Valve leakage diagnosis method and system based on improved Hippo algorithm optimized SVM

Abstract

The present invention proposes a valve leakage diagnosis method and system based on an improved Hippo algorithm to optimize SVM, which relates to the field of valve leakage diagnosis technology, including: collecting original signals; decomposing the collected original signals to obtain a finite number of IMF components; calculating fuzzy entropy and constructing feature vectors; using the feature vectors as data samples and dividing them into a training set, a validation set, and a test set; using the training set to train an SVM model, and iteratively optimizing the parameters of the SVM model based on the improved Hippo algorithm according to the classification error rate of the validation set on the trained SVM model to obtain an optimized SVM model; inputting the test set into the optimized SVM model to obtain a valve leakage diagnosis model; obtaining a new acoustic emission signal and inputting it into the valve leakage diagnosis model to obtain a diagnosis result. The present invention can not only improve the classification accuracy and generalization ability of the diagnosis model, but also enhance the stability of the model under different working conditions and environmental noise.

Description

Technical Field

[0001] The present invention relates to the technical field of valve leakage diagnosis, and in particular to a valve leakage diagnosis method and system based on an improved Hippo algorithm to optimize SVM. Background Art

[0002] Valves, as essential components in everyday life, play a crucial role in regulating the flow direction and volume of media, opening and closing pipelines, and other important functions. Their sealing performance directly impacts their functionality. During valve operation, internal leakage, once detected, is more concealed than external leakage, making accurate diagnosis of valve internal leakage crucial. Early research focused on fault classification using traditional signal processing methods and unoptimized support vector machines (SVMs). While these methods lay the foundation for valve internal leakage detection, they exhibit limitations when processing complex, nonlinear, and nonstationary signals, and their classification accuracy needs to be improved.

[0003] As research deepened, scholars began to introduce more advanced signal processing techniques, such as empirical mode decomposition (EMD) and complete ensemble empirical mode decomposition (CEEMDAN), to better extract fault features. At the same time, to improve the classification performance of SVMs, researchers began exploring various intelligent optimization algorithms to optimize key SVM parameters. These algorithms, including the particle swarm optimization (PSO) and the grey wolf algorithm (GWO), have significantly improved the efficiency of SVM parameter selection and classification accuracy.

[0004] However, despite the introduction of these optimization algorithms, valve internal leakage fault diagnosis technology still faces several challenges. First, existing optimization algorithms still have room for improvement in terms of global search capabilities and convergence speed, and are prone to falling into local optimal solutions. Second, these algorithms are sensitive to the quality of the initial population, which may affect the final optimization results. In addition, during the SVM parameter optimization process, existing methods often only focus on classification accuracy. These issues limit the overall performance and practical application effects of the fault diagnosis system. More efficient and robust optimization methods are urgently needed to further improve the accuracy and reliability of valve internal leakage fault diagnosis. Summary of the Invention

[0005] In view of this, the present invention proposes a valve leakage diagnosis method and system based on an improved Hippo algorithm to optimize the SVM. This method improves the efficiency and accuracy of SVM parameter optimization by using an improved Hippo algorithm (IHO). It also combines CEEMDAN signal decomposition and fuzzy entropy feature extraction techniques to better capture the nonlinear and nonstationary characteristics of valve leakage signals. This method not only improves the classification accuracy and generalization ability of the diagnostic model, but also enhances the model's stability under different operating conditions and environmental noise. Through these improvements, the solution aims to provide a more reliable and efficient solution for valve leakage fault diagnosis in industrial practice.

[0006] The technical solution of the present invention is achieved as follows:

[0007] In one aspect, the present invention provides a valve leakage diagnosis method based on an improved Hippo algorithm optimized SVM, comprising:

[0008] S1. Collect acoustic emission signals under different pressure and valve opening conditions as original signals;

[0009] S2, performing adaptive white noise complete empirical mode decomposition on the collected original signal to obtain a finite number of IMF components;

[0010] S3, calculating the correlation coefficient between each IMF component and the original signal, selecting the IMF component whose correlation coefficient is greater than a preset threshold, calculating the fuzzy entropy of the selected IMF component, and constructing a feature vector;

[0011] S4. Use the feature vector as a data sample and divide it into training set, validation set and test set;

[0012] S5. Use the training set to train the SVM model. According to the classification error rate of the validation set on the trained SVM model, the parameters of the SVM model are iteratively optimized based on the improved Hippo algorithm to obtain the optimal parameters and the optimized SVM model.

[0013] S6. Input the test set into the optimized SVM model, predict the valve status, calculate the classification accuracy, and evaluate the model performance. Fine-tune the model based on the evaluation results to obtain a valve leakage diagnosis model.

[0014] S7. Obtain a new acoustic emission signal, decompose and extract features from the signal, and input it into a valve leakage diagnosis model to obtain a diagnosis result.

[0015] Based on the above solution, preferably, step S5 includes:

[0016] S51, selecting the penalty parameter c and the kernel parameter g of the SVM model as target optimization parameters, and logarithmically encoding the penalty parameter c and the kernel parameter g into individual positions in the Hippo algorithm;

[0017] S52, setting the population size N, the maximum number of iterations T and upper and lower boundary parameters, and initializing the population using a combined chaotic map;

[0018] S53, for each parameter combination (c i ,g i ), use the training set to train the SVM model and calculate the fitness value on the validation set;

[0019] S54, iteratively updating the population according to the improved Hippo algorithm, performing a local search for the current optimal solution in each iteration, and outputting the optimal solution, i.e., the optimal parameter combination, when the iteration stop condition is reached;

[0020] S55. Perform antilogarithmic transformation on the optimal parameter combination to obtain optimal parameters c and g.

