Complex Structure Impact Location Method and System Based on Hybrid Kernel Extreme Learning Machine

Abstract

The present invention relates to a complex structure impact positioning method and system based on a hybrid kernel extreme learning machine. The method includes: constructing a sample data set; the sample data set includes multiple groups of sample data and corresponding label information, and the sample data is the arrival time of a shock wave; then constructing a hybrid kernel extreme learning machine model, and the hidden layer feature mapping of the hybrid kernel extreme learning machine model is a hybrid kernel function, and the hybrid kernel function includes a global kernel function and a local kernel function; training the hybrid kernel extreme learning machine model with the sample data set to obtain a complex structure impact positioning model; and finally, performing impact positioning on the complex structure corresponding to the sample data to be measured through the complex structure impact positioning model to determine the impact position. The present invention can improve the impact positioning accuracy of complex structures.

Description

Technical Field

[0001] The present invention relates to the technical field of structural health monitoring, and particularly to a method for shock location of complex structures based on a hybrid kernel extreme learning machine. Background Art

[0002] During flight, an aