Storage tank bottom plate corrosion acoustic emission detection noise identification method based on SSA-CatBoost model

Abstract

The invention provides a storage tank bottom plate corrosion acoustic emission detection noise recognition method based on an SSA-CatBoost model. The storage tank bottom plate corrosion acoustic emission detection noise recognition method comprises the following steps that 1, acoustic emission signal data are collected; 2, collecting pure corrosion signals and original waveform signals of different noise sources by an acoustic emission detection system; 3, extracting time domain and frequency domain noise from the noise-reduced acoustic emission waveform signal to form multi-dimensional feature data; 4, finally, before acoustic emission characteristic evaluation, an SSA-CatBoost model is used for acoustic emission source recognition, non-corrosion signals are removed, and then bottom plate state grading judgment is carried out; by applying the technical scheme, the reliability of corrosion state evaluation of the storage tank bottom plate can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of acoustic emission detection, in particular to a method for identifying noise in acoustic emission detection of corrosion of a storage tank bottom plate based on an SSA-CatBoost model. Background Art

[0002] Storage tanks are important infrastructure in the petrochemical field. Since the media stored in them are mostly toxic, flammable, and explosive substances, once corrosion perforation and leakage occur, it will cause incalculable heavy losses. As the most vulnerable part to corrosion and difficult to detect, the bottom plate of the storage tank must be strictly controlled. As a new type of passive on-line detection method, acoustic emission detection has great economic and time advantages such as simple process and no need to open the tank compared with traditional non-destructive testing methods such as magnetic particle, magnetic flux leakage, and ultrasonic testing. Therefore, it has received extensive attention from the industrial and academic circles.

[0003] The grading evaluation of the corrosion state of the storage tank bottom plate based on acoustic emission detection technology mainly depends on the density distribution of time difference positioning points or the number of impacts of each independent channel within the effective detection time. However, due to the complexity of the on-site environment, acoustic emission detection operations are often interfered by noises such as mechanical impacts, strong electromagnetic interference, and condensate dripping. These noises will also be detected by the acoustic emission system and form impacts or positioning points, thus affecting the detection results. At the same time, there are also noise components in the effective corrosion signals, which interfere with the feature extraction effect and classification effect of the acoustic emission source. Therefore, how to reduce the interference of noise is an important problem faced by acoustic emission detection of storage tank corrosion. To eliminate noise interference, generally in production, the detection threshold is increased to filter out low-amplitude noise signals, or a guard sensor is introduced to eliminate the acoustic emission signal source in a specific direction. Although these hardware filtering and signal acquisition technologies can effectively eliminate the interference of specific noises, it is difficult to cope with such a complex detection environment as the corrosion detection of the oil tank bottom plate.

[0004] At present, the research on the noise reduction method for acoustic emission detection signals mainly focuses on eliminating the noise components in the acoustic emission signals and improving the signal-to-noise ratio of the acoustic emission signals. There is no relevant noise reduction method for acoustic emission detection of storage tank bottom plates based on machine learning sound source recognition. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method for identifying noise in acoustic emission detection of corrosion of a storage tank bottom plate based on an SSA-CatBoost model, so as to improve the reliability of the corrosion state evaluation of the storage tank bottom plate.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A method for identifying noise in acoustic emission detection of corrosion of storage tank bottom plates based on the SSA-CatBoost model, comprising the following steps: Step 1: Conduct a sensitivity test on the acoustic emission detection equipment and a noise test on the ambient noise. After determining the hardware parameters of the acoustic emission detection through the sensitivity test results and the ambient noise detection results, collect the acoustic emission signal data;

[0007] Step 2: For the pure corrosion signals and the original waveform signals of different noise sources collected by the acoustic emission detection system, use the established acoustic emission signal denoising model to perform denoising processing on the officially collected acoustic emission data;

[0008] Step 3: Extract time-domain and frequency-domain noise from the denoised acoustic emission waveform signals to form multi-dimensional feature data;

[0009] Step 4: Randomly divide the feature data into a training set and a test set in a ratio of 1:1. The training set is used for the training of the SSA-CatBoost model; the test set is used to verify whether the performance of the model meets the standard. Finally, before the acoustic emission feature evaluation, use the SSA-CatBoost model to identify the acoustic emission source, eliminate non-corrosion signals, and then conduct the bottom plate state grading judgment.