[0021] Based on the above solution, preferably, in step S52, the combined chaotic map includes a Logistic map and a Tent map, and initializing the population using the combined chaotic map includes:

[0022] S521. Design a hybrid mapping function to determine the mapping to be used through conditional judgment. The hybrid mapping function is as follows:

[0023]

[0024] Where r is a random number between [0,1], μ is the control parameter of the Logistic mapping;

[0025] S522. Determine the dimension m of the problem space, generate an initial value x0 for each dimension, repeatedly apply the above hybrid mapping function multiple times for each dimension to generate a chaotic sequence, and take the final value as the initialization chaotic sequence value of the dimension;

[0026] S523, mapping the generated chaotic sequence value to the upper and lower bounds of the hippo population search space, the position X of each hippo individual i Expressed as: X i =(x i1 ,x i2 ,…,x im ), let z be the generated chaotic sequence value, then the initial position of the hippopotamus individual is calculated as follows:

[0027] X i :x ii =ll j +z·(ul j -ll j ), i=1,2,...,N; j=1,2,...,m

[0028] Where xij is the position of the i-th hippopotamus in the j-th dimension, ll j and ul j are the lower and upper bounds of the j-th dimension respectively;

[0029] S524. Repeat steps S522-S523 for each hippopotamus individual and each dimension in the population until the entire population is initialized.

[0030] Based on the above solution, preferably, the calculation formula of the fitness value is:

[0031]

[0032] Where E is the classification error rate on the validation set, n SV is the number of support vectors, M is the total number of data samples, and w is the weight coefficient, which balances the classification error rate and model complexity.

[0033] Based on the above solution, preferably, step S54 includes:

[0034] S541, based on the generated initialization population X={X1, X2,…, X N}, calculate each individual X in the population i The fitness value F i ;

[0035] S542, according to the size of the fitness value, record the current optimal individual X best and its fitness F best ;

[0036] S543, for the current optimal individual X best Perform local search, obtain local candidate solutions, calculate the fitness value of the local candidate solutions, and if it is greater than F best , then the local candidate solution and its fitness value are replaced by the current optimal individual X best and its fitness F best ;

[0037] S544. For each individual X i , updates the position according to a nonlinear adaptive weight strategy;

[0038] S545, using the preset probability to identify individual X i Make mutations;

[0039] S546. Calculate the fitness value of the new individual. If it is better than the current optimal individual, update X. best and F best ;

[0040] S547. If the maximum number of iterations is reached or the accuracy requirement is met, the optimal solution X is output.best , otherwise, return to step S542 and iterate again.

[0041] Based on the above solution, preferably, in step S544, the nonlinear adaptive weight strategy includes a nonlinear adaptive weight factor ω, and the position of the hippopotamus individual is updated according to the nonlinear adaptive weight factor ω. The hippopotamus individual position update formula and the weight factor ω are expressed as follows:

[0042]

[0043]

[0044] Where T is the maximum number of iterations, t is the current number of iterations, and are two different individuals randomly selected from the population, β is the learning factor, is the individual's current location, Update the location for the individual, is the optimal individual in the current iteration.

[0045] Based on the above solution, preferably, in step S545, for individual X i The formula for mutation is as follows:

[0046]

[0047] Among them, rand is a random number between [0,1], X min and N max The value range of the variable.

[0048] Based on the above solution, preferably, in step S543, performing a local search for the current optimal individual includes:

[0049]

[0050] Where δ is a small step size and N(0,1) is the standard normal distribution.

[0051] Based on the above solution, preferably, step S2 includes:

[0052] S21. Add standard normally distributed Gaussian white noise to the original signal x(t) to construct the sequence to be decomposed x i (t):

[0053] x i (t)=x(t)+εδ i (t)

[0054] Where ε is the Gaussian white noise weight coefficient, δ i(t) is the Gaussian white noise generated during the i-th processing;

[0055] S22, treat the decomposition sequence x i (t) performs EMD decomposition, and takes the average value of the first-order modal component IMF1 obtained by decomposition as the first-order IMF component IMF1(t) of the complete empirical mode decomposition of adaptive white noise:

[0056]

[0057] r1(t)=x(t)-IMF1(t)

[0058] Where IMF1(t) represents the first modal component obtained by the complete empirical mode decomposition of adaptive white noise; r1(t) represents the residual signal after the first decomposition;

[0059] S23, after adding a specific Gaussian white noise to the j-th order residual signal obtained after decomposition, continue to perform EMD decomposition to obtain a new IMF component and residual signal:

[0060]

[0061] r j (t) = r j-1 (t)-IMF j (t)

[0062] Where, IMF j (t) represents the j-th order modal component obtained by the complete empirical mode decomposition of adaptive white noise; E j-1 () represents the j-1th order IMF component after EMD decomposition of the sequence; ε j-1 represents the weight coefficient of the noise added to the complete empirical mode of adaptive white noise at the j-1th iteration; r j (t) represents the j-th order residual signal;

[0063] S24. When the residual signal of the EMD decomposition becomes a monotonic signal or meets other preset conditions, the iteration stops, and the adaptive white noise complete empirical mode algorithm decomposition ends.

[0064] On the other hand, the present invention also provides a valve leakage diagnosis system based on an improved Hippo algorithm to optimize SVM, the system is used to perform any of the above methods, the system comprising:

[0065] Data acquisition module, used to collect acoustic emission signals under different pressure and valve opening conditions;

[0066] The data processing module is used to perform adaptive white noise complete empirical mode decomposition on the collected signal to obtain a finite number of IMF components;

[0067] The feature extraction module is used to calculate the correlation coefficient between the IMF component and the original signal, select the IMF component with a correlation coefficient greater than a preset threshold, and calculate its fuzzy entropy to construct a feature vector;

[0068] The model optimization module is used to train the SVM model and optimize the parameters of the SVM model using the improved Hippo algorithm to obtain a valve leakage diagnosis model;

[0069] The fault diagnosis module is used to classify and predict the feature vector using the valve leakage diagnosis model to realize valve leakage fault diagnosis.

[0070] The present invention has the following beneficial effects compared to the prior art:

[0071] (1) The valve leakage diagnosis method proposed in this paper, based on an improved Hippo algorithm optimized SVM, effectively addresses the difficulties in feature extraction caused by the nonlinear and non-stationary characteristics of valve leakage signals, as well as the poor classification performance caused by insufficient parameter optimization of traditional SVMs, by combining CEEMDAN signal decomposition, fuzzy entropy feature extraction, and an improved IHO-SVM classifier. This method can accurately identify valve leakage faults under different pressure and valve opening conditions, significantly improving diagnostic accuracy and stability.

