Analog circuit fault diagnosis method based on IHO-KELM
By using the IHO-KELM method in analog circuit fault diagnosis, using KPCA weighted Mahayana distance and IHO algorithm to find the best, the problem of KELM classification accuracy and efficiency improvement is solved, and a higher fault diagnosis accuracy is achieved.
Patent Information
- Application Number
- CN202510195725.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-21
- Publication Date
- 2025-05-30
AI Technical Summary
In the prior art, in the diagnosis of analog circuit faults, KELM classification accuracy and classification efficiency need to be further improved.
The IHO-KELM-based analog circuit fault diagnosis method is adopted, and the statistical feature set is constructed by collecting the voltage values in different fault modes of the analog circuit, and the Mahayana distance is weighted by KPCA, and the feature set is fused as the feature set for fault diagnosis. At the same time, the IHO algorithm is used to optimize the regularization coefficient C and kernel function parameter g of the KELM network to improve the classification accuracy and efficiency.
The diagnostic accuracy of analog circuit fault diagnosis has been significantly improved. Compared with the ELM, KELM and HO-KELM diagnostic models, the diagnostic accuracy of the IHO-KELM fault diagnosis model has been significantly improved.
Smart Images
Figure CN120067946A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault diagnosis, and particularly to a fault diagnosis method for analog circuits based on IHO-KELM. Background Art
[0002] Fault diagnosis of analog circuits faces many challenges, including the diversity of fault modes, the complexity of fault detection, and the accuracy of fault location.
[0003] Fault diagnosis of analog circuits includes two key steps: feature extraction and fault classification; feature extraction includes: signal decomposition methods such as empirical mode decomposition (EMD), local mean decomposition (LMD), and wavelet analysis; dimensionality reduction methods such as K-L transform, principal component analysis (PCA), and local linear embedding (LLE); statistical methods such as variance, kurtosis, and peak value; distance metric methods such as Euclidean distance (ED), dynamic time warping (DTW), and Mahalanobis distance (MD).
[0004] The method for soft fault diagnosis of pulse power supply based on multi-feature fusion proposed by Zhou Tongyu et al. presents a method of multi-feature fusion and the feasibility of the feature extraction method that fuses weighted Mahalanobis distance features and statistical features; however, the classification accuracy of the Mahalanobis distance weighting method of this method needs to be further improved.
[0005] Fault classification includes: decision trees and CNNs, RNNs and LSTMs, etc., which are all commonly used classification models. However, they have the disadvantages of overfitting when there are many data features and underfitting vice versa; KELM introduces a kernel function, enabling ELM not to set the number of hidden layer neurons, thus reducing the network complexity; KELM introduces adjustable kernel function parameters (g) and regularization coefficients (C), which have an obvious impact on the performance of KELM; however, these two parameters are usually fixed values, resulting in a local optimum problem. Summary of the Invention
[0006] Aiming at the deficiencies of the existing methods, the present invention solves the problem that the classification accuracy and classification efficiency of KELM need to be further improved.
[0007] The technical solution adopted by the present invention is: a fault diagnosis method for analog circuits based on IHO-KELM includes the following steps:
[0008] Step 1: Collect voltage values under different fault modes of the analog circuit and construct a statistical feature set.
[0009] As a preferred embodiment of the present invention, the statistical feature components include: peak value, average value, sample root mean square, kurtosis, and ripple voltage.
[0010] As a preferred embodiment of the present invention, the analog circuit includes a BUCK circuit.
[0011] Step 2: Use KPCA to weight the Mahalanobis distance and construct a fused feature set;
[0012] As a preferred embodiment of the present invention, the formula for using KPCA to weight the Mahalanobis distance is:
[0013]
[0014] where tr(Λ) is the trace of the matrix, S is the 1 covariance matrix of G, μ is the vector composed of the mean values of the G 1 statistical components, G i is the statistical feature set, and C is the covariance matrix of the weight matrix B of KPCA.
[0015] Step 3: Divide the fused feature set into a training set and a test set. Based on the training set, use the IHO algorithm to optimize the regularization coefficient C and the kernel function parameter g of the KELM network; use the test set to test the IHO-KELM model and output the fault classification result; in the IHO algorithm, use the Sobol sequence to initialize the population, use the position update strategy of the sine-cosine algorithm's oscillation and the randomness of the Cauchy distribution, introduce a dynamic Rice step size strategy to balance the global search and local search capabilities, and adjust the shape parameter;
[0016] As a preferred embodiment of the present invention, the formula for using the Sobol sequence to initialize the population is:
[0017]
[0018] where is the upper and lower bounds of the target variable, represents the current position of the i-th individual.