[0010] In a preferred embodiment, in Step 2, the wavelet threshold method is used to denoise the acoustic emission signal; a simulated signal and white noise are mixed to form a noisy acoustic emission signal; the performance of the denoising algorithm is optimized by adjusting the wavelet basis function, the number of wavelet decomposition layers, and the wavelet threshold adjustment parameters.

[0011] In a preferred embodiment, Step 2 specifically includes the following steps:

[0012] Step 21: For the noisy emission signal iD1(t)∈L 2 (R) , select a wavelet basis function to perform orthogonal wavelet transform on it to obtain the wavelet coefficients of the signal cj,k :

[0013]

[0014] where a0 = 2, b0 = 1; the wavelet basis function Adopt the wavelet threshold method to denoise the acoustic emission signal;

[0015] Step 22: Use the soft threshold for threshold adjustment, and its formula is as follows:

[0016]

[0017] where: x0 is the wavelet threshold,cj,k is the wavelet coefficient after wavelet transform, is the wavelet coefficient after threshold selection;

[0018] Step 23: Re-select the threshold of the wavelet coefficient and obtain the new wavelet coefficient

[0019] Step 24: Use the inverse wavelet transform to reconstruct the new wavelet coefficient and recover the effective signal

[0020]

[0021] Step 25: Signal-to-noise ratio SNR and root mean square error RMSE are used as evaluation indicators

[0022]

[0023] In a preferred embodiment, the calculation formula for the target statistical value of the samples in the SSA-CatBoost model in step 4 is:

[0024]

[0025] Where: x j , y i are both training samples; D k is the data set before the k-th sample; a is the weight coefficient, and P is the prior value.

[0026] In a preferred embodiment, in step 4, the SSA optimization algorithm is used in the SSA-CatBoost model to optimize the six key parameters of the learning rate, the number of trees, the depth of the trees, the L2 regularization term, the minimum number of samples for splitting, and the maximum number of features of CatBoost.

[0027] In a preferred embodiment, the use of the SSA-CatBoost model for acoustic emission source identification in step 4 specifically includes: First, according to the number of individuals N of the salp swarm, a decision variable number D is defined to define an N×D dimensional space; the position of each salp individual is represented by X i =[X i1 ,X i2 ,...,X iD , i = 1, 2, 3,..., N; determine the initial position of the population according to the value range of the decision variables:

[0028] X N×D = rand(N, D)×(Ub - Lb)+Lb

[0029] where $U_b = [u_{b1}, u_{b2},..., u_{b D $ and $L_b = [l_{b1}, l_{b2},..., l_{b D $ represent the upper and lower bounds of each decision variable range;

[0030] Secondly, update the position of the leader; the position of the leader is related to the position of the target, while maintaining a certain degree of randomness to ensure that it can lead the followers towards the target position; in a population, the position of the leader in each dimension is represented by and the followers are represented by $n = 2, 3,..., N$; the expression for updating the leader's position is:

[0031]

[0032] where: $F d $ represents the target position in the $d$-th dimension; $c_1$, $c_2$ and $c_3$ are the control parameters of the salp swarm algorithm, $c_1 = 2e -(4×l / lmax) $ is the convergence factor, which is affected by the iteration number $l$ and is used to balance the convergence speed of the iteration process; $c_2$ and $c_3$ are random numbers in the interval $[0, 1]$ and are used to determine the update direction and step size of the leader's position;

[0033] Then, update the positions of the followers based on the position of the leader; in practical applications, according to Newton's laws of motion, update the positions of the followers based on velocity and acceleration

[0034]

[0035] where: $t$ represents the iteration number, and the acceleration $a$ can be expressed as $a = (v f - v_0) / t$; $v_0$ represents the initial velocity, and the initial velocity of the followers is 0 at the beginning of each iteration;

[0036] Since the position of the followers only depends on the position of the previous salp, therefore, the velocity $v f $ is expressed as:

[0037]

[0038] Therefore, the expression for updating the position of the followers is:

[0039]

[0040] where $n \geq 2$, the position corresponding to the $n$-th salp in the $d$-th dimension.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1) The method of the present invention extracts multi-domain features composed of time domain and frequency domain, effectively compensating for the problem of insufficient extraction of traditional acoustic emission parameter pattern features, and improving the ability of acoustic emission signals to describe acoustic emission source information.

[0043] 2) The present invention introduces the SSA algorithm to optimize the CatBoost acoustic emission source recognition model. The BES algorithm has strong global search ability and good convergence, and can quickly find the optimal parameter combination of CatBoost, greatly reducing the time cost and labor cost of training the SVM model.