[0072] (2) The present invention uses multiple chaotic maps to initialize the IHO algorithm population, effectively solving the problem of the traditional IHO algorithm being single-initialized and prone to falling into local optimality. By introducing a diverse initial population, the global search capability of the algorithm is enhanced, and the probability of finding the optimal solution is increased. This improvement enables the IHO algorithm to more comprehensively explore the parameter space when optimizing SVM parameters, avoiding performance degradation caused by improper initialization, thereby providing higher-quality SVM model parameters for subsequent valve leakage fault diagnosis;

[0073] (3) The nonlinear adaptive weight strategy and elite retention mechanism designed in this invention effectively solve the problems of poor balance between global search and local development and slow convergence speed in the traditional IHO algorithm during the iteration process. The nonlinear adaptive weight factor expands the search range in the early stage of iteration and enhances the global exploration ability; it gradually decreases in the later stage, which is conducive to local refined search. The elite retention mechanism ensures that the optimal solution is not replaced by a new solution, thereby improving the stability of the algorithm. These improvements significantly improve the convergence speed and optimization accuracy of the IHO algorithm, making it possible to find the optimal parameter combination faster and more accurately when optimizing SVM parameters;

[0074] (4) The present invention introduces a mutation operator, combined with an improved fitness function, to effectively solve the problem that the IHO algorithm is prone to falling into local optimality and the SVM model is overfitting. The mutation operation increases the diversity of the population and prevents the algorithm from converging prematurely; the improved fitness function not only considers the classification accuracy, but also the model complexity, balancing the performance and generalization ability of the model. These improvements make the optimized SVM model have stronger generalization ability and robustness, and can better adapt to the valve leakage fault diagnosis task under different working conditions and environmental noise;

[0075] (5) The present invention uses the CEEMDAN algorithm to decompose the valve leakage signal and combines it with the fuzzy entropy feature extraction method to effectively solve the problem of difficult feature extraction caused by the nonlinear and non-stationary characteristics of the valve leakage signal. The CEEMDAN algorithm overcomes the modal aliasing problem in the traditional EMD method and improves the accuracy and reliability of signal decomposition; the fuzzy entropy feature can effectively capture the complexity and irregularity of the signal. This combination not only fully utilizes the time-frequency characteristics of the valve leakage signal, but also enhances the feature's anti-noise ability, providing a more effective and robust feature input for the subsequent SVM classifier. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0077] Figure 1 is a flow chart of the method of the present invention;

[0078] Figure 2 It is a technical implementation diagram of the present invention;

[0079] Figure 3 This is a system framework diagram of the present invention. DETAILED DESCRIPTION

[0080] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0081] like Figure 1 As shown, the present invention provides a valve leakage diagnosis method based on an improved Hippo algorithm to optimize SVM, comprising:

[0082] S1. Collect acoustic emission signals under different pressure and valve opening conditions as original signals;

[0083] S2, performing adaptive white noise complete empirical mode decomposition on the collected original signal to obtain a finite number of IMF components;

[0084] S3, calculating the correlation coefficient between each IMF component and the original signal, selecting the IMF component with a correlation coefficient greater than a threshold τ, calculating the fuzzy entropy of the selected IMF component, and constructing a feature vector;

[0085] S4. Use the feature vector as a data sample and divide it into training set, validation set and test set;

[0086] S5. Use the training set to train the SVM model. According to the classification error rate of the validation set on the trained SVM model, the parameters of the SVM model are iteratively optimized based on the improved Hippo algorithm to obtain the optimal parameters and the optimized SVM model.

[0087] S6. Input the test set into the optimized SVM model, predict the valve status, calculate the classification accuracy, and evaluate the model performance. Fine-tune the model based on the evaluation results to obtain a valve leakage diagnosis model.

[0088] S7. Obtain a new acoustic emission signal, decompose and extract features from the signal, and input it into a valve leakage diagnosis model to obtain a diagnosis result.

[0089] See also Figure 2The technical route of the present invention includes: 1) collecting acoustic emission signals under different pressures and different valve opening conditions, that is, original signals; 2) using CEEMDAN algorithm to decompose the original signal to obtain a finite number of IMF components; 3) calculating the correlation coefficient between each IMF component and the original signal, that is, similarity, using Pearson correlation coefficient, setting a similarity threshold τ, and taking the IMF components greater than the threshold as selected IMF components, calculating the fuzzy entropy of the selected IMF components, and using the fuzzy entropy of these IMF components as extracted features to construct a feature vector; 4) using the feature vector as a data sample, dividing the training set, validation set, and test set according to a certain ratio; 5) selecting the SVM model as the classification model, using the radial basis function as the kernel function, and selecting the parameters to be optimized in the SVM model as the penalty parameter c and the kernel parameter g, that is, the dimension in the problem space of the optimization algorithm is set to 2, and using the improved Hippo algorithm to iteratively optimize the parameters to be optimized of the SVM model. During the optimization, each The SVM model corresponding to the group optimization parameters is trained on the training set, and then the fitness value is calculated on the validation set. The optimal individual is determined according to the descending order of fitness, and the position of the hippopotamus individual is updated according to the designed nonlinear adaptive weight factor and position update strategy to update the hippopotamus population. The optimized parameters are iteratively optimized until the iteration stop condition is reached. At this time, the optimal solution is output, and the SVM model corresponding to the optimal solution is the optimized model; 6) the test set is input into the optimized model for final tuning, the classification accuracy of the optimized model on the test set is calculated, the model performance is evaluated, and the model is fine-tuned if necessary to obtain the final SVM model, which is used as the valve leakage diagnosis model; 7) thereafter, the acoustic emission signal under the new pressure and valve opening conditions is obtained, the CEEMDAN algorithm is used to decompose the signal, and the fuzzy entropy of the selected IMF component is calculated to obtain the eigenvector of the signal, and the eigenvector is input into the valve leakage diagnosis model to obtain the diagnosis result.

[0090] Specifically, in one embodiment of the present invention, the collected acoustic emission signals are mainly fault signals, so as to facilitate the training and optimization of subsequent models.