[0019] As a preferred embodiment of the present invention, the position update formula for the oscillation of the sine-cosine algorithm is:
[0020]
[0021] where q is the oscillation parameter, T is the maximum number of iterations, represents the first position update in the stage.
[0022] As a preferred embodiment of the present invention, the position update formula for the randomness of the Cauchy distribution is:
[0023]
[0024] where is the new position finally generated in the stage, and s is the scale parameter of the Cauchy distribution.
[0025] As a preferred embodiment of the present invention, the formula for the shape parameter is:
[0026]
[0027] Where T is the maximum number of iterations, and t is the current number of iterations.
[0028] As a preferred embodiment of the present invention, the analog circuit fault diagnosis system based on IHO-KELM includes: a memory for storing instructions executable by a processor; and a processor for executing the instructions to implement the analog circuit fault diagnosis method based on IHO-KELM.
[0029] As a preferred embodiment of the present invention, a computer-readable medium storing computer program code, the computer program code implementing the analog circuit fault diagnosis method based on IHO-KELM when executed by a processor.
[0030] Advantages of the present invention:
[0031] 1. The present invention extracts the statistical features of the output voltage signal of the analog circuit, and uses the weighted Mahalanobis distance method to process the statistical features, obtains the weighted Mahalanobis distance features, and fuses the two features as the feature set for fault diagnosis;
[0032] 2. By introducing the Sobol sequence initialization, dynamic Rice step size, the oscillation of the sine-cosine algorithm and the randomness strategy of the Cauchy distribution to improve HO, the convergence speed of the algorithm is accelerated and the optimization ability is improved;
[0033] 3. Compared with the ELM, KELM and HO-KELM diagnostic models, the diagnostic accuracy of the IHO-KELM fault diagnosis model has been significantly improved, indicating the feasibility and efficiency of the method of the present invention in analog circuit fault diagnosis. Description of the Drawings
[0034] Figure 1 is the flow chart of the analog circuit fault diagnosis method based on IHO-KELM of the present invention;
[0035] Figure 2 is the optimization effect diagram of the Sobol sequence improved optimization of the present invention;
[0036] Figure 3 is the optimization effect diagram of the position update strategy of the oscillation of the sine-cosine algorithm and the randomness of the Cauchy distribution of the present invention;
[0037] Figure 4 is the optimization effect diagram of the shape parameter of the Rice flight of the present invention;
[0038] Figure 5It is the simulation schematic diagram of the BUCK circuit;
[0039] Figure 6 It is the IHO of the present invention to find the best fitness curve graph;
[0040] Figure 7 It is the comparison graph of the diagnostic results of different diagnostic models. Specific implementation manners
[0041] The present invention will be further described below in conjunction with the accompanying drawings and embodiments. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner. Therefore, it only shows the components related to the present invention.
[0042] As Figure 1 shown, a simulation circuit fault diagnosis method based on IHO-KELM includes the following steps:
[0043] Step 1: Collect the voltage values under different fault modes of the analog circuit and construct a statistical feature set;
[0044] The analog circuit can be a BUCK circuit or other circuits, as long as the circuit can obtain the voltage values under different fault modes is within the protection scope of the present invention.
[0045] The statistical feature components include: peak value p, average value a, root mean square of samples σ, kurtosis k u and ripple voltage r; the corresponding formulas are:
[0046]
[0047] Among them, X is the output voltage signal of the analog circuit; E(X) is the expected value of the data taken, representing the mean value of X; N is the number of samples taken within the same time period; ε is the standard deviation of X; max(X) and min(X) represent the maximum and minimum values within a sample.
[0048] The statistical feature set is m represents the dimension of the statistical feature components, and i represents the number of circuit modes.
