Method for classifying, identifying and detecting free metals in GIS (Geographic Information System) by optimizing support vector machine through sparrow algorithm

Abstract

The invention aims to provide a method for classifying, identifying and detecting free metals in a GIS (Geographic Information System) by optimizing a support vector machine through a sparrow algorithm, which comprises the following steps of: firstly, acquiring ultrasonic and vibration signals sent by metal foreign matters in a GIS tank body by using acquisition equipment, and preprocessing data; feature extraction is carried out on the preprocessed data, and frequency domain and time domain feature values are extracted; the method comprises the following steps: extracting characteristic values of metal foreign matters in a GIS tank body, inputting the extracted characteristic values into a support vector machine (SVM) model, setting an initial hyper-parameter kernel scale, optimizing hyper-parameters gamma and C of the SVM by using a sparrow algorithm, finally classifying the metal foreign matters in the GIS tank body through the optimized SVM model, and further optimizing the model for different types of foreign matters to obtain type combination data for identifying the metal foreign matters in the GIS tank body. The method is used for identifying the geometric parameters and the material parameters of the metal foreign matters in the GIS equipment, so that guidance is provided for insulation of the GIS equipment, and the economic burden of planned maintenance is reduced.

Description

Technical Field

[0001] The invention belongs to the technical field of GIS equipment status analysis, and in particular relates to a method for classifying, identifying and detecting free metals inside a GIS by using a sparrow algorithm to optimize a support vector machine. Background Art

[0002] Gas Insulated Switchgear (GIS) encapsulates high-voltage devices such as busbars, circuit breakers, disconnectors, and lightning arresters in a grounded metal casing. It is a switchgear currently widely used in my country's high-voltage substations. Because the sealed components in GIS are not affected by the external environment, it has the advantages of high reliability, long maintenance cycle, simple maintenance, etc., and has a compact structure and small footprint. With the laying of high-voltage transmission lines in my country, the number of GIS put into operation has increased year by year.

[0003] Since it operates at a high voltage level, once a fault occurs, the impact range is also larger, often causing the entire substation to lose power, or even causing the collapse of the power system. Because of the high degree of integration of GIS equipment and the sealed structure of multiple components, its internal fault location and maintenance work is difficult and complicated; internal fault maintenance often requires the disassembly of non-faulty units, resulting in long repair time and a large power outage range. GIS equipment will inevitably have various abnormalities and fault phenomena. The main cause of insulation failure is particles and foreign matter. Under the action of applied voltage, they move inside the tank, causing the electric field strength inside the GIS to be distorted, which may cause partial discharge. Long-term partial discharge will further deteriorate the insulation performance and eventually lead to insulation breakdown. Summary of the invention

[0004] The purpose of the present invention is to provide a free metal classification and identification detection method for GIS by optimizing support vector machine with sparrow algorithm, which is used to identify the geometric parameters and material parameters of metal foreign bodies inside GIS equipment, thereby providing guidance for the insulation of GIS equipment and reducing the economic burden of planned maintenance.

[0005] The technical solution adopted by the present invention is to optimize the support vector machine with the sparrow algorithm to classify and identify the free metals in the GIS, which is specifically implemented in the following steps:

[0006] Step 1: Use acquisition equipment to obtain ultrasonic and vibration signals emitted by metal foreign bodies in the GIS tank, and pre-process the data;

[0007] Step 2: Perform feature extraction on the preprocessed data to extract frequency domain and time domain feature values;

[0008] Step 3: Input the extracted eigenvalues ​​into the support vector machine (SVM) model and set the initial hyperparameter kernel scale: γ and box constraint C, which are used to identify and classify the geometric parameters and material parameters of foreign bodies;

[0009] Step 4, using the Sparrow algorithm to optimize the hyperparameters γ and C of SVM, and obtaining the Sparrow algorithm optimized support vector machine for free metal classification, identification and detection method inside GIS;

[0010] Step 5: Use the optimized SVM model to classify metal foreign bodies in the GIS tank, and then further optimize the model for different types of foreign bodies to obtain type combination data for metal foreign body identification in the GIS.

[0011] The present invention is also characterized in that:

[0012] Step 1 is implemented according to the following steps:

[0013] Step 1.1, place a metal foreign body with known parameters inside the GIS tank, fill the GIS equipment with 0.35-0.65MPa SF6, apply 220kV industrial frequency AC voltage, and collect and record ultrasonic and vibration signals inside the GIS equipment cavity;

[0014] Step 1.2: Given a one-dimensional finite vector, the discrete Fourier transform is expressed as:

[0015]

[0016] in, The computational complexity is O(N 2 ), assuming N = 2 M , M is a positive integer, x(n) is divided into two groups according to duality, that is, let n = 2r and n = 2r + 1, then It is expressed as:

[0017]

[0018] According to the decomposition method of the above formula, x(k) is finally decomposed into subsequences of 2 points;

[0019] Step 1.3, ultrasonic signal preprocessing:

[0020] Remove outliers: Identify and remove abnormal data caused by equipment failure or interference;

[0021] Distance-based outlier detection: Use the distance between a data point and its nearest neighbor to identify outliers that are far away from other data points. Outlier processing uses the standard deviation method, and the formula is as follows:

[0022] |x-μ|>kσ

[0023] x is the data point, μ is the mean of the data, σ is the standard deviation of the data, k is the threshold, and the value of k is 2 to 4;

[0024] Step 1.4: Signal smoothing: Apply filtering or smoothing techniques to reduce noise interference:

[0025] Use an exponentially weighted moving average (EMA) to smooth the data and give more weight to recent data points. The formula is as follows:

[0026] EMA t =α·x t +(1-α) EMA t-1

[0027] Among them, EMA t is the exponentially weighted moving average at time t, x t is the original data point at time t, α is the smoothing coefficient, which ranges from 0 to 1;

[0028] Step 1.5: Based on the characteristics of ultrasonic signals, wavelet transform is selected for denoising:

[0029] First, soft threshold processing is performed on the wavelet coefficients, and the coefficients less than the threshold are processed using the soft threshold function to reduce noise:

[0030] SoftThreshold(X,T)=sign(X)*max(|X|-T,0)

[0031] Where X is the wavelet coefficient, T is the threshold, and sign(X) is the sign function, indicating the positive or negative sign of the coefficient;

[0032] Then, the signal is reconstructed, and the processed wavelet coefficients are used for inverse wavelet transform, and the processed coefficients are reconstructed into an ultrasonic signal with noise removed;

[0033] Step 1.6: Vibration signal denoising process, outlier processing, Zscore calculation:

[0034] Calculate the Zscore value of the vibration signal data using the mean and standard deviation of the vibration data. The formula is as follows:

[0035]

[0036] Step 1.7, determine the outliers:

[0037] Set a threshold, and when the absolute value of Zscore is greater than 3, identify the data points exceeding the threshold as outliers; replace the outliers with interpolation for the data points identified as outliers;

[0038] Step 1.8: Filtering operation to remove noise components within a specific frequency range:

[0039] The IIR filter form using the difference equation form is expressed as:

[0040] y[n]=b0x[n]+b1x[n-1]+...+b m x[nm]-a1y[n-1]-...-a n y[nn]

[0041] Where x[n] is the value of the input signal at time n, y[n] is the value of the output signal at time n, b0, b1, …b m are the coefficients of the input signal, a1…a n is the coefficient of the output signal, and the digital filter is applied to the vibration signal data and processed using the filter;

[0042] Step 1.9: Integrate the above processed data, and record the sample data set as X = [x1, x2, x3, …, x i ],i∈[1,N],x i Represents the set of preprocessed signals.