[0091] Specifically, in one embodiment of the present invention, step S2 includes:

[0092] S21. Add standard normally distributed Gaussian white noise to the original signal x(t) to construct the sequence to be decomposed x i (t):

[0093] x i (t)=x(t)+εδ i (t)

[0094] Where ε is the Gaussian white noise weight coefficient, δ i (t) is the Gaussian white noise generated during the i-th processing;

[0095] S22, treat the decomposition sequence x i (t) performs EMD decomposition, and takes the average value of the first-order modal component IMF1 obtained by decomposition as the first-order IMF component IMF1(t) of the complete empirical mode decomposition of adaptive white noise:

[0096]

[0097] r1(t)=x(t)-IMF1(t)

[0098] Where IMF1(t) represents the first modal component obtained by the complete empirical mode decomposition of adaptive white noise; r1(t) represents the residual signal after the first decomposition;

[0099] S23, after adding a specific Gaussian white noise to the j-th order residual signal obtained after decomposition, continue to perform EMD decomposition to obtain a new IMF component and residual signal:

[0100]

[0101] r j (t) = r j-1 (t)-IMF j (t)

[0102] Where, IMF j (t) represents the j-th modal component obtained by the complete empirical mode decomposition of adaptive white noise; E j-1 () represents the j-1th order IMF component after EMD decomposition of the sequence; ε j-1 represents the weight coefficient of the noise added to the complete empirical mode of adaptive white noise at the j-1th iteration; r j (t) represents the j-th order residual signal;

[0103] S24. When the residual signal of the EMD decomposition becomes a monotonic signal or meets other preset conditions, the iteration stops, and the adaptive white noise complete empirical mode algorithm decomposition ends.

[0104] The CEEMDAN algorithm improves upon the traditional EMD (Empirical Mode Decomposition) method. By introducing adaptive white noise and multiple ensemble averaging, it effectively addresses the modal aliasing problem inherent in traditional EMD methods. CEEMDAN not only preserves the time-frequency localization of the signal but also improves the accuracy and reliability of signal decomposition. The CEEMDAN algorithm reduces modal aliasing by adding white noise of varying intensities. Multiple decompositions are averaged to reduce the impact of noise and improve decomposition stability. IMF components are gradually extracted to ensure the physical significance of each component.

[0105] Through CEEMDAN decomposition, meaningful IMF components can be extracted from complex valve leakage signals, laying the foundation for subsequent feature extraction and fault diagnosis. This method is particularly suitable for processing nonlinear and non-stationary fault signals such as valve leakage, and can effectively capture subtle changes in the signal, improving the accuracy and reliability of fault diagnosis.

[0106] Specifically, in one embodiment of the present invention, step S3 includes:

[0107] For each IMF component, calculate the Pearson correlation coefficient between it and the original signal. The calculation formula of the Pearson correlation coefficient is as follows:

[0108]

[0109] Among them, ρ X,Y is the Pearson correlation coefficient of X and Y, cov(X,Y) is the covariance of X and Y, σ X , σ Y is the standard deviation of X and Y, E represents the expected value, v X 、μ Y are the means of X and Y respectively.

[0110] Set a threshold τ and select IMF components with correlation coefficients greater than τ. This step can filter out the IMF components that are most relevant to the original signal and remove those that are irrelevant or have large noise components.

[0111] For the selected IMF components, the fuzzy entropy is calculated. Fuzzy entropy is a measure of time series complexity and is particularly effective for nonlinear and nonstationary signals. Calculating fuzzy entropy involves the following steps: a) converting the time series data into fuzzy sets; b) calculating the fuzzy similarity between each data point in the time series and other data points; and c) defining the fuzzy entropy based on the similarity of the fuzzy vectors.

[0112] Fuzzy entropy is defined as:

[0113]

[0114] In the formula, b is the embedding dimension, a is the data length, R is the similarity discrimination threshold, Φ b (R) is the proportion of template matching. The larger the fuzzy entropy value, the higher the probability of generating new patterns in the sequence, reflecting the greater the complexity of the sequence.

[0115] The calculated fuzzy entropy values are combined into a feature vector. The fuzzy entropy value of each IMF component is used as a dimension of the feature vector.

[0116] In this embodiment, fuzzy entropy is used to extract the features of the IMF components obtained after CEEMDAN decomposition. By calculating the fuzzy entropy of each IMF component, a vector that can effectively characterize the valve leakage fault characteristics can be constructed.

[0117] After obtaining the feature vector, the feature vector is used as a data sample and divided into a training set, a validation set, and a test set. The training set and the validation set are used to train and optimize the SVM model.

[0118] Support vector machines (SVMs), a powerful binary classification model, are based on constructing a linear classifier in feature space. This classifier is unique in its pursuit of margin maximization, a characteristic that significantly distinguishes it from traditional perceptron methods. To achieve broader applicability, SVMs incorporate the kernel technique. This innovation enables SVMs to transcend linear boundaries, becoming essentially nonlinear classifiers capable of effectively handling complex data distributions.

[0119] SVM uses kernel function to perform nonlinear mapping of original data to solve multi-classification problems in high-dimensional space. The present invention uses radial basis function as kernel function.

[0120] The performance of a support vector machine (SVM) depends on the choice of the penalty parameter c and the kernel parameter g. In order to improve the classification performance of SVM, it is necessary to find the optimal parameter combination (c, g).

[0121] Specifically, the present invention uses an improved Hippo algorithm to optimize the parameters of the SVM model. During the optimization process, each set of parameters is trained on the training set, and the fitness value is calculated on the validation set to iteratively optimize the SVM model. In this embodiment, step S5 includes:

[0122] S51. Select the penalty parameter c and kernel parameter g of the SVM model as target optimization parameters, and logarithmically encode the penalty parameter c and kernel parameter g into individual positions in the Hippo algorithm:

[0123] X i =(log2c i ,log2g i )

[0124] Using a logarithmic scale helps to better explore the parameter space, because the parameters of SVMs usually have better performance at different orders of magnitude.

[0125] S52, setting the population size N, the maximum number of iterations T and upper and lower boundary parameters, and initializing the population using a combined chaotic map.