[0044] 3) The present invention uses the SSA algorithm to optimize the CatBoost acoustic emission source recognition model, separates effective signals from environmental noise, can effectively reduce the influence of environmental factors on the grading evaluation of the corrosion state of storage tanks, and improve the reliability of acoustic emission detection. Description of the Drawings

[0045] Figure 1 It is a schematic diagram of the overall process of the preferred embodiment of the present invention.

[0046] Figure 2 It is a schematic diagram of the SSA-CatBoost model process of the preferred embodiment of the present invention.

[0047] Figure 3 It is a method for denoising acoustic emission signals based on wavelet threshold of the preferred embodiment of the present invention.

[0048] Figure 4 It is a schematic diagram of the simulation experiment platform for acoustic emission sources of the bottom plate corrosion of storage tanks of the preferred embodiment of the present invention.

[0049] Figure 5 It is a schematic diagram of the recognition result of acoustic emission sources of the preferred embodiment of the present invention. Detailed Embodiments

[0050] The following further describes the present invention in conjunction with the drawings and embodiments.

[0051] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present application belongs.

[0052] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0053] Method for identifying noise in acoustic emission detection of corrosion of storage tank bottom plate based on SSA-CatBoost model, referring to Figures 1-5 , in the acoustic emission detection process of the storage tank bottom plate in the solution of the present invention, compared with the traditional acoustic emission detection process, a module for denoising acoustic emission signals and identifying and eliminating noise signals is added. First, by introducing wavelet threshold-based denoising, the signal-to-noise ratio of the acoustic emission signals is improved, so that the extracted features are more distinguishable. Subsequently, by introducing the SSA-CatBoost sound source identification model, the problem that the traditional acoustic emission detection process is vulnerable to environmental noise interference is solved. The complete technical solution of the acoustic emission source noise identification model for corrosion of storage tank bottom plate based on the SSA-CatBoost algorithm is as Figure 1 shown

[0054] Specifically, first, before the formal detection, the sensitivity of the acoustic emission detection equipment is tested and the environmental noise is tested. The hardware parameters of the acoustic emission detection are determined through the sensitivity test results and the environmental noise detection results. This is the same as the traditional acoustic emission detection method here. After ensuring that the hardware settings are correct, the formal acquisition of acoustic emission signal data is carried out.

[0055] Secondly, for the pure corrosion signals and the original waveform signals of different noise sources collected by the acoustic emission detection system, the established acoustic emission signal denoising model is used to denoise the formally collected acoustic emission data.

[0056] Subsequently, time-domain and frequency-domain noises are extracted from the denoised acoustic emission waveform signals to form multi-dimensional feature data. The names and extraction formulas of the time-domain and frequency-domain features are shown in Table 1 and Table 2.

[0057] Table 1 Time-domain feature parameters

[0058]

[0059]

[0060] Note: x(i), i = 1, 2,..., N is the discrete acoustic emission waveform signal, and N represents the sampling length of the acoustic emission signal x(i).

[0061] Table 2 Frequency-domain feature parameters

[0062]

[0063] Note: p i , f i respectively represent the frequency-domain amplitude and frequency corresponding to the signal data of the sample signal x(i) after Fourier transform.

[0064] Finally, the feature data is randomly divided into a training set and a test set at a ratio of 1:1. The training set is used for the training of the SSA-CatBoost model. The test set is used to verify whether the performance of the model meets the standard. Finally, before the acoustic emission feature evaluation, the acoustic emission source identification model is used to identify the acoustic emission source, and the non-corrosion signals are removed, and then the bottom plate state grading judgment is carried out.

[0065] Specifically, to avoid numerical problems caused by different parameter distribution ranges, the original feature data is normalized before feature selection, and the variable is transformed into the range of [-1,1]. The normalization process is shown in the following formula:

[0066]

[0067] In the formula, x n represents the normalized input variable, x max and x min represent the maximum and minimum values of the input variable respectively.