[0043] Step 2 is implemented as follows:

[0044] Step 2.1, feature selection, extract the eigenvalue of the signal after Fourier transformation after denoising, and let the eigenvalue extracted from the ultrasonic signal, the pulse factor: Margin Factor: Crest Factor: X peakF =X rms / X peak , and the kurtosis of the eigenvalues ​​extracted by the vibration sensor: Skewness: As [a i1 ,a i2 ,a i3 ,a i4 ,a i5 ], recorded as a set of sample data a i =[a i1 ,a i2 ,a i3 ,a i4 ,a i5 ] T , i∈[1,N], sample data a i Corresponding to the actual result Y i =[y i1 ,y i2 ,y i3 ,…,y in ] T , where Yi Represents the signal y generated by each free metal foreign body in A collection of;

[0045] Step 2.2, the sample data set is recorded as X = [x1, x2, x3, ..., xi], i∈[1, N], the metal foreign bodies inside the GIS include n kinds, the material parameters at least include copper metal foreign bodies, iron metal foreign bodies, and aluminum metal foreign bodies; the geometric parameters at least include spherical metal foreign bodies, linear metal foreign bodies, sheet metal foreign bodies, spindle-shaped metal foreign bodies, and irregular metal foreign bodies;

[0046] Normalization process, for each feature, calculate its mean μ and standard deviation σ:

[0047] For each eigenvalue X, use the normalization method Zscore to get the normalized eigenvalue X normalized , the corresponding formula is:

[0048]

[0049] Step 2.3, update the data set and normalize the eigenvalue a' i =[a' i1 ,a' i2 ,a' i3 ,a' i4 ,a' i5 ] T , i∈[1,N], replace the original eigenvalue a i =[a i1 ,a i2 ,a i3 ,a i4 ,a i5 ] T , i∈[1,N], update the data set for subsequent machine learning model training and prediction;

[0050] Step 2.4: Construct feature vectors, label the data set, and add corresponding labels and category identifiers to each feature vector.

[0051] Step 3 is implemented as follows:

[0052] Step 3.1: Convert to convex function

[0053]

[0054] sty i (w T x i +b)≥1,i=1,2,...,m

[0055] Then Convert to Solve for convex functions,

[0056] Use Lagrange multiplier method and KKT conditions to solve the optimal value:

[0057] Step 3.2, after solving, substitute into the hyperplane model:

[0058] Available

[0059] The above is the dual form of the SVM dual problem;

[0060] Step 3.3: If the data is linearly inseparable, then introduce a slack variable ξ≥0, so that the function interval plus the slack variable ≥1, then the objective function is:

[0061]

[0062] Dual problem:

[0063]

[0064] C is the penalty parameter. The larger the C value, the greater the penalty for classification. First, use the Lagrange multiplier method to obtain the Lagrange function, and then solve its dual problem.

[0065] Step 4 is implemented according to the following steps:

[0066] Step 4.1, use the sparrow algorithm to optimize the support vector machine parameters, initialize the sampling point set and the proxy model construction, randomly select n0 sampling points within the set parameter search range, and use the average test accuracy of ten-fold cross validation as the objective function f. Different parameter combinations are used as independent variables x to form the proxy model framework, and the initial distribution of the objective function and the sampling point set D are obtained;

[0067] Step 4.2: Sampling point selection: select the next sampling point x by maximizing the acquisition function t , to obtain the function value f(x t ), the sparrow algorithm simulates the behavior of sparrows looking for food and selects the next optimal sampling point according to the current sampling point set D;

[0068] Step 4.3: Update the proxy model and set the new sampling point [x t ,f(x t )] is added to the sampling point set D, and the proxy model is updated so that the proxy model is more consistent with the distribution of the objective function;

[0069] Step 4.4, iterative optimization, set a maximum number of iterations, when the number of iterations reaches the maximum number, stop the algorithm iteration, in each iteration, repeat steps 2 and 3, find the next optimal sampling point, and finally output the optimal sampling point and the corresponding optimal value of the objective function, that is, the optimal parameters C and γ of the SVM model.

[0070] Step 5 is implemented according to the following steps:

[0071] Step 5.1: Use the parameters of metal foreign bodies in GIS to identify and diagnose, and use the test sample x in the test sample set. i Perform the test and get the test result Y text ;

[0072] Step 5.2: According to the test sample x i The actual result Y i Evaluation is performed to determine the test results Y text the accuracy of the information provided and draw conclusions based on it;

[0073] Step 5.3: Calculate the accuracy of the test sample set and compare whether it exceeds the preset accuracy threshold. When the accuracy is higher than this threshold, it can be regarded that the GIS internal insulation fault diagnosis model has high accuracy:

[0074] Calculate the accuracy Z text :

[0075]

[0076] Among them, Y i =[y i1 ,y i2 ,y i3 ,…,y in ] T , Y text =[y text1 ,y text2 ,y text3 ,…,y textn ] T ;

[0077] If the test result is Y text The accuracy Z text If it is not less than z%, then the test result is judged to be Y text To be accurate;

[0078] Step 5.4: Determine whether the model meets the termination conditions: If not, update the parameters. If the error meets the preset accuracy or reaches the maximum number of iterations, output the type of metal foreign body.

[0079] The beneficial effects of the present invention are as follows: the sparrow algorithm optimizes the support vector machine to classify, identify and detect free metals inside GIS; based on the optimization of the support vector machine by the sparrow algorithm, a free metal classification, identification and detection method inside GIS is established; a model is established for the types of metal foreign bodies inside GIS, material parameters: copper metal foreign bodies, iron metal foreign bodies, aluminum metal foreign bodies, and geometric parameters: spherical metal foreign bodies, linear metal foreign bodies, flaky metal foreign bodies, spindle-shaped metal foreign bodies, and irregular metal foreign bodies; the sparrow algorithm is used to optimize the hyperparameters γ and C of SVM to improve the accuracy of the optimized classification model; and a free metal classification, identification and detection method inside GIS by the sparrow algorithm optimized support vector machine is obtained, which can effectively improve the identification of the types of metal foreign bodies inside GIS and improve the accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is a flow chart of the steps of the method for classifying and identifying free metals in GIS by optimizing the support vector machine using the sparrow algorithm in the present invention;

[0081] Figure 2 The overall scheme of the foreign body recognition model in the free metal classification and recognition detection method inside GIS by optimizing the support vector machine using the sparrow algorithm of the present invention;

[0082] Figure 3 The present invention is a flow chart of a foreign body recognition model based on SVM in the free metal classification and recognition detection method inside GIS using the sparrow algorithm to optimize the support vector machine. DETAILED DESCRIPTION

[0083] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0084] The present invention uses the sparrow algorithm to optimize the support vector machine to classify and identify free metals in GIS. The flow chart is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0085] Step 1: Use acquisition equipment to obtain the ultrasonic and vibration signals emitted by metal foreign bodies in the GIS tank, and preprocess the data, including denoising and signal transformation (Fourier and modal analysis) to prepare for feature extraction.