[0049] Step 2: Use KPCA to weight the Mahalanobis distance and construct a fusion feature set;
[0050] The Mahalanobis Distance (MD) is an effective method for calculating the similarity of two unknown sample sets; then G i to G 1 The squared form of the Mahalanobis distance can be expressed as:
[0051] D 2 (m) =(G i -μ)S-1 (G i -μ) T (2)
[0052] where μ is the vector composed of the mean values of the statistical components of G 1 and S is the covariance matrix of G 1
[0053] Since MD visualizes the impact of changes in each statistical feature component on circuit performance and does not consider the degree of influence of different statistical feature components on circuit performance, when calculating the Mahalanobis distance, different weights should be assigned to them according to the sensitivity of different feature components to circuit performance, so as to more accurately describe the performance changes of the circuit
[0054] Compared with principal component analysis (PCA), kernel principal component analysis (KPCA) can better preserve the global structure of the data. Therefore, KPCA is used to weight the Mahalanobis distance for further extraction of circuit features. Considering the weight matrix B, assume the squared form of the weighted Mahalanobis distance is
[0055] D 2 (m) =(G i -μ)BS -1 B(G i -μ) T (3)
[0056] Determine B through KPCA
[0057] Map the statistical feature set to a new space through the mapping function φ(x) to obtain a new kernel matrix, and thus analyze the degree of influence on the fault mode. The covariance matrix C after transformation of the new kernel matrix is given by the formula
[0058]
[0059] where n is the number of statistical components. Based on the kernel function K = φφ T , perform eigenvalue decomposition on C
[0060] Assume the non-zero eigenvalues of C: λ 1 ≥λ 2 ≥…≥λ p >0 and the corresponding unit eigenvectors v 1 , v 2 ,…v p . Denote: Λ = diag(λ 1 ,λ 2 ,…,λ p ), V = diag(v 1 ,v 2 ,…,v p ), we get:
[0061]
[0062] The formula for KPCA to weight the Mahalanobis distance is:
[0063]
[0064] Among them, tr(Λ) is the trace of the matrix.
[0065] Thus, the weighted Mahalanobis distance feature set of each fault mode of the circuit is obtained Fusing the fault mode statistical feature set and the weighted Mahalanobis distance feature set of the circuit, the fused feature set F of each fault mode of the circuit is obtained i =(G i ,D i ).
[0066] KPCA maps the original data to a high-dimensional feature space through a kernel function. In this space, the non-linear structure of the data is linearized; this means that in the high-dimensional space, the distribution of the data is closer to the linearly separable state, so that the weighting based on the Mahalanobis distance can more effectively capture the internal structure of the data. In contrast, PCA can only handle linear relationships and cannot effectively remove the noise and redundant information in non-linear data. Therefore, the KPCA of the present invention to weight the Mahalanobis distance can improve the classification accuracy; the results of the comparative experiment are shown in the following table:
[0067] Table 1 Comparison of KPCA and PCA for weighting MD
[0068]
[0069]
[0070] Step 3: Divide the fused feature set into a training set and a test set. Based on the training set, use the IHO algorithm to optimize the regularization coefficient C and the kernel function parameter g of the KELM network; use the test set to test the IHO-KELM model and output the fault classification result; in the IHO algorithm, use the Sobol sequence to initialize the population, introduce the sine-cosine algorithm for oscillation for position update, introduce the dynamic Rice step size strategy to balance the global search and local search capabilities, and adjust the shape parameter;
[0071] KELM is a new machine learning method developed by introducing a kernel function into the extreme learning machine, and its principle is as follows:
[0072] The weight matrix of the extreme learning machine is as the formula:
[0073]
[0074] Among them, γ is the weight matrix; H is the output of the hidden layer nodes; E is the identity matrix; C is the regularization coefficient, which is used to control the complexity of the model and mainly prevent the model from overfitting; Y is the target vector.
[0075] Introducing the kernel matrix, we can get:
[0076] Ω ELM = HH T = h(x i )h(x j ) = K(x i , x j )(8)
[0077] Among them, Ω ELM is the kernel matrix, x i and x j are input vectors, and h(x i ) and h(x j ) are the output functions of the hidden layer.
[0078] The kernel function has a great influence on the model performance. To improve the generalization of KELM, the radial basis function (RBF) is selected. g is the parameter of the kernel function, and its expression is:
[0079]
[0080] The Hippopotamus Optimization Algorithm (HO) is a swarm intelligence optimization algorithm. It draws inspiration from the inherent behaviors of hippopotamus individuals (abbreviated as "individuals"), including: 1. The movement of individuals in the river; 2. The defense of individuals against predators; 3. The escape of individuals from predators; The position update methods in these three stages are used to achieve multi-objective optimization.