[0068] The present invention uses Catboost as the classification model for the acoustic emission source of the storage tank. CatBoost is an ensemble learning algorithm based on gradient boosting decision trees, which has excellent accuracy and high computational speed when dealing with categorical data, and optimizes the processing ability of large-scale data sets. Compared with other GBDT algorithms such as XGBoost and LightGBM, CatBoost has a greater advantage in classification accuracy and better performance when dealing with feature categories. Compared with traditional gradient boosting, CatBoost processes the training data sequence to ensure that the model will not cause bias due to the data sequence during each round of training, that is, the ordered target statistic. The ordered target statistic is a key technology for processing categorical features. It introduces a randomly arranged order for the target statistic to ensure that the target statistic value of each sample only depends on historical samples, reducing the impact of target leakage. The calculation formula for the target statistic value of the sample is:

[0069]

[0070] In the formula: x j 、 y i are both training samples; D k is the data set before the kth sample; a is the weight coefficient, and P is the prior value.

[0071] CatBoost has numerous hyperparameters. Using the trial-and-error method requires a large amount of time cost and it is difficult to obtain the optimal parameter combination. Therefore, the present invention introduces the SSA optimization algorithm to optimize six key parameters of CatBoost, namely the learning rate, the number of trees, the depth of the tree, the L2 regularization term, the minimum number of samples for splitting, and the maximum number of features, in order to obtain the optimal acoustic emission source identification model.

[0072] First, according to the number of individuals N in the salp swarm population, define an N×D dimensional space representing the number of decision variables D (number of Decision variables). In the present invention, D is 6. The position of each salp individual is represented by X i =[X i1 ,X i2 ,...,X iD , where i = 1, 2, 3,..., N. Determine the initial position of the population according to the numerical range of the decision variables:

[0073] X N×D =rand(N,D)×(Ub - Lb)+Lb (2)

[0074] In the formula, Ub = [ub1, ub2,..., ub D and Lb = [lb1, lb2,..., lb D represent the upper and lower bounds of the range of each decision variable.

[0075] Secondly, update the position of the leader. The position of the leader is related to the position of the target, and at the same time maintains a certain degree of randomness to ensure that it can lead the followers to move forward along the target position. In a population, the position of the leader in each dimension is represented by represented by, and the followers are represented by n = 2, 3,..., N. The expression for updating the position of the leader is:

[0076]

[0077] In the formula: F d represents the target position in the d-th dimension. c1, c2, and c3 are the control parameters of the salp algorithm. c1 = 2e -(4×l / lmax) is the convergence factor, which is affected by the number of iterations l and is used to balance the convergence speed of the iterative process. c2 and c3 are random numbers in the interval [0, 1] and are used to determine the update direction and step size of the leader's position.

[0078] Then, update the positions of the followers based on the position of the leader. In practical applications, according to Newton's laws of motion, update the positions of the followers based on velocity and acceleration

[0079]

[0080] In the formula: t represents the number of iterations, and a is the acceleration, which can be expressed as a = (v f - v0) / t. v0 represents the initial velocity, and at the beginning of each iteration, the velocity follower's initial velocity is 0.

[0081] Since the position of the follower only depends on the position of the previous salp, therefore, the velocity v f can be expressed as:

[0082]

[0083] Therefore, the position update formula of the follower is:

[0084]

[0085] where n ≥ 2, the position corresponding to the nth salp in the dth dimension.

[0086] The present invention uses the wavelet threshold method to reduce the noise of the acoustic emission signal. By utilizing the decorrelation of data in wavelet transform, the energy of the acoustic emission signal is mainly concentrated in the larger wavelet coefficients, while the energy of the noise is relatively small. Therefore, by setting an appropriate threshold, the signal coefficients can be retained and most of the noise can be removed. First, a noisy acoustic emission signal is formed by mixing a simulated signal and white noise. Then, by adjusting parameters such as the wavelet basis function, the number of wavelet decomposition levels, and the wavelet threshold adjustment, the performance of the denoising algorithm is optimized.

[0087] 1) For the noisy emission signal i D1 (t) ∈ L 2 (R), select a wavelet basis function to perform orthogonal wavelet transform on it to obtain the wavelet coefficients c j,k :

[0088]

[0089] In the formula, a0 = 2, b0 = 1; the wavelet basis function The present invention uses the wavelet threshold method to reduce the noise of the acoustic emission signal, by utilizing the decorrelation of data in wavelet transform.

[0090] 2) Threshold adjustment method, the present invention uses the soft threshold, and its formula is as follows:

[0091]

[0092] In the formula: x0 is the wavelet threshold, c j,k is the wavelet coefficient after wavelet transform, is the wavelet coefficient after threshold selection.

[0093] 2) Threshold selection: Re-select the threshold of the wavelet coefficients to obtain new wavelet coefficients.

[0094] 3) Use the inverse wavelet transform for the new wavelet coefficients to perform reconstruction and recover the effective signal.