[0086] Combination Figure 2 , step 1 is implemented according to the following steps:

[0087] Step 1.1, place a metal foreign body with known parameters inside the GIS tank, fill the GIS equipment with 0.35-0.65MPa SF6, apply 220kV industrial frequency AC voltage, and collect and record ultrasonic and vibration signals inside the GIS equipment cavity;

[0088] Step 1.2: The signal collected by the sensor is a discrete signal. Since computers can only process sequences of finite length, given a one-dimensional finite vector, the discrete Fourier transform is expressed as:

[0089]

[0090] in, The computational complexity is O(N 2 ), when N is large, the computational complexity of the above Fourier transform is huge. Using Cooley and Tukey fast Fourier transform algorithm, assuming N = 2 M , M is a positive integer, x(n) is divided into two groups according to duality, that is, let n = 2r and n = 2r + 1, then It is expressed as:

[0091]

[0092] According to the decomposition method of the above formula, x(k) is finally decomposed into subsequences of 2 points;

[0093] Step 1.3, ultrasonic signal preprocessing:

[0094] Remove outliers: Identify and remove abnormal data caused by equipment failure or interference;

[0095] Distance-based outlier detection: Use the distance between a data point and its nearest neighbor to identify outliers that are far away from other data points. Outlier processing uses the standard deviation method, and the formula is as follows:

[0096] |x-μ|>kσ

[0097] x is the data point, μ is the mean of the data, σ is the standard deviation of the data, k is the threshold, and the value of k is 2 to 4;

[0098] Step 1.4: Signal smoothing: Apply filtering or smoothing techniques to reduce noise interference:

[0099] Use an exponentially weighted moving average (EMA) to smooth the data and give more weight to recent data points. The formula is as follows:

[0100] EMA t =α·x t +(1-α) EMA t-1

[0101] Among them, EMA t is the exponentially weighted moving average at time t, x t is the original data point at time t, α is the smoothing coefficient, which ranges from 0 to 1;

[0102] Step 1.5: Based on the characteristics of ultrasonic signals, wavelet transform is selected for denoising:

[0103] First, soft threshold processing is performed on the wavelet coefficients, and the coefficients less than the threshold are processed using the soft threshold function to reduce noise:

[0104] SoftThreshold(X,T)=sign(X)*max(|X|-T,0)

[0105] Where X is the wavelet coefficient, T is the threshold, and sign(X) is the sign function, indicating the positive or negative sign of the coefficient;

[0106] Then, the signal is reconstructed, and the processed wavelet coefficients are used for inverse wavelet transform (Inverse Wavelet Transform), and the processed coefficients are reconstructed into an ultrasonic signal with noise removed;

[0107] Step 1.6: Vibration signal denoising process, outlier processing, Zscore calculation:

[0108] Calculate the Zscore value of the vibration signal data using the mean and standard deviation of the vibration data. The formula is as follows:

[0109]

[0110] Step 1.7, determine the outliers:

[0111] Set a threshold, and when the absolute value of Zscore is greater than 3, identify the data points exceeding the threshold as outliers; replace the outliers with interpolation for the data points identified as outliers;

[0112] Step 1.8: Filtering operation to remove noise components within a specific frequency range:

[0113] The noise frequency range to be removed, the digital filter uses a difference equation to describe the behavior of the discrete time system. Using the IIR filter (InfiniteImpulseResponse) in the form of a difference equation, the form is expressed as:

[0114] y[n]=b0x[n]+b1x[n-1]+...+b m x[nm]-a1y[n-1]-...-a n y[nn]

[0115] Where x[n] is the value of the input signal at time n, y[n] is the value of the output signal at time n, b0, b1, …b m are the coefficients of the input signal, a1…a nis the coefficient of the output signal, and the digital filter is applied to the vibration signal data and processed using the filter;

[0116] Step 1.9: Integrate the above processed data, and record the sample data set as X = [x1, x2, x3, …, x i ],i∈[1,N],x i Represents the set of preprocessed signals.

[0117] Combination Figure 3 ,Step 2, perform feature extraction on the preprocessed data, and use the transformed signal to extract frequency domain and time domain eigenvalues ​​for the parameter input layer of SVM.

[0118] Step 2 is implemented as follows:

[0119] Step 2.1, feature selection, extract the eigenvalue of the signal after Fourier transformation after denoising, and let the eigenvalue extracted from the ultrasonic signal, the pulse factor: Margin Factor: Crest Factor: X peakF =X rms / X peak , and the kurtosis of the eigenvalues ​​extracted by the vibration sensor: Skewness: As [a i1 ,a i2 ,a i3 ,a i4 ,a i5 ], recorded as a set of sample data a i =[a i1 ,a i2 ,a i3 ,a i4 ,a i5 ] T , i∈[1,N], sample data a i Corresponding to the actual result Y i =[y i1 ,y i2 ,y i3 ,…,y in ] T , where Y i Represents the signal y generated by each free metal foreign body in A collection of;

[0120] Step 2.2, the sample data set is recorded as X = [x1, x2, x3, ..., xi], i∈[1, N], the metal foreign bodies inside the GIS include n kinds, the material parameters at least include copper metal foreign bodies, iron metal foreign bodies, and aluminum metal foreign bodies; the geometric parameters at least include spherical metal foreign bodies, linear metal foreign bodies, sheet metal foreign bodies, spindle-shaped metal foreign bodies, and irregular metal foreign bodies;

[0121] Normalization process, for each feature, calculate its mean μ and standard deviation σ:

[0122] For each eigenvalue X, use the normalization method Zscore to get the normalized eigenvalue X normalized , the corresponding formula is:

[0123]

[0124] Step 2.3, update the data set and normalize the eigenvalue a' i =[a' i1 ,a' i2 ,a' i3 ,a' i4 ,a' i5 ] T , i∈[1,N], replace the original eigenvalue a i =[a i1 ,a i2 ,a i3 ,a i4 ,a i5 ] T , i∈[1,N], update the data set for subsequent machine learning model training and prediction;

[0125] Step 2.4: Construct feature vectors, label the data set, and add corresponding labels and category identifiers to each feature vector.

[0126] Step 3: Input the extracted eigenvalues ​​into the support vector machine (SVM) model and set the initial hyperparameters kernel scale: γ (KernelScale) and box constraint: C (BoxConstraint) to identify and classify the geometric parameters and material parameters of foreign bodies.