[0126] This embodiment improves the mapping and combines the Logistic mapping and the Tent mapping to initialize the population, so as to increase the diversity and uniformity of the initial population.

[0127] Specifically, initializing the population using the combined chaotic map includes:

[0128] S521. Design a hybrid mapping function to determine the mapping to be used through conditional judgment. The hybrid mapping function is as follows:

[0129]

[0130] Where r is a random number between [0,1], μ is the control parameter of the Logistic mapping;

[0131] S522. Determine the dimension m of the problem space, generate an initial value x0 for each dimension, repeatedly apply the above hybrid mapping function multiple times for each dimension to generate a chaotic sequence, and take the final value as the initialization chaotic sequence value of the dimension;

[0132] S523, mapping the generated chaotic sequence value to the upper and lower bounds of the hippo population search space, the position X of each hippo individual i Expressed as: X i =(x i1 ,x i2 ,…,x im ), let z be the generated chaotic sequence value, then the initial position of the hippopotamus individual is calculated as follows:

[0133] X i :x ij =ll j +z·(ul j -ll j ), i=1,2,...,N; j=1,2,...,m

[0134] Where x ij is the position of the i-th hippopotamus in the j-th dimension, ll j and ul j are the lower and upper bounds of the j-th dimension respectively;

[0135] S524. Repeat steps S522-S523 for each hippopotamus individual and each dimension in the population until the entire population is initialized.

[0136] Specifically, m represents the dimension, and each dimension represents an independent variable in the problem space. For IHO, the position X of each hippopotamus individual is i Expressed as: X i =(x i1 ,x i2 ,…,x imDuring the initialization process, an initial value needs to be generated for each dimension. This means that for each individual hippopotamus, m initial values need to be generated, one for each dimension. In the hybrid mapping initialization method provided by the present invention, a hybrid mapping function is independently applied to each dimension to generate initial values. This ensures that a good initial distribution can be obtained in each dimension.

[0137] S53, for each parameter combination (c i ,g i ), use the training set to train the SVM model, and calculate the fitness value on the validation set.

[0138] Specifically, to avoid overfitting, the fitness function not only considers the classification accuracy, but also the complexity of the model (the number of support vectors). The fitness function is defined as:

[0139]

[0140] Where E is the classification error rate on the validation set, n SV is the number of support vectors, M is the total number of data samples, and w is the weight coefficient, which balances the classification error rate and model complexity.

[0141] S54. Iteratively update the population according to the improved Hippo algorithm. In each iteration, perform a local search for the current optimal solution. When the iteration stop condition is reached, output the optimal solution, that is, the optimal parameter combination.

[0142] The Hippopotamus algorithm (HO) is a swarm intelligence optimization algorithm inspired by the social behavior of hippos. It simulates the position updates, defense strategies, and escape strategies of hippos in rivers or ponds to find the optimal solution to the optimization problem. The complete optimization process of the Hippopotamus algorithm is as follows:

[0143] 1) Population initialization

[0144] The hippo population is mathematically represented by a matrix. Each hippo's position corresponds to a candidate solution, and its position update represents the value of the decision variable.

[0145] 2) Update of the hippopotamus's position in the river or pond (exploration phase)

[0146] Hippopotamus groups have a complex structure, consisting of females, young, multiple males, and a dominant male. The dominant male guards the territory, while the females are protected by the dominant male. Adult males may be expelled after mating, forcing them to establish their own groups, either by attracting females or by competing for dominance. The following equation describes the position of male hippopotamus within the group.

[0147]

[0148] in, represents the position of the male hippopotamus, Dhippo represents the position of the dominant hippopotamus, y1 is a random number in the range of 0-1, and I1 is an integer between 1 and 2.

[0149]

[0150] Where h1 and h2 are numbers or vectors randomly selected from the formula of h, I1 and I2 are integers between 1 and 2, and Q1 and Q2 are random integers 0 or 1. is a random vector between 0 and 1, r5, r6, r7 are random numbers between 0 and 1, MG i is the average of some randomly selected hippos.

[0151] 3) Hippopotamus defends against predators (Exploration)

[0152] 3) Hippopotamus defends against predators (Exploration)

[0153] When faced with a threat, a hippopotamus's defense mechanism includes suddenly turning to approach the predator and emitting a deafening roar as a warning signal intended to force the predator away.

[0154]

[0155] in Represents a random vector from 0 to 1. Predator represents the position of the predator, represents the distance from the i-th hippopotamus to the predator. During this period, the hippopotamus took a Factors of the body engage in defensive behavior to protect themselves from predators. Less than F i , when the predator is very close to the hippopotamus, the hippopotamus will quickly turn towards the predator and move towards it, causing it to retreat. Greater than F i , it means that the predator is relatively far away from the hippopotamus.

[0156]

[0157] in, is a random vector with Levy distribution, Levy(θ) is its calculation formula; w and v are random numbers between 0 and 1, f is a uniform random number between 2 and 4, and c is a random number between 1 and 1.5. is a random number between 2 and 3. g represents a random number between -1 and 1. is a random vector of dimension 1*m.

[0158] 4) When hippos encounter a group of predators or are unable to successfully drive away predators through defensive means, they will adopt an escape strategy and choose to leave the current area. At this time, the hippopotamus's position is updated.

[0159] This paper considers that the Hippo algorithm may not strike a good balance between global search and local exploitation, making it prone to falling into local optima, resulting in insufficient search capabilities. Convergence can be slow during iterations, particularly in complex, high-dimensional optimization problems, impacting the algorithm's efficiency. The quality of the algorithm's initial population significantly influences the final optimization results; improper initialization can lead to performance degradation.

[0160] Therefore, the aforementioned combined chaotic mapping is employed to address the initial population quality issue. Furthermore, an innovative nonlinear adaptive weight factor, ω, is designed to dynamically adjust the algorithm's exploration and exploitation balance, enhancing the comprehensiveness and depth of the search. The currently optimal individual is retained in each iteration to ensure that the optimal solution is not replaced by a new one, improving the algorithm's stability and search efficiency.