[0095]

[0096] 4) To compare the noise reduction effects of different wavelet denoising parameters, the signal-to-noise ratio (SNR) and the root mean square error (RMSE) are introduced as evaluation indicators.

[0097]

[0098] In this embodiment, the data is from the corrosion acoustic emission detection experimental platform of the storage tank bottom plate. The experimental platform is as Figure 4 shown. The common on-site noises in the acoustic emission detection of the storage tank are classified and experimentally simulated: start the circulation pump to run idly to simulate the interference generated by the operation of on-site mechanical equipment; start the circulation pump to make the medium in the tank circulate in the pipeline to simulate fluid disturbance; rub the tank wall with sandpaper to simulate the friction noise between mechanical structures; turn on the shower to wash the tank wall to simulate the interference caused by rainy weather; gently tap the tank wall with fingers to simulate the interference caused by accidental touch of the operator or the stay of birds touching the tank body; open the bottom needle valve of the tank to simulate the interference caused by the leakage of the medium in the tank.

[0099] First, set the acoustic emission detection hardware device according to the sensitivity test and the noise test. According to the experimental results, the hardware parameters are set as shown in Table 3, and 75 groups of acoustic emission signals from seven different sources, namely bottom corrosion, mechanical vibration, fluid disturbance, friction noise, rain interference, tank body impact, and bottom leakage of the tank, are collected, a total of 525 groups of data.

[0100] Secondly, use the above wavelet threshold method to denoise the acoustic emission signals. In this example, use s(t)=A(t)e jθ(t) to construct a simulated acoustic emission signal and add white noise to form a noisy simulated signal. Calculate the noise reduction effects of different set parameters. It is found that when using the soft threshold method, the wavelet basis is Sym7, and the decomposition level is 5, the denoising effect on the simulated acoustic emission signal is the best, and its SNR and RMSE reach 13.285 and 0.0357 respectively. According to this result, denoise the seven different acoustic emission source signals collected, and extract 14 time-domain features and 6 frequency-domain features from the denoised waveform signals to form 20 time-frequency domain features.

[0101] Finally, the trained SSA-CatBoost classification model is used to identify the collected data, and the obtained recognition results are used to eliminate the noise signals.

[0102] Table 3 Settings of AE detection hardware parameters

[0103]

[0104] Recognition results and performance comparison: The recognition results of the corresponding AE sources on the storage tank bottom plate are as Figure 5 shown. It can be seen that the SSA-SVM classification model proposed in the present invention has good AE source recognition ability for the storage tank bottom plate, and the recognition accuracy rate on the test set reaches 90.10%. The number of iterations of each model is set to 20, 30, 40, and 50 times respectively. At the same time, in order to avoid the influence of accidental errors, 10 repeated experiments are carried out on the sample data of the AE signals of the storage tank. The average value and standard deviation of the test set accuracy rates of 10 repeated experiments of each model are shown in Table 4. When the number of iterations is 40, the average accuracy rate of the SSA-CatBoost model reaches the highest 91.85%. At the same time, the standard deviation of 10 repeated experiments of SSA-CatBoost is 0.33%, indicating that BES has strong global search ability, and the SSA-CatBoost model has better stability and can well identify the types of AE sources on the storage tank bottom plate.

[0105] Table 4

[0106]

[0107] When the AE data set collected by AE is mixed with pure corrosion and various types of noise signals, the noise signals can be identified and eliminated by using the model of the present invention, so as to achieve the purpose of reducing environmental noise interference. The remaining data is used for the corrosion grading evaluation of the storage tank bottom plate.

Claims

1. An acoustic emission detection noise recognition method for the corrosion of the storage tank bottom plate based on the SSA-CatBoost model, characterized in that, It includes the following steps: Step 1: Conduct a sensitivity test on the acoustic emission detection device and a noise test on the ambient noise. After determining the hardware parameters of the acoustic emission detection based on the sensitivity test results and the ambient noise detection results, collect the acoustic emission signal data; Step 2: For the pure corrosion signals and the original waveform signals of different noise sources collected by the acoustic emission detection system, use the established acoustic emission signal denoising model to perform denoising processing on the officially collected acoustic emission data; Step 3: Extract the time-domain and frequency-domain noises from the denoised acoustic emission waveform signals to form multi-dimensional feature data; Step 4: Randomly divide the feature data into a training set and a test set at a ratio of 1:

1. The training set is used for the training of the SSA-CatBoost model; the test set is used to verify whether the performance of the model meets the standard. Finally, before the acoustic emission feature evaluation, use the SSA-CatBoost model to identify the acoustic emission source, eliminate the non-corrosion signals, and then conduct the floor state grading judgment.