[0127] Step 3 is implemented as follows:

[0128] Step 3.1: Convert to convex function

[0129]

[0130] sty i (w T x i +b)≥1,i=1,2,...,m

[0131] Then Convert to When solving convex functions, 1 / 2 is used to facilitate calculation after differentiation.

[0132] Use Lagrange multiplier method and KKT conditions to solve the optimal value:

[0133] Step 3.2, after solving, substitute into the hyperplane model:

[0134] Available

[0135] The above is the dual form of the SVM dual problem;

[0136] Step 3.3: If the data is linearly inseparable, then introduce a slack variable ξ≥0, so that the function interval plus the slack variable ≥1, then the objective function is:

[0137]

[0138] Dual problem:

[0139]

[0140] C is the penalty parameter. The larger the C value, the greater the penalty for classification. First, use the Lagrange multiplier method to obtain the Lagrange function, and then solve its dual problem.

[0141] Step 4: Use the Sparrow algorithm to optimize the hyperparameters γ and C of SVM to improve the accuracy of the optimized classification model, and obtain the Sparrow algorithm optimized support vector machine for free metal classification and recognition detection method inside GIS; ensure the accurate recognition of the geometric parameters and material parameters of metal foreign bodies.

[0142] Step 4 is implemented according to the following steps:

[0143] Step 4.1, use the sparrow algorithm to optimize the support vector machine parameters, initialize the sampling point set and the proxy model construction, randomly select n0 sampling points within the set parameter search range, and use the average test accuracy of ten-fold cross validation as the objective function f. Different parameter combinations are used as independent variables x to form the proxy model framework, and the initial distribution of the objective function and the sampling point set D are obtained;

[0144] Step 4.2, sampling point selection, select the next sampling point x by maximizing the acquisition function t , to obtain the function value f(x t ), the sparrow algorithm simulates the behavior of sparrows looking for food and selects the next optimal sampling point according to the current sampling point set D;

[0145] Step 4.3: Update the proxy model and set the new sampling point [x t ,f(x t )] is added to the sampling point set D, and the proxy model is updated so that the proxy model is more consistent with the distribution of the objective function. This can be achieved by adjusting the sparrow behavior in the sparrow algorithm to better search the parameter space;

[0146] Step 4.4, iterative optimization, set a maximum number of iterations, when the number of iterations reaches the maximum number, stop the algorithm iteration, in each iteration, repeat steps 2 and 3, find the next optimal sampling point, and finally output the optimal sampling point and the corresponding optimal value of the objective function, that is, the optimal parameters C and γ of the SVM model.

[0147] Step 5: Use the optimized SVM model to classify metal foreign bodies in GIS tanks, and then further optimize the model for different types of foreign bodies to improve the accuracy and reliability of classification, verify the GIS internal metal foreign body recognition model, optimize the support vector machine based on the sparrow algorithm to classify and identify free metals in GIS, and obtain the type combination data for the recognition of metal foreign bodies in GIS.

[0148] Step 5 is implemented according to the following steps:

[0149] Step 5.1: Use the parameters of metal foreign bodies in GIS to identify and diagnose, and use the test sample x in the test sample set. i Perform the test and get the test result Y text ;

[0150] Step 5.2: According to the test sample x i The actual result Y i Evaluation is performed to determine the test results Y text the accuracy of the information provided and draw conclusions based on it;

[0151] Step 5.3: Calculate the accuracy of the test sample set and compare whether it exceeds the preset accuracy threshold. When the accuracy is higher than this threshold, it can be regarded that the GIS internal insulation fault diagnosis model has high accuracy:

[0152] Calculate the accuracy Z text :

[0153]

[0154] Among them, Y i =[y i1 ,y i2 ,y i3 ,…,y in ] T , Y text =[y text1,y text2 ,y text3 ,…,y textn ] T ;

[0155] If the test result is Y text The accuracy Z text If it is not less than z%, then the test result is judged to be Y text To be accurate;

[0156] Step 5.4: Determine whether the model meets the termination conditions: If not, update the parameters. If the error meets the preset accuracy or reaches the maximum number of iterations, output the type of metal foreign body.

[0157] Step 5.5: By statistically analyzing the test sample set, if the accuracy rate exceeds the pre-set standard, the free metal classification and identification detection method model within the GIS is considered to be accurate.

[0158] Example 1

[0159] The present invention uses the sparrow algorithm to optimize the support vector machine to classify and identify free metals in GIS. The flow chart is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0160] Step 1: Use acquisition equipment to obtain the ultrasonic and vibration signals emitted by metal foreign bodies in the GIS tank, and preprocess the data, including denoising and signal transformation (Fourier and modal analysis) to prepare for feature extraction.

[0161] Step 2: Perform feature extraction on the preprocessed data, and use the transformed signal to extract frequency domain and time domain feature values ​​for the parameter input layer of SVM.

[0162] Step 3: Input the extracted eigenvalues ​​into the support vector machine (SVM) model and set the initial hyperparameters kernel scale: γ (KernelScale) and box constraint: C (BoxConstraint) to identify and classify the geometric parameters and material parameters of foreign bodies.

[0163] Step 4: Use the Sparrow algorithm to optimize the hyperparameters γ and C of SVM to improve the accuracy of the optimized classification model, and obtain the Sparrow algorithm optimized support vector machine for free metal classification and recognition detection method inside GIS; ensure the accurate recognition of the geometric parameters and material parameters of metal foreign bodies.

[0164] Step 5: Use the optimized SVM model to classify metal foreign bodies in GIS tanks, and then further optimize the model for different types of foreign bodies to improve the accuracy and reliability of classification, verify the GIS internal metal foreign body recognition model, optimize the support vector machine based on the sparrow algorithm to classify and identify free metals in GIS, and obtain the type combination data for the recognition of metal foreign bodies in GIS.

[0165] Example 2

[0166] The present invention uses the sparrow algorithm to optimize the support vector machine to classify and identify free metals in GIS. The flow chart is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0167] Step 1: Use acquisition equipment to obtain the ultrasonic and vibration signals emitted by metal foreign bodies in the GIS tank, and preprocess the data, including denoising and signal transformation (Fourier and modal analysis) to prepare for feature extraction.