[0161] By applying this nonlinear adaptive weight factor to the Hippo position update formula, the search process can be guided more flexibly, allowing the algorithm to achieve a better balance between exploring unknown areas and refining the search for the optimal solution. This improvement not only broadens the search range but also significantly enhances the algorithm's optimization capabilities, providing a more effective and powerful tool for solving practical optimization problems. The Hippo position update formula and weight factor ω are expressed as follows:

[0162]

[0163]

[0164] Where T is the maximum number of iterations, t is the current number of iterations, and are two different individuals randomly selected from the population, β is the learning factor, is the individual’s current location, Update the location for the individual, is the optimal individual in the current iteration. In the early stages of the algorithm's iterations, the weight factor ω is designed to be large and have a steep slope. This strategy aims to rapidly expand the search range of the male hippo population, enhancing the algorithm's global exploration capabilities, thereby more broadly covering the solution space and increasing the chances of discovering potentially high-quality solutions. As the iterations progress, towards the later stages, the slope of the weight factor ω decreases, and its change gradually slows. This adjustment helps the algorithm refine its search within already explored, optimal areas, reducing blind exploration and increasing the likelihood of escaping local optima. This nonlinear weight factor strategy not only improves algorithm accuracy but also ultimately helps it find a more suitable and optimized solution location.

[0165] In addition, some individuals are randomly mutated to increase population diversity and prevent premature convergence.

[0166] The mutation operation is as follows:

[0167] With a certain probability, i Perform mutation:

[0168]

[0169] Among them, rand is a random number between [0,1], X min and X max The value range of the variable.

[0170] When using IHO to optimize SVM, we also need to consider the problem of insufficient parameter optimization accuracy, that is, the inability to fully search for the optimal parameter combination, which affects the classification performance. Therefore, in each iteration, a local search is performed on the current optimal solution to improve accuracy:

[0171]

[0172] Where δ is a small step size and N(0,1) is the standard normal distribution.

[0173] In summary, in this embodiment, the specific process of step S54 is as follows:

[0174] S541, based on the generated initialization population X={X1, X2,…, X N}, calculate each individual X in the population i The fitness value F i ;

[0175] S542, according to the size of the fitness value, record the current optimal individual X best and its fitness F best ;

[0176] S543, for the current optimal individual X best Perform local search, obtain local candidate solutions, calculate the fitness value of the local candidate solutions, and if it is greater than F best , then the local candidate solution and its fitness value are replaced by the current optimal individual X best and its fitness F best ;

[0177] S544. For each individual X i , updates the position according to a nonlinear adaptive weight strategy;

[0178] S545, using the preset probability to identify individual X i Make mutations;

[0179] S546. Calculate the fitness value of the new individual. If it is better than the current optimal individual, update X. best and F best ;

[0180] S547. If the maximum number of iterations is reached or the accuracy requirement is met, the optimal solution X is output. best , otherwise, return to step S542 and iterate again.

[0181] S55. Perform antilogarithmic transformation on the optimal parameter combination to obtain optimal parameters c and g.

[0182] Specifically, in one embodiment of the present invention, after iteratively training and optimizing the SVM model using the training set and validation set based on the improved Hippo algorithm, the optimal parameters are obtained. The model corresponding to the optimal parameters is the optimized SVM model. Subsequently, the pre-retained test set data is input into the optimized SVM model. The specific steps are as follows:

[0183] The feature vector of the test set is input into the optimized SVM model.

[0184] The model is used to predict each test sample to obtain the predicted valve state.

[0185] Compare the predictions with the true labels of the test set.

[0186] After obtaining the prediction results, multiple performance indicators need to be calculated to comprehensively evaluate the performance of the model. The main performance indicators include:

[0187] Classification accuracy: The number of correctly classified samples divided by the total number of samples. Confusion matrix: This provides a detailed breakdown of the classifications between categories, helping to identify which categories the model performs better or worse on. Precision, recall, and F1 score: These metrics provide a more detailed picture of the model's performance across categories. ROC curve and AUC value: These are used to assess the model's classification ability and robustness.

[0188] Based on the calculated performance indicators, a comprehensive evaluation of the model's overall performance is required. This evaluation includes the model's performance under different valve leakage conditions and whether the model is overfitting or underfitting. If any evaluation results are deficient, further fine-tuning can be performed. The resulting SVM model, known as the trained model, can then be retrained using all data samples to obtain a valve leakage diagnosis model. This model is packaged in a deployable format. Subsequently, new acoustic emission signals are acquired, decomposed, and feature extracted. These signals are then fed into the valve leakage diagnosis model to obtain diagnostic results.

[0189] In the present invention, the classification decision function of the SVM model is:

[0190]

[0191] Where, α i is the Lagrange multiplier, y i is the category label of the data sample, K(χ i ,χ) is the kernel function, and the present invention adopts the radial basis kernel function (RBF):

[0192]

[0193] b is the bias term.

[0194] Specifically, the present invention experimentally verifies the proposed improved Hippo algorithm optimized SVM model (IHO-SVM):

[0195] By simulating varying degrees of internal leakage in gas tank valves, acoustic emission signals were collected under multiple air pressure and valve opening conditions for subsequent classification experiments. The experimental platform is divided into two parts: the first is the gas tank valve internal leakage test platform, which integrates a compressor, gas storage tank, adjustable valve, and air pressure gauge to ensure stable and accurate simulation of various internal leakage scenarios; the second is the acoustic emission acquisition system, which includes an acoustic emission sensor, preamplifier, high-speed data acquisition card, and acoustic emission signal acquisition software. This system collects acoustic emission signals in real time and at high speed.

[0196] During the experiment, the gas tank was first pressurized using a compressor. After the pressure gauge stabilized at a preset value of 0.5 bar, the valve opening was gradually adjusted to various levels, ranging from 1 / 20 (slight internal leakage), 3 / 40 (minor internal leakage), 1 / 10 (moderate internal leakage), and 1 / 8 (severe leakage), to simulate varying degrees of internal leakage. Simultaneously, the acoustic emission acquisition system remained operational, capturing and recording the acoustic emission signals under each operating condition in real time. Subsequently, the gas tank pressure was raised to another preset value of 1 bar, and the valve opening adjustment and signal acquisition steps were repeated to fully cover internal leakage under different air pressure conditions. Accurate recording of the acoustic emission signals for each operating condition (i.e., different air pressure and valve opening combinations) was ensured. The classification of valve internal leakage faults is shown in Table 1. Each of the eight fault types in Table 1 contains 100 data sets, with 5,000 data points sampled per set.

