b!; Step 5.4 Calculate the permutation entropy at different scales: Among them, P v Let H be the probability of the v-th symbol sequence occurring; from the above formula, we get: H P (b) When the maximum value ln(b!) is taken, For H P (b) After normalization, we get:

5. The method for extracting features of lightning strike faults in transmission lines based on the Wild Dog optimization algorithm according to claim 4, characterized in that, The specific steps in Step 5 for optimizing the scale factor s, embedding dimension b, and delay time λ using the Wild Dog optimization algorithm are as follows: First, the time series SGC1(X={x 11 ,x 12 ,x 13 ,…,x 1r Substituting the unoptimized multi-scale permutation entropy model, the square function of the kurtosis of the multi-scale permutation entropy is selected as the fitness function. Then, according to the Wild Dog optimization algorithm, new combinations of [s, b, λ] are continuously updated. The fitness value after each update is calculated, and the size of all fitness values ​​is compared. The combination of [s, b, λ] corresponding to the smallest fitness value is taken as the optimal parameter combination for the multi-scale permutation entropy algorithm, thus constructing a new multi-scale permutation entropy algorithm optimized by the Wild Dog optimization algorithm. The process of generating the fitness function in the above parameter optimization process is as follows: Let the time series X = {x i The permutation entropy of the sequence H at all scales of {i = 1, 2, ..., r} constitutes the sequence H. p (X)={H p (1),H p (2),…,H p The formula for calculating the kurtosis of the multi-scale permutation entropy is: In the above formula, Represents sequence H p The mean of (X); Represents sequence H p The standard deviation of (X); E(·) represents the expected value; The square function of the entropy kurtosis of a multi-scale arrangement is: F(X)=Kurt 2 。 6. The method for extracting features of lightning strike faults in transmission lines based on the Wild Dog optimization algorithm according to claim 5, characterized in that, The specific steps of the Wild Dog optimization algorithm are as follows: (1) Population initialization The wild dog population is randomly initialized within the search boundary: Among them, lb i and ub i rand represents the upper and lower bounds of an individual, respectively. i It is a random number between [0,1], where i represents the i-th stray dog; (2) Strategy 1: Mass Attack Predators employ highly intelligent hunting techniques. Wild dogs typically hunt small prey alone, but when hunting large prey, they can locate the prey and surround it, exhibiting behavior as shown in the following formula: Where t represents the current iteration number. na is the new location of the wild dogs; na is a random integer generated in reverse order of [2, SizePop / 2], where SizePop is the size of the wild dog population. It is a subset of the wild dogs that are attacked, where k represents the k-th wild dog population, where x i (t) is a randomly generated wild dog population; It is the best wild dog found in the previous iteration; β1 is a random number uniformly generated in [-2,2], which is a scaling factor that can change the size of the wild dog's trajectory; (3) Strategy 2: Persecution and Attack Wild dogs typically hunt small prey until they catch it alone; the following formula simulates this behavior: in, This is the new location for stray dogs. This is the best wild dog found in the previous iteration. The value of β1 is the same as the value in the first policy formula. β2 is a random number uniformly generated in the interval [-1, 1]. r1 is a random number generated in the interval from 1 to the maximum search agent size. It is the r1th wild dog randomly selected, where i ≠ r1; (4) Strategy 3: Cleaning Scavenging behavior is defined as the act of wild dogs finding and eating carrion while wandering freely in their habitat; the following formula is used to simulate this behavior: in, This is the new location for the wild dog, and r1 is a random number generated within the range of 1 to the maximum search agent size. σ is the r1th wild dog randomly selected, where i ≠ r1, and σ is a binary number randomly generated by the algorithm, where σ ∈ [0, 1]. (5) Strategy 4: Survival rate of wild dogs In DOA, the survival rate of wild dogs is given by the following formula: Among them, fitness max and fitness min These are the worst and best fitness values ​​in the current generation, respectively, while fitness(i) is the current fitness value of the i-th wild dog; the survival vector in the above formula includes the normalized fitness in the interval [0,1]; the following formula is applied to low survival rates using strategy three: in, The wild dogs with lower survival rates will be updated. r1 and r2 are random numbers generated within the range from 1 to the maximum size of the search agent, where r1 ≠ r2. and It is the r1th or r2th wild dog randomly selected.

7. The method for extracting features of lightning strike faults in transmission lines based on the Wild Dog optimization algorithm according to claim 2, characterized in that, Step 6 includes the following steps: In Step 3, the eigenvalues ​​λ of the real symmetric matrix A constructed from the trajectory matrix X are... i i = 1, 2, ..., d, λ i The distribution represents the spectral distribution of matrix A, then the probability of energy distribution in different directions is defined as P1, P2, P3, ..., P d The calculation formula is: Where d represents the embedding dimension, 0≤p i ≤1, p i This describes the uncertainty of entropy in different directions; therefore, the definition of symplectic geometric entropy is obtained as follows:

Images