[0168] Step 1 is implemented according to the following steps:

[0169] Step 1.1, place a metal foreign body with known parameters inside the GIS tank, fill the GIS equipment with 0.35-0.65MPa SF6, apply 220kV industrial frequency AC voltage, and collect and record ultrasonic and vibration signals inside the GIS equipment cavity;

[0170] Step 1.2: The signal collected by the sensor is a discrete signal. Since computers can only process sequences of finite length, given a one-dimensional finite vector, the discrete Fourier transform is expressed as:

[0171]

[0172] in, The computational complexity is O(N 2 ), when N is large, the computational complexity of the above Fourier transform is huge. Using Cooley and Tukey fast Fourier transform algorithm, assuming N = 2 M , M is a positive integer, x(n) is divided into two groups according to duality, that is, let n = 2r and n = 2r + 1, then It is expressed as:

[0173]

[0174] According to the decomposition method of the above formula, x(k) is finally decomposed into subsequences of 2 points;

[0175] Step 1.3, ultrasonic signal preprocessing:

[0176] Remove outliers: Identify and remove abnormal data caused by equipment failure or interference;

[0177] Distance-based outlier detection: Use the distance between a data point and its nearest neighbor to identify outliers that are far away from other data points. Outlier processing uses the standard deviation method, and the formula is as follows:

[0178] |x-μ|>kσ

[0179] x is the data point, μ is the mean of the data, σ is the standard deviation of the data, k is the threshold, and the value of k is 2 to 4;

[0180] Step 1.4: Signal smoothing: Apply filtering or smoothing techniques to reduce noise interference:

[0181] Use an exponentially weighted moving average (EMA) to smooth the data and give more weight to recent data points. The formula is as follows:

[0182] EMA t =α·x t +(1-α) EMA t-1

[0183] Among them, EMA t is the exponentially weighted moving average at time t, x t is the original data point at time t, α is the smoothing coefficient, which ranges from 0 to 1;

[0184] Step 1.5: Based on the characteristics of ultrasonic signals, wavelet transform is selected for denoising:

[0185] First, soft threshold processing is performed on the wavelet coefficients, and the coefficients less than the threshold are processed using the soft threshold function to reduce noise:

[0186] SoftThreshold(X,T)=sign(X)*max(|X|-T,0)

[0187] Where X is the wavelet coefficient, T is the threshold, and sign(X) is the sign function, indicating the positive or negative sign of the coefficient;

[0188] Then, the signal is reconstructed, and the processed wavelet coefficients are used for inverse wavelet transform (Inverse Wavelet Transform), and the processed coefficients are reconstructed into an ultrasonic signal with noise removed;

[0189] Step 1.6: Vibration signal denoising process, outlier processing, Zscore calculation:

[0190] Calculate the Zscore value of the vibration signal data using the mean and standard deviation of the vibration data. The formula is as follows:

[0191]

[0192] Step 1.7, determine the outliers:

[0193] Set a threshold, and when the absolute value of Zscore is greater than 3, identify the data points exceeding the threshold as outliers; replace the outliers with interpolation for the data points identified as outliers;

[0194] Step 1.8: Filtering operation to remove noise components within a specific frequency range:

[0195] The noise frequency range to be removed, the digital filter uses a difference equation to describe the behavior of the discrete time system. Using the IIR filter (InfiniteImpulseResponse) in the form of a difference equation, the form is expressed as:

[0196] y[n]=b0x[n]+b1x[n-1]+...+b m x[nm]-a1y[n-1]-...-a n y[nn]

[0197] Where x[n] is the value of the input signal at time n, y[n] is the value of the output signal at time n, b0, b1, …b m are the coefficients of the input signal, a1…a n is the coefficient of the output signal, and the digital filter is applied to the vibration signal data and processed using the filter;

[0198] Step 1.9: Integrate the above processed data, and record the sample data set as X = [x1, x2, x3, …, x i ],i∈[1,N],x i Represents the set of preprocessed signals.

[0199] Step 2: Perform feature extraction on the preprocessed data, and use the transformed signal to extract frequency domain and time domain feature values ​​for the parameter input layer of SVM.

[0200] Step 2 is implemented as follows:

[0201] Step 2.1, feature selection, extract the eigenvalue of the signal after Fourier transformation after denoising, and let the eigenvalue extracted from the ultrasonic signal, the pulse factor: Margin Factor: Crest Factor: X peakF =X rms / X peak , and the kurtosis of the eigenvalues ​​extracted by the vibration sensor: Skewness: As [a i1 ,a i2 ,ai3 ,a i4 ,a i5 ], recorded as a set of sample data a i =[a i1 ,a i2 ,a i3 ,a i4 ,a i5 ] T , i∈[1,N], sample data a i Corresponding to the actual result Y i =[y i1 ,y i2 ,y i3 ,…,y in ] T , where Y i Represents the signal y generated by each free metal foreign body in A collection of;

[0202] Step 2.2, the sample data set is recorded as X = [x1, x2, x3, ..., xi], i∈[1, N], the metal foreign bodies inside the GIS include n kinds, the material parameters at least include copper metal foreign bodies, iron metal foreign bodies, and aluminum metal foreign bodies; the geometric parameters at least include spherical metal foreign bodies, linear metal foreign bodies, sheet metal foreign bodies, spindle-shaped metal foreign bodies, and irregular metal foreign bodies;

[0203] Normalization process, for each feature, calculate its mean μ and standard deviation σ:

[0204] For each eigenvalue X, use the normalization method Zscore to get the normalized eigenvalue X normalized , the corresponding formula is:

[0205]

[0206] Step 2.3, update the data set and normalize the eigenvalue a' i =[a' i1 ,a' i2 ,a' i3 ,a' i4 ,a' i5 ] T , i∈[1,N], replace the original eigenvalue a i =[a i1 ,a i2 ,a i3 ,a i4 ,a i5 ] T , i∈[1,N], update the data set for subsequent machine learning model training and prediction;

[0207] Step 2.4: Construct feature vectors, label the data set, and add corresponding labels and category identifiers to each feature vector.

[0208] Step 3: Input the extracted eigenvalues ​​into the support vector machine (SVM) model and set the initial hyperparameters kernel scale: γ (KernelScale) and box constraint: C (BoxConstraint) to identify and classify the geometric parameters and material parameters of foreign bodies.

[0209] Step 4: Use the Sparrow algorithm to optimize the hyperparameters γ and C of SVM to improve the accuracy of the optimized classification model, and obtain the Sparrow algorithm optimized support vector machine for free metal classification and recognition detection method inside GIS; ensure the accurate recognition of the geometric parameters and material parameters of metal foreign bodies.

[0210] Step 4 is implemented according to the following steps:

[0211] Step 4.1, use the sparrow algorithm to optimize the support vector machine parameters, initialize the sampling point set and the proxy model construction, randomly select n0 sampling points within the set parameter search range, and use the average test accuracy of ten-fold cross validation as the objective function f. Different parameter combinations are used as independent variables x to form the proxy model framework, and the initial distribution of the objective function and the sampling point set D are obtained;

[0212] Step 4.2: Sampling point selection: select the next sampling point x by maximizing the acquisition function t , to obtain the function value f(x t ), the sparrow algorithm simulates the behavior of sparrows looking for food and selects the next optimal sampling point according to the current sampling point set D;

[0213] Step 4.3: Update the proxy model and set the new sampling point [x t ,f(x t )] is added to the sampling point set D, and the proxy model is updated so that the proxy model is more consistent with the distribution of the objective function. This can be achieved by adjusting the sparrow behavior in the sparrow algorithm to better search the parameter space;

[0214] Step 4.4, iterative optimization, set a maximum number of iterations, when the number of iterations reaches the maximum number, stop the algorithm iteration, in each iteration, repeat steps 2 and 3, find the next optimal sampling point, and finally output the optimal sampling point and the corresponding optimal value of the objective function, that is, the optimal parameters C and γ of the SVM model.