[0197] Table 1 Fault classification and labels

[0198] Fault type Label 0.5bar slight internal leakage 0 0.5bar small amount of internal leakage 1 0.5 bar medium internal leakage 2 0.5bar serious leakage 3 1 bar slight internal leakage 4 1bar small amount of internal leakage 5 1 bar medium internal leakage 6 1 bar serious leakage 7

[0199] The CEEMDAN algorithm is used to decompose the original signal under 8 states to obtain the IMF under each state. Since the number of IMFs obtained under different states is different, the Pearson correlation coefficient between the original signal and the decomposed IMFs of each order is calculated, and the IMFs with the top 5 correlation coefficients are taken.

[0200] The fuzzy entropy of the 5th-order IMF of each fault type constitutes a 100×5 fault feature matrix and is input into the IHO-SVM model. 60% of the samples of each fault type are taken as training samples, 20% as validation sets, and 20% as test samples. The number of hippo populations is set to 20, and the maximum number of iterations is 50. At the same time, in order to verify the superiority of IHO-SVM in valve fault diagnosis and classification, it is compared with HO-SVM, PSO-SVM and GWO-SVM models. In the PSO-SVM model, c1=2 and c2=2.

[0201] The accuracy of each model obtained is as follows:

[0202] IHO-SVM accuracy: 98.74%; HO-SVM accuracy: 96.85%; PSO-SVM accuracy: 96.22%; GWO-SVM accuracy: 96.22%; traditional SVM accuracy: 91.19%.

[0203] By comparing the results of 10 independent runs, IHO-SVM shows higher accuracy and stability.

[0204] In summary, the valve leakage diagnosis method based on the improved IHO-SVM proposed in this paper demonstrates superior accuracy and stability. By combining CEEMDAN decomposition and fuzzy entropy feature extraction with an IHO-optimized SVM classifier, this method can effectively identify valve leakage at different pressures and openings, providing a reliable solution for fault diagnosis in industrial practice.

[0205] In addition, if Figure 3 As shown, the present invention also provides a valve leakage diagnosis system based on an improved Hippo algorithm to optimize SVM, the system is used to perform any of the above methods, the system comprising:

[0206] The data acquisition module is used to collect acoustic emission signals under different pressure and valve opening conditions; and perform preliminary digital processing and storage on the collected signals.

[0207] The data processing module is used to perform adaptive white noise complete empirical mode decomposition on the collected signal to obtain a finite number of IMF components to achieve more effective signal representation. The specific steps include: adding standard normally distributed Gaussian white noise to the original signal to construct the sequence to be decomposed; performing EMD decomposition on the sequence to be decomposed to obtain the first-order modal components; iteratively performing EMD decomposition, each time adding a specific Gaussian white noise to the residual signal; terminating the decomposition process when the residual signal becomes monotonic or meets other preset conditions; and ultimately obtaining a series of IMF (intrinsic mode function) components that can better represent the time-frequency characteristics of the signal.

[0208] The feature extraction module is used to calculate the correlation coefficient between the IMF component and the original signal, select the IMF component with a correlation coefficient greater than a preset threshold, and calculate its fuzzy entropy to construct a feature vector; the calculation of the fuzzy entropy takes into account the complexity and uncertainty of the signal and can better capture the nonlinear and non-stationary characteristics of the valve leakage signal.

[0209] The model optimization module is used to train the SVM model and optimize the parameters of the SVM model using the improved Hippo algorithm to obtain a valve leakage diagnosis model;

[0210] The feature vector is divided into training set, validation set and test set. The SVM model is preliminarily trained using the training set. The penalty parameter c and kernel parameter g of the SVM model are optimized using the improved Hippo algorithm (IHO). The optimization process includes: parameter encoding: logarithmically encode c and g; population initialization: initialize the population using a mixed mapping function; fitness calculation: consider the classification error rate and model complexity; population update: update the individual position using a nonlinear adaptive weight strategy; local search: perform local search on the current optimal solution to improve accuracy; mutation operation: increase population diversity to prevent premature convergence; iterative optimization until the maximum number of iterations is reached or the accuracy requirements are met. Output the optimal parameter combination to obtain the optimized SVM model. Finally, fine-tune the model using the test set to obtain the valve leakage diagnosis model.

[0211] The fault diagnosis module is used to classify and predict the feature vector using the valve leakage diagnosis model to realize valve leakage fault diagnosis.