[0081] In the HO algorithm, the population initialization is generated by the traditional random method, which has uneven distribution and affects the global optimization ability; the position update methods in the first and third stages rely too much on random quantities, which will lose the population diversity, making the algorithm easy to fall into local optimum or have too low convergence accuracy; the position update method in the second stage simulates the defense of individuals against predators through Lévy flight, and the fixed Lévy step size makes the algorithm lack flexibility in the optimization process, resulting in low solution accuracy and poor convergence performance; therefore, the following improvements are made for these shortcomings.
[0082] The IHO algorithm includes:
[0083] Step 31: Initialize the population using the Sobol sequence;
[0084] The Sobol sequence is a low-discrepancy sequence, also known as a quasi-random sequence, which generates points in a multi-dimensional space that are more evenly distributed than traditional random sequences. Assuming the hippopotamus population is X, then d represents the variable dimension, represents the current position of the i-th individual;
[0085] Let the random number S generated by the Sobol sequence n ∈ [0, 1], then the initial position of the population is defined as:
[0086]
[0087] Among them, are the upper and lower bounds of the target variable.
[0088] The population generated by the Sobol sequence is evenly distributed in the solution space; this uniformity enables the algorithm to more effectively explore the global in the initial stage, avoiding insufficient local search caused by uneven population distribution, thereby accelerating the convergence speed, and testing with a unimodal function;
[0089]
[0090] The dimension is uniformly set to 10, the range of the target variable is [-100, 100], and the theoretical value is 0; the test results are as Figure 2 shown, it can be seen that the number of iterations after initialization of the Sobol sequence is significantly reduced and the convergence speed is accelerated.
[0091] Step 32: Utilize the position update strategy of the oscillation of the sine-cosine algorithm and the randomness of the Cauchy distribution;
[0092] The individual moves in the river and evades predators to update its position, as shown in the following formula:
[0093]
[0094] Among them, t is the current iteration number, is the global optimal individual position at the t-th iteration, is the next-generation individual position, and are the lower and upper bounds of the local dimension after t iterations.
[0095] And:
[0096]
[0097] Among them, d represents the dimension; r and I represent random vectors.
[0098] Step 321: Introduce the oscillation of the sine-cosine algorithm to update the position, and the formula is:
[0099]
[0100] Among them, q is the oscillation parameter, and T is the maximum number of iterations. Indicates the first position update in the phase.
[0101] Step 332: Introduce the Cauchy distribution to perform the second position update in a certain phase;
[0102]
[0103] Among them, is the new position finally generated in the phase, s is the scale parameter of the Cauchy distribution, which affects the strength of randomness. Combining the oscillation of the sine-cosine algorithm with the Cauchy random distribution helps the algorithm jump out of the local optimal solution and explore new regions.
[0104] The values of the sine-cosine function fluctuate between [-1, 1]. When the function value is close to 0, the candidate solution will perform local search in the neighborhood of the current optimal solution; while when the function value is close to ±1, the candidate solution will move away from the current optimal solution, thus realizing global exploration; the random perturbation range of the Cauchy distribution is relatively large, which can make a large-scale adjustment to the current solution; this adjustment enables the individual to jump out of the neighborhood of the local optimal solution and then explore new regions, thereby enhancing the global search ability; perform test experiments with multimodal functions
[0105]
[0106] The dimension is uniformly set to 10, the range of the target variable is [-100, 100], the theoretical value is 0, and the test results are as Figure 3 shown.
[0107] Step 34: Introduce a dynamic Lévy flight step size;
[0108] The position update of the individual defending against the predator is as follows:
[0109]
[0110] Among them, is the randomly generated position of the predator, represents the Lévy flight step size of the i-th individual, which is used to simulate the long-distance jumping random movement of the individual when defending against the predator; β is the shape parameter of the Lévy distribution, and c is the step size adjustment factor.
[0111] and is a d-dimensional vector randomly selected from the normal distribution random variables u and v:
[0112]
[0113] Among them, τ u is the extended parameter of the Lévy flight and is affected by β.
[0114] A dynamic Lévy step - size strategy is added to balance the global search and local search capabilities, and the shape parameter β is adjusted as follows:
[0115]
[0116] Among them, T is the maximum number of iterations, and t is the current iteration number.
[0117] Using the sin function can avoid excessive amplitude changes during the iteration process, thus missing potential better solutions.