[0215] Step 5: Use the optimized SVM model to classify metal foreign bodies in GIS tanks, and then further optimize the model for different types of foreign bodies to improve the accuracy and reliability of classification, verify the GIS internal metal foreign body recognition model, optimize the support vector machine based on the sparrow algorithm to classify and identify free metals in GIS, and obtain the type combination data for the recognition of metal foreign bodies in GIS.

[0216] Step 5 is implemented according to the following steps:

[0217] Step 5.1: Use the parameters of metal foreign bodies in GIS to identify and diagnose, and use the test sample x in the test sample set. i Perform the test and get the test result Y text ;

[0218] Step 5.2: According to the test sample x i The actual result Y i Evaluation is performed to determine the test result Y text the accuracy of the information provided and draw conclusions based on it;

[0219] Step 5.3: Calculate the accuracy of the test sample set and compare whether it exceeds the preset accuracy threshold. When the accuracy is higher than this threshold, it can be regarded that the GIS internal insulation fault diagnosis model has high accuracy:

[0220] Calculate the accuracy Z text :

[0221]

[0222] Among them, Y i =[y i1 ,y i2 ,y i3 ,…,y in ] T , Y text =[y text1 ,y text2 ,y text3 ,…,y textn ] T ;

[0223] If the test result is Y text The accuracy Z text If it is not less than z%, then the test result is judged to be Y text To be accurate;

[0224] Step 5.4: Determine whether the model meets the termination conditions: If not, update the parameters. If the error meets the preset accuracy or reaches the maximum number of iterations, output the type of metal foreign body.

[0225] Step 5.5: By statistically analyzing the test sample set, if the accuracy rate exceeds the pre-set standard, the free metal classification and identification detection method model within the GIS is considered to be accurate.

[0226] Example 3

[0227] The present invention uses the sparrow algorithm to optimize the support vector machine to classify and identify free metals in GIS. The flow chart is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0228] Step 1: Use acquisition equipment to obtain the ultrasonic and vibration signals emitted by metal foreign bodies in the GIS tank, and preprocess the data, including denoising and signal transformation (Fourier and modal analysis) to prepare for feature extraction.

[0229] Step 1 is implemented according to the following steps:

[0230] Step 1.1, place a metal foreign body with known parameters inside the GIS tank, fill the GIS equipment with 0.35-0.65MPa SF6, apply 220kV industrial frequency AC voltage, and collect and record ultrasonic and vibration signals inside the GIS equipment cavity;

[0231] Step 1.2: The signal collected by the sensor is a discrete signal. Since computers can only process sequences of finite length, given a one-dimensional finite vector, the discrete Fourier transform is expressed as:

[0232]

[0233] in, The computational complexity is O(N 2 ), when N is large, the computational complexity of the above Fourier transform is huge. Using Cooley and Tukey fast Fourier transform algorithm, assuming N = 2 M , M is a positive integer, x(n) is divided into two groups according to duality, that is, let n = 2r and n = 2r + 1, then It is expressed as:

[0234]

[0235] According to the decomposition method of the above formula, x(k) is finally decomposed into subsequences of 2 points;

[0236] Step 1.3, ultrasonic signal preprocessing:

[0237] Remove outliers: Identify and remove abnormal data caused by equipment failure or interference;

[0238] Distance-based outlier detection: Use the distance between a data point and its nearest neighbor to identify outliers that are far away from other data points. Outlier processing uses the standard deviation method, and the formula is as follows:

[0239] |x-μ|>kσ

[0240] x is the data point, μ is the mean of the data, σ is the standard deviation of the data, k is the threshold, and the value of k is 2 to 4;

[0241] Step 1.4: Signal smoothing: Apply filtering or smoothing techniques to reduce noise interference:

[0242] Use an exponentially weighted moving average (EMA) to smooth the data and give more weight to recent data points. The formula is as follows:

[0243] EMA t =α·x t +(1-α) EMA t-1

[0244] Among them, EMA t is the exponentially weighted moving average at time t, x t is the original data point at time t, α is the smoothing coefficient, which ranges from 0 to 1;

[0245] Step 1.5: Based on the characteristics of ultrasonic signals, wavelet transform is selected for denoising:

[0246] First, soft threshold processing is performed on the wavelet coefficients, and the coefficients less than the threshold are processed using the soft threshold function to reduce noise:

[0247] SoftThreshold(X,T)=sign(X)*max(|X|-T,0)

[0248] Where X is the wavelet coefficient, T is the threshold, and sign(X) is the sign function, indicating the positive or negative sign of the coefficient;

[0249] Then, the signal is reconstructed, and the processed wavelet coefficients are used for inverse wavelet transform (Inverse Wavelet Transform), and the processed coefficients are reconstructed into an ultrasonic signal with noise removed;

[0250] Step 1.6: Vibration signal denoising process, outlier processing, Zscore calculation:

[0251] Calculate the Zscore value of the vibration signal data using the mean and standard deviation of the vibration data. The formula is as follows:

[0252]

[0253] Step 1.7, determine the outliers:

[0254] Set a threshold, and when the absolute value of Zscore is greater than 3, identify the data points exceeding the threshold as outliers; replace the outliers with interpolation for the data points identified as outliers;

[0255] Step 1.8: Filtering operation to remove noise components within a specific frequency range:

[0256] The noise frequency range to be removed, the digital filter uses a difference equation to describe the behavior of the discrete time system. Using the IIR filter (InfiniteImpulseResponse) in the form of a difference equation, the form is expressed as:

[0257] y[n]=b0x[n]+b1x[n-1]+...+b m x[nm]-a1y[n-1]-...-a n y[nn]

[0258] Where x[n] is the value of the input signal at time n, y[n] is the value of the output signal at time n, b0, b1, …b m are the coefficients of the input signal, a1…a n is the coefficient of the output signal, and the digital filter is applied to the vibration signal data and processed using the filter;

[0259] Step 1.9: Integrate the above processed data, and record the sample data set as X = [x1, x2, x3, …, x i ],i∈[1,N],x i Represents the set of preprocessed signals.

[0260] Step 2: Perform feature extraction on the preprocessed data, and use the transformed signal to extract frequency domain and time domain feature values ​​for the parameter input layer of SVM.