2. The method for identifying noise in acoustic emission detection of corrosion of storage tank bottom plates based on the SSA-CatBoost model according to claim 1, wherein In Step 2, the wavelet threshold method is used for the denoising of the acoustic emission signal; a simulated signal and white noise are mixed to form a noisy acoustic emission signal; the performance of the denoising algorithm is optimized by adjusting the wavelet basis function, the number of wavelet decomposition layers, and the wavelet threshold adjustment parameters.

3. The method for identifying noise in acoustic emission detection of corrosion of storage tank bottom plates based on the SSA-CatBoost model according to claim 1, characterized in that Step 2 specifically includes the following steps: Step 21: For the noisy emission signal i D1 (t) ∈ L 2 (R), select a wavelet basis function to perform orthogonal wavelet transform on it to obtain the wavelet coefficients c j,k : where a0 = 2, b0 = 1; wavelet basis function Wavelet threshold method is used to reduce the noise of acoustic emission signals; Step 22: Use the soft threshold for threshold adjustment, and its formula is as follows: Where: \(x_0\) is the wavelet threshold, \(c\) j,k is the wavelet coefficient after wavelet transform, and \(c'\) is the wavelet coefficient after threshold selection; Step 23: Re-select the threshold of the wavelet coefficients and obtain new wavelet coefficients Step 24: Reconstruct the new wavelet coefficients using the inverse wavelet transform to recover the effective signal ​ Step 25: The signal-to-noise ratio SNR and the root mean square error RMSE are used as evaluation indicators 4. The method for identifying noise in acoustic emission detection of corrosion of storage tank bottom plates based on the SSA-CatBoost model according to claim 1, wherein, In Step 4, the calculation formula for the target statistical value of the samples in the SSA-CatBoost model is: Wherein: x j , y i are both training samples; D k is the data set before the k-th sample; a is the weight coefficient, and P is the prior value.

5. The method for identifying the noise in the acoustic emission detection of the corrosion of the storage tank bottom plate based on the SSA-CatBoost model according to claim 1, characterized in that, In Step 4, in the SSA-CatBoost model, the SSA optimization algorithm is used to optimize the six key parameters of the CatBoost, namely the learning rate, the number of trees, the depth of the tree, the L2 regularization term, the minimum number of samples for splitting, and the maximum number of features.

6. The method for identifying the noise in the acoustic emission detection of the corrosion of the storage tank bottom plate based on the SSA-CatBoost model according to claim 5, wherein In step 4, the SSA-CatBoost model is used for acoustic emission source identification, which specifically includes: First, according to the number of individuals N in the salp population, a dimensional space of N×D is defined, representing the number of decision variables D; the position of each salp individual is represented by X i =[X i1 , X i2 ,..., X iD , where i = 1, 2, 3,..., N; the initial position of the population is determined according to the value range of the decision variables: X N×D = rand(N,D) × (Ub - Lb) + Lb where Ub = [ub1, ub2,..., ub D and Lb = [lb1, lb2,..., lb D represent the upper and lower bounds of each decision variable range; Secondly, update the position of the leader; the position of the leader is related to the position of the target, while maintaining a certain degree of randomness to ensure that it can lead the followers towards the target position; in a population, the position of the leader in each dimension is represented by and the followers are represented by ; the position update expression of the leader is: Where: F d represents the target position of the d-th dimension; c1, c2, and c3 are the control parameters of the salp swarm algorithm, c1 = 2e -(4×l / lmax) is the convergence factor, which is affected by the iteration number l and is used to balance the convergence speed of the iteration process; c2 and c3 are random numbers in the interval [0, 1] and are used to determine the update direction and step size of the leader position; Then, update the positions of the followers based on the position of the leader; in actual applications, update the positions of the followers based on the velocity and acceleration according to Newton's laws of motion where: t represents the number of iterations, and a is the acceleration, which can be expressed as a = (v f - v0) / t; v0 represents the initial velocity, and at the beginning of each iteration, the velocity of the follower is 0 for both initial velocities. Since the position of the follower depends only on the position of the previous salp, the velocity v f is expressed as: Therefore, the position update formula of the follower is: where n ≥ 2, The position corresponding to the nth salp in the dth dimension.

Images