[0118] Lévy flight is a random - walk strategy based on non - Gaussian random distribution. Its step - size follows the Lévy distribution and has long - tail characteristics. The algorithm of Lévy flight combines a dynamic adjustment mechanism, and dynamically adjusts the search step - size according to the situation during the iteration process (such as the quality of the current solution, the number of iterations, etc.). This can effectively break the aggregation phenomenon of the population and increase population diversity. The increase in diversity enables the algorithm to explore the solution space from multiple directions, avoiding search stagnation caused by the population prematurely aggregating near the local optimal solution, and a unimodal function is used for the test experiment; the test results are as Figure 4 shown.
[0119] Experimental process:
[0120] Use Simulink to simulate different fault modes of the experimental circuit; each fault mode is subjected to 100 Monte Carlo simulations, the time interval is set to 10 units, and the output voltage signal is monitored. Obtain 1000 output voltage signals of each fault mode of the circuit as the original data.
[0121] Calculate the statistical characteristics of each simulation test of each fault mode, that is, 5 statistical components, with 100 data for each statistical component, as the statistical - feature set. Calculate the weighted Mahalanobis distance of the fault - free mode set itself, and then calculate the weighted Mahalanobis distances of other fault modes based on the mean and covariance matrix of the statistical components of the fault - free mode set; fuse the statistical - feature set and the weighted Mahalanobis - distance feature set to obtain the fused - feature sample set of each fault mode of the circuit.
[0122] Based on the fused - feature sample set, divide it into a training set and a test set. Input the training set into the improved HO - optimized KELM model to find the optimal C and g to make the model more accurate, and thus establish an IHO - KELM analog - circuit fault - diagnosis model.
[0123] Use the IHO - KELM model to classify the faults of the test set and obtain the results of analog - circuit fault diagnosis.
[0124] To verify the effectiveness of the method of the present invention in the soft fault diagnosis of analog circuits, a non-ideal BUCK analog circuit with an input voltage of 20V - 10V is selected for simulation experiments; a BUCK circuit simulation model is established in Simulink, and SW1 and SW2 are controlled by a signal generator. The parameters of each device in the circuit are as Figure 5 shown:
[0125] Circuit fault mode setting:
[0126] Perform sensitivity analysis on the BUCK circuit, and set C, L, R ON , R L , R F , R ESR as the faulty components for discussion; among them, R ON is the equivalent resistance of the MOSFET power transistor, R L is the equivalent resistance of the inductor L, R F is the equivalent resistance of the diode, R ESR is the equivalent resistance of the capacitor. Set the component fault value to be ±50% beyond the nominal value of the component. Set the resistor tolerance to 5%, and the inductor and capacitor tolerances to 10%; F1 represents the no-fault mode, and F2 - F7 represent the other 6 fault modes. The corresponding fault mode settings of the circuit are shown in Table 1:
[0127] Table 2 Fault Modes of the BUCK Circuit
[0128]
[0129] Circuit feature extraction:
[0130] Perform 100 Monte Carlo simulation experiments on different modes of the BUCK circuit through Simulink, and set the simulation time to 10 unit lengths each time. Calculate the statistical features of the original data of each mode, and then calculate the weighted Mahalanobis distance using the statistical features. There are a total of 700 groups of samples for 7 fault modes; a partial feature fusion set is shown in Table 3 below.
[0131] Table 3 Partial Fusion Feature Set of Statistical Features and Weighted Mahalanobis Distance
[0132]
[0133] Fault diagnosis and result analysis
[0134] To verify the superiority of the method of the present invention, experiments select ELM, KELM, HO-KELM, and IHO-KELM for comparison; randomly select 75% of the fusion feature set (700 groups of samples) as the training set and 25% as the test set to test these four fault diagnosis models.
[0135] First, the HO-KELM and IHO-KELM models are trained using the training set. Taking the error rate of the training set as the fitness, the fitness change curve is as Figure 6 shown.
[0136] As can be seen from Figure 6 , the IHO-KELM model can find the optimal fitness faster than the HO-KELM, and the fitness of the IHO-KELM is better. This shows that the IHO-KELM improves the global search ability compared with the HO-KELM, while accelerating the convergence speed and improving the fault diagnosis efficiency.
[0137] Finally, the diagnostic results of the four models are as Figure 7 shown.