[0261] Step 2 is implemented as follows:

[0262] Step 2.1, feature selection, extract the eigenvalue of the signal after Fourier transformation after denoising, and let the eigenvalue extracted from the ultrasonic signal, the pulse factor: Margin Factor: Crest Factor: X peakF =X rms / X peak , and the kurtosis of the eigenvalues ​​extracted by the vibration sensor: Skewness: As [a i1 ,a i2 ,a i3 ,a i4 ,ai5 ], recorded as a set of sample data a i =[a i1 ,a i2 ,a i3 ,a i4 ,a i5 ] T , i∈[1,N], sample data a i Corresponding to the actual result Y i =[y i1 ,y i2 ,y i3 ,…,y in ] T , where Y i Represents the signal y generated by each free metal foreign body in A collection of;

[0263] Step 2.2, the sample data set is recorded as X = [x1, x2, x3, ..., xi], i∈[1, N], the metal foreign bodies inside the GIS include n kinds, the material parameters at least include copper metal foreign bodies, iron metal foreign bodies, and aluminum metal foreign bodies; the geometric parameters at least include spherical metal foreign bodies, linear metal foreign bodies, sheet metal foreign bodies, spindle-shaped metal foreign bodies, and irregular metal foreign bodies;

[0264] Normalization process, for each feature, calculate its mean μ and standard deviation σ:

[0265] For each eigenvalue X, use the normalization method Zscore to get the normalized eigenvalue X normalized , the corresponding formula is:

[0266]

[0267] Step 2.3, update the data set and normalize the eigenvalue a' i =[a' i1 ,a' i2 ,a' i3 ,a' i4 ,a' i5 ] T , i∈[1,N], replace the original eigenvalue a i =[a i1 ,a i2 ,a i3 ,a i4 ,a i5 ] T , i∈[1,N], update the data set for subsequent machine learning model training and prediction;

[0268] Step 2.4: Construct feature vectors, label the data set, and add corresponding labels and category identifiers to each feature vector.

[0269] Step 3: Input the extracted eigenvalues ​​into the support vector machine (SVM) model and set the initial hyperparameters kernel scale: γ (KernelScale) and box constraint: C (BoxConstraint) to identify and classify the geometric parameters and material parameters of foreign bodies.

[0270] Step 3 is implemented as follows:

[0271] Step 3.1: Convert to convex function

[0272]

[0273] sty i (w T x i +b)≥1,i=1,2,...,m

[0274] Then Convert to When solving convex functions, 1 / 2 is used to facilitate calculation after differentiation.

[0275] Use Lagrange multiplier method and KKT conditions to solve the optimal value:

[0276] Step 3.2, after solving, substitute into the hyperplane model:

[0277] Available

[0278] The above is the dual form of the SVM dual problem;

[0279] Step 3.3: If the data is linearly inseparable, then introduce a slack variable ξ≥0, so that the function interval plus the slack variable ≥1, then the objective function is:

[0280]

[0281] Dual problem:

[0282]

[0283] C is the penalty parameter. The larger the C value, the greater the penalty for classification. First, use the Lagrange multiplier method to obtain the Lagrange function, and then solve its dual problem.

[0284] Step 4: Use the Sparrow algorithm to optimize the hyperparameters γ and C of SVM to improve the accuracy of the optimized classification model, and obtain the Sparrow algorithm optimized support vector machine for free metal classification and recognition detection method inside GIS; ensure the accurate recognition of the geometric parameters and material parameters of metal foreign bodies.

[0285] Step 4 is implemented according to the following steps:

[0286] Step 4.1, use the sparrow algorithm to optimize the support vector machine parameters, initialize the sampling point set and the proxy model construction, randomly select n0 sampling points within the set parameter search range, and use the average test accuracy of ten-fold cross validation as the objective function f. Different parameter combinations are used as independent variables x to form the proxy model framework, and the initial distribution of the objective function and the sampling point set D are obtained;

[0287] Step 4.2, sampling point selection, select the next sampling point x by maximizing the acquisition function t , to obtain the function value f(x t ), the sparrow algorithm simulates the behavior of sparrows looking for food and selects the next optimal sampling point according to the current sampling point set D;

[0288] Step 4.3: Update the proxy model and set the new sampling point [x t ,f(x t )] is added to the sampling point set D, and the proxy model is updated so that the proxy model is more consistent with the distribution of the objective function. This can be achieved by adjusting the sparrow behavior in the sparrow algorithm to better search the parameter space;

[0289] Step 4.4, iterative optimization, set a maximum number of iterations, when the number of iterations reaches the maximum number, stop the algorithm iteration, in each iteration, repeat steps 2 and 3, find the next optimal sampling point, and finally output the optimal sampling point and the corresponding optimal value of the objective function, that is, the optimal parameters C and γ of the SVM model.

[0290] Step 5: Use the optimized SVM model to classify metal foreign bodies in GIS tanks, and then further optimize the model for different types of foreign bodies to improve the accuracy and reliability of classification, verify the GIS internal metal foreign body recognition model, optimize the support vector machine based on the sparrow algorithm to classify and identify free metals in GIS, and obtain the type combination data for the recognition of metal foreign bodies in GIS.

Claims

1. The sparrow algorithm optimizes the support vector machine to classify and identify free metals in GIS, which is characterized by: Follow the steps below to implement it: Step 1: Use acquisition equipment to obtain ultrasonic and vibration signals emitted by metal foreign bodies in the GIS tank, and pre-process the data; Step 2: Perform feature extraction on the preprocessed data to extract frequency domain and time domain feature values; Step 3: Input the extracted eigenvalues ​​into the support vector machine (SVM) model, and set the initial hyperparameters kernel scale γ and box constraint C to identify and classify the geometric parameters and material parameters of foreign bodies; Step 4, using the Sparrow algorithm to optimize the hyperparameters γ and C of SVM, and obtaining the Sparrow algorithm optimized support vector machine for free metal classification, identification and detection method inside GIS; Step 5: Use the optimized SVM model to classify metal foreign bodies in the GIS tank, and then further optimize the model for different types of foreign bodies to obtain type combination data for metal foreign body identification in the GIS.