[0212] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A valve leakage diagnosis method based on the improved Hippo algorithm and optimized SVM is characterized by: include: S1. Collect acoustic emission signals under different pressure and valve opening conditions as original signals; S2, performing adaptive white noise complete empirical mode decomposition on the collected original signal to obtain a finite number of IMF components; S3, calculating the correlation coefficient between each IMF component and the original signal, selecting the IMF component whose correlation coefficient is greater than a preset threshold, calculating the fuzzy entropy of the selected IMF component, and constructing a feature vector; S4. Use the feature vector as a data sample and divide it into training set, validation set and test set; S5. Use the training set to train the SVM model. According to the classification error rate of the validation set on the trained SVM model, the parameters of the SVM model are iteratively optimized based on the improved Hippo algorithm to obtain the optimal parameters and the optimized SVM model. Step S5 includes: S51, selecting the penalty parameter c and the kernel parameter g of the SVM model as target optimization parameters, and logarithmically encoding the penalty parameter c and the kernel parameter g into individual positions in the Hippo algorithm; S52, setting the population size N, the maximum number of iterations T and upper and lower boundary parameters, and initializing the population using a combined chaotic map; S53, for each parameter combination (c i ,g i ), use the training set to train the SVM model and calculate the fitness value on the validation set; S54, iteratively updating the population according to the improved Hippo algorithm, performing a local search for the current optimal solution in each iteration, and outputting the optimal solution, i.e., the optimal parameter combination, when the iteration stop condition is reached; S55, performing antilogarithmic transformation on the optimal parameter combination to obtain optimal parameters c and g; Step S54 includes: S541, based on the generated initialization population X={X1, X2,…, X N }, calculate each individual X in the population i The fitness value F i ; S542, according to the size of the fitness value, record the current optimal individual X best and its fitness F best ; S543, for the current optimal individual X best Perform local search, obtain local candidate solutions, calculate the fitness value of the local candidate solutions, and if it is greater than F best , then the local candidate solution and its fitness value are replaced by the current optimal individual X best and its fitness F best ; S544. For each individual X i , updates the position according to a nonlinear adaptive weight strategy; S545, using the preset probability to identify individual X i Make mutations; S546. Calculate the fitness value of the new individual. If it is better than the current optimal individual, update X. best and F best ; S547. If the maximum number of iterations is reached or the accuracy requirement is met, the optimal solution X is output. best Otherwise, return to step S542 and iterate again; In step S544, the nonlinear adaptive weight strategy includes a nonlinear adaptive weight factor ω. The position of the hippopotamus individual is updated according to the nonlinear adaptive weight factor ω. The hippopotamus individual position update formula and the weight factor ω are expressed as follows: Where T is the maximum number of iterations, t is the current number of iterations, and are two different individuals randomly selected from the population, β is the learning factor, is the individual's current location, Update the location for the individual, is the optimal individual in the current iteration; In step S545, for individual X i The formula for mutation is as follows: Among them, rand is a random number between [0,1], X min and X max The value range of the variable; S6. Input the test set into the optimized SVM model, predict the valve status, calculate the classification accuracy, and evaluate the model performance. Fine-tune the model based on the evaluation results to obtain a valve leakage diagnosis model. S7. Obtain a new acoustic emission signal, decompose and extract features from the signal, and input it into a valve leakage diagnosis model to obtain a diagnosis result.

2. The valve leakage diagnosis method based on the improved Hippo algorithm optimized SVM according to claim 1, characterized in that: In step S52, the combined chaotic map includes a Logistic map and a Tent map, and initializing the population using the combined chaotic map includes: S521. Design a hybrid mapping function to determine the mapping to be used through conditional judgment. The hybrid mapping function is as follows: Where r is a random number between [0,1], μ is the control parameter of the Logistic mapping; S522. Determine the dimension m of the problem space, generate an initial value x0 for each dimension, repeatedly apply the above hybrid mapping function multiple times for each dimension to generate a chaotic sequence, and take the final value as the initialization chaotic sequence value of the dimension; S523, mapping the generated chaotic sequence value to the upper and lower bounds of the hippo population search space, the position X of each hippo individual i Expressed as: X i =(x i1 ,x i2 ,…,x im ), let z be the generated chaotic sequence value, then the initial position of the hippopotamus individual is calculated as follows: X i :x ij =ll j +z·(ul j -ll j ),i=1,2,…,N;j=1,2,…,m Where x ij is the position of the i-th hippopotamus in the j-th dimension, ll j and ul j are the lower and upper bounds of the j-th dimension respectively; S524. Repeat steps S522-S523 for each hippopotamus individual and each dimension in the population until the entire population is initialized.

3. The valve leakage diagnosis method based on the improved Hippo algorithm optimized SVM according to claim 1, characterized in that: The calculation formula of fitness value is: Where E is the classification error rate on the validation set, n SV is the number of support vectors, M is the total number of data samples, and w is the weight coefficient, which balances the classification error rate and model complexity.

4. The valve leakage diagnosis method based on the improved Hippo algorithm optimized SVM according to claim 1, characterized in that: In step S543, performing a local search for the current optimal individual includes: Where δ is a small step size and N(0,1) is the standard normal distribution.

5. The valve leakage diagnosis method based on the improved Hippo algorithm optimized SVM according to claim 1, characterized in that: Step S2 includes: S21. Add standard normally distributed Gaussian white noise to the original signal x(t) to construct the sequence to be decomposed x i (t): x i (t)=x(t)+εδ i (t) Where ε is the Gaussian white noise weight coefficient, δ i (t) is the Gaussian white noise generated during the i-th processing; S22, treat the decomposition sequence x i (t) performs EMD decomposition, and takes the average value of the first-order modal component IMF1 obtained by decomposition as the first-order IMF component IMF1(t) of the complete empirical mode decomposition of adaptive white noise: r1(t)=x(t)-IMF1(t) Where IMF1(t) represents the first modal component obtained by the complete empirical mode decomposition of adaptive white noise; r1(t) represents the residual signal after the first decomposition; S23, after adding a specific Gaussian white noise to the j-th order residual signal obtained after decomposition, continue to perform EMD decomposition to obtain a new IMF component and residual signal: r j (t)=r j-1 (t)-IMF j (t) Where, IMF j (t) represents the j-th modal component obtained by the complete empirical mode decomposition of adaptive white noise; E j-1 () represents the j-1th order IMF component after EMD decomposition of the sequence; ε j-1 represents the weight coefficient of the noise added to the complete empirical mode of adaptive white noise at the j-1th iteration; r j (t) represents the j-th order residual signal; S24. When the residual signal of the EMD decomposition becomes a monotonic signal or meets other preset conditions, the iteration stops, and the adaptive white noise complete empirical mode algorithm decomposition ends.

6. The valve leakage diagnosis system based on the improved Hippo algorithm to optimize SVM is characterized by: The system is used to perform the method according to any one of claims 1 to 5, and the system includes: Data acquisition module, used to collect acoustic emission signals under different pressure and valve opening conditions; The data processing module is used to perform adaptive white noise complete empirical mode decomposition on the collected signal to obtain a finite number of IMF components; The feature extraction module is used to calculate the correlation coefficient between the IMF component and the original signal, select the IMF component with a correlation coefficient greater than a preset threshold, and calculate its fuzzy entropy to construct a feature vector; The model optimization module is used to train the SVM model and optimize the parameters of the SVM model using the improved Hippo algorithm to obtain a valve leakage diagnosis model; The fault diagnosis module is used to classify and predict the feature vector using the valve leakage diagnosis model to realize valve leakage fault diagnosis.

Images