[0138] From the comparison of the experimental results, it can be seen that the classification problems of the ELM (accuracy rate 92%) and KELM (accuracy rate 93.7143%) diagnostic models are relatively prominent in the case of multi-fault modes, resulting in a low diagnostic accuracy rate; on the premise of optimizing KELM with HO, the diagnostic accuracy rate of HO-KELM (accuracy rate 95.4286%) is significantly higher than the above two models, but there are still some ambiguities in the classification of F1, F2, and F7; through the improvement of the HO algorithm, the IHO-KELM (accuracy rate 98.2857%) diagnostic model further clarifies the classification problems of F1, F2, and F7. Compared with the above three models, the diagnostic accuracy rate is significantly improved, and the diagnostic performance is better than the other three models.
[0139] The present invention extracts the statistical features of the output voltage signal of the analog circuit, processes the statistical features using the weighted Mahalanobis distance method to obtain the weighted Mahalanobis distance features, and fuses the two features as the feature set for fault diagnosis; then, by introducing the Sobol sequence initialization, dynamic Rice step size, the oscillation of the sine-cosine algorithm, and the randomness strategy of the Cauchy distribution to improve HO, the convergence speed of the algorithm is accelerated and the optimization ability is improved; the experimental results show that compared with the ELM, KELM, and HO-KELM diagnostic models, the diagnostic accuracy rate of the IHO-KELM fault diagnosis model is significantly improved, proving the feasibility and efficiency of the method of the present invention in analog circuit fault diagnosis.
[0140] Inspired by the ideal embodiments of the present invention described above, through the above description, relevant staff can completely make various changes and modifications without departing from the technical idea of the present invention. The technical scope of the present invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.
Claims
1. An analog circuit fault diagnosis method based on IHO-KELM, characterized in that: The following steps are involved: Step 1: Collect voltage values of analog circuits under different fault modes and construct a statistical feature set; Step 2: Use KPCA to weight the Mahalanobis distance and construct a fusion feature set; Step 3: Use the IHO algorithm to optimize the regularization coefficient C and kernel function parameter g of the KELM network; use statistical features and fusion features to train the improved KELM network and output the fault classification results; In the IHO algorithm, the Sobol sequence is used to initialize the population, the position update strategy is based on the oscillation of the sine and cosine algorithm and the randomness of the Cauchy distribution, and the dynamic Rice step strategy is introduced to balance the capabilities of global search and local search, and the shape parameters are adjusted.
2. The analog circuit fault diagnosis method based on IHO-KELM according to claim 1, characterized in that: The formula for weighting the Mahalanobis distance using KPCA is: Among them, tr(Λ) is the trace of the matrix, S is the G1 covariance matrix, μ is the vector composed of the means of the statistical components of G1, G i is the statistical feature set, and C is the covariance matrix of the KPCA weight matrix B.
3. The analog circuit fault diagnosis method based on IHO-KELM according to claim 1, characterized in that: The formula for initializing the population using the Sobol sequence is: in, are the upper and lower bounds of the target variable, Represents the current position of the i-th individual.
4. The analog circuit fault diagnosis method based on IHO-KELM according to claim 1, characterized in that: The oscillatory position update formula of the sine-cosine algorithm is: Among them, q is the oscillation parameter, T is the maximum number of iterations, Indicates the first position update in the phase.
5. The analog circuit fault diagnosis method based on IHO-KELM according to claim 4, characterized in that: The position update formula of Cauchy distribution randomness is: in, is the new position generated at the end of the stage, and s is the scale parameter of the Cauchy distribution.
6. The analog circuit fault diagnosis method based on IHO-KELM according to claim 1, characterized in that: The formula for the shape parameter is: Among them, T is the maximum number of iterations, and t is the current number of iterations.
7. The analog circuit fault diagnosis method based on IHO-KELM according to claim 1, characterized in that: The statistical characteristic components include: peak value, average value, sample RMS value, kurtosis and ripple voltage.
8. The analog circuit fault diagnosis method based on IHO-KELM according to claim 1, characterized in that: The analog circuit includes a BUCK circuit.
9. The analog circuit fault diagnosis system based on IHO-KELM is characterized by: include: a memory for storing instructions executable by a processor; A processor, configured to execute instructions to implement the analog circuit fault diagnosis method based on IHO-KELM as described in any one of claims 1 to 8.
10. A computer readable medium storing computer program code, characterized in that: When the computer program code is executed by a processor, the computer program code implements the analog circuit fault diagnosis method based on IHO-KELM as described in any one of claims 1 to 8.