2. The method for classifying and identifying free metals in GIS by using the sparrow algorithm optimized support vector machine according to claim 1 is characterized in that: The step 1 is specifically implemented according to the following steps: Step 1.1, place a metal foreign body with known parameters inside the GIS tank, fill the GIS equipment with 0.35-0.65MPa SF6, apply 220kV industrial frequency AC voltage, and collect and record ultrasonic and vibration signals inside the GIS equipment cavity; Step 1.2: Given a one-dimensional finite vector, the discrete Fourier transform is expressed as: in, The computational complexity is O(N 2 ), assuming N = 2 M , M is a positive integer, x(n) is divided into two groups according to duality, that is, let n = 2r and n = 2r + 1, then It is expressed as: According to the decomposition method of the above formula, x(k) is finally decomposed into subsequences of 2 points; Step 1.3, ultrasonic signal preprocessing: Remove outliers: Identify and remove abnormal data caused by equipment failure or interference; Distance-based outlier detection: Use the distance between a data point and its nearest neighbor to identify outliers that are far away from other data points. Outlier processing uses the standard deviation method, and the formula is as follows: |x-μ|>kσ x is the data point, μ is the mean of the data, σ is the standard deviation of the data, k is the threshold, and the value of k is 2 to 4; Step 1.4: Signal smoothing: Apply filtering or smoothing techniques to reduce noise interference: Use an exponentially weighted moving average (EMA) to smooth the data and give more weight to recent data points. The formula is as follows: EMA t =α·x t +(1-a)·EMA t-1 Among them, EMA t is the exponentially weighted moving average at time t, x t is the original data point at time t, α is the smoothing coefficient, which ranges from 0 to 1; Step 1.5: Based on the characteristics of ultrasonic signals, wavelet transform is selected for denoising: First, soft threshold processing is performed on the wavelet coefficients, and the coefficients less than the threshold are processed using the soft threshold function to reduce noise: SoftThreshold(X,T)=sign(X)*max(|X|-T,0) Where X is the wavelet coefficient, T is the threshold, and sign(X) is the sign function, indicating the positive or negative sign of the coefficient; Then, the signal is reconstructed, and the processed wavelet coefficients are used for inverse wavelet transform, and the processed coefficients are reconstructed into an ultrasonic signal with noise removed; Step 1.6: Vibration signal denoising process, outlier processing, Zscore calculation: Calculate the Zscore value of the vibration signal data using the mean and standard deviation of the vibration data. The formula is as follows: Step 1.7, determine the outliers: Set a threshold, and when the absolute value of Zscore is greater than 3, identify the data points exceeding the threshold as outliers; replace the outliers with interpolation for the data points identified as outliers; Step 1.8: Filtering operation to remove noise components within a specific frequency range: The IIR filter form using the difference equation form is expressed as: y[n]=b0x[n]+b1x[n-1]+...+b m x[n-m]-a1y[n-1]-...-a n y[n-n] Where x[n] is the value of the input signal at time n, y[n] is the value of the output signal at time n, b0, b1, …b m are the coefficients of the input signal, a1…a n is the coefficient of the output signal, and the digital filter is applied to the vibration signal data and processed using the filter; Step 1.9: Integrate the above processed data, and record the sample data set as X = [x1, x2, x3, …, x i ],i∈[1,N],x i Represents the set of preprocessed signals.

3. The method for classifying and identifying free metals in GIS by using the sparrow algorithm optimized support vector machine according to claim 2 is characterized in that: The step 2 is specifically implemented according to the following steps: Step 2.1, feature selection, extract the eigenvalue of the signal after Fourier transformation after denoising, and let the eigenvalue extracted from the ultrasonic signal, the pulse factor: Margin Factor: Crest Factor: X peakF =X rms / X peak , and the kurtosis of the eigenvalues ​​extracted by the vibration sensor: Skewness: As [a i1 ,a i2 ,a i3 ,a i4 ,a i5 ], recorded as a set of sample data a i =[a i1 ,a i2 ,a i3 ,a i4 ,a i5 ] T , i∈[1,N], sample data a i Corresponding to the actual result Y i =[y i1 ,y i2 ,y i3 ,…,y in ] T , where Y i Represents the signal y generated by each free metal foreign body in A collection of; Step 2.2, the sample data set is recorded as X = [x1, x2, x3, ..., xi], i∈[1, N], the metal foreign bodies inside the GIS include n kinds, the material parameters at least include copper metal foreign bodies, iron metal foreign bodies, and aluminum metal foreign bodies; the geometric parameters at least include spherical metal foreign bodies, linear metal foreign bodies, sheet metal foreign bodies, spindle-shaped metal foreign bodies, and irregular metal foreign bodies; Normalization process, for each feature, calculate its mean μ and standard deviation σ: For each eigenvalue X, use the normalization method Zscore to get the normalized eigenvalue X normalized , the corresponding formula is: Step 2.3, update the data set and normalize the eigenvalue a' i =[a' i1 ,a' i2 ,a' i3 ,a' i4 ,a' i5 ] T , i∈[1,N], replace the original eigenvalue a i =[a i1 ,a i2 ,a i3 ,a i4 ,a i5 ] T , i∈[1,N], update the data set for subsequent machine learning model training and prediction; Step 2.4: Construct feature vectors, label the data set, and add corresponding labels and category identifiers to each feature vector.

4. The method for classifying and identifying free metals in GIS by using the sparrow algorithm optimized support vector machine according to claim 3 is characterized in that: The step 3 is specifically implemented according to the following steps: Step 3.1: Convert to convex function s.t.y i (w T x i +b)≥1,i=1,2,...,m Then Convert to Solve for convex functions, Use Lagrange multiplier method and KKT conditions to solve the optimal value: Step 3.2, after solving, substitute into the hyperplane model: Available The above is the dual form of the SVM dual problem; Step 3.3: If the data is linearly inseparable, then introduce a slack variable ξ≥0, so that the function interval plus the slack variable ≥1, then the objective function is: Dual problem: C is the penalty parameter. The larger the C value is, the greater the penalty for classification. First, use the Lagrange multiplier method to obtain the Lagrangian function, and then solve its dual problem.

5. The method for classifying and identifying free metals in GIS by using the sparrow algorithm optimized support vector machine according to claim 4 is characterized in that: The step 4 is specifically implemented according to the following steps: Step 4.1, use the sparrow algorithm to optimize the support vector machine parameters, initialize the sampling point set and the proxy model construction, randomly select n0 sampling points within the set parameter search range, take the average test accuracy of ten-fold cross validation as the objective function f, and use different parameter combinations as independent variables x to form the proxy model framework, and obtain the initial distribution of the objective function and the sampling point set D; Step 4.2: Sampling point selection: select the next sampling point x by maximizing the acquisition function t , to obtain the function value f(x t ), the sparrow algorithm simulates the behavior of sparrows looking for food and selects the next optimal sampling point according to the current sampling point set D; Step 4.3: Update the proxy model and set the new sampling point [x t ,f(x t )] is added to the sampling point set D, and the proxy model is updated so that the proxy model is more consistent with the distribution of the objective function; Step 4.4, iterative optimization, set a maximum number of iterations, when the number of iterations reaches the maximum number, stop the algorithm iteration, in each iteration, repeat steps 2 and 3, find the next optimal sampling point, and finally output the optimal sampling point and the corresponding optimal value of the objective function, that is, the optimal parameters C and γ of the SVM model.

6. The method for classifying and identifying free metals in GIS by using the sparrow algorithm optimized support vector machine according to claim 5 is characterized in that: The step 5 is specifically implemented according to the following steps: Step 5.1: Use the parameters of metal foreign bodies in GIS to identify and diagnose, and use the test sample x in the test sample set. i Perform the test and get the test result Y text ; Step 5.2: According to the test sample x i The actual result Y i Evaluation is performed to determine the test result Y text the accuracy of the information provided and draw conclusions based on it; Step 5.3: Calculate the accuracy of the test sample set and compare whether it exceeds the preset accuracy threshold. When the accuracy is higher than this threshold, it can be regarded that the GIS internal insulation fault diagnosis model has high accuracy: Calculate the accuracy Z text : Among them, Y i = [y i1 , y i2 , y i3 , …, y in T , Y text = [y text1 , y text2 , y text3 , …, y textn T ;​​ If the test result is Y text The accuracy Z text If it is not less than z%, then the test result is judged to be Y text To be accurate; Step 5.4: Determine whether the model meets the termination conditions: If not, update the parameters. If the error meets the preset accuracy or reaches the maximum number of iterations, output the type of metal foreign body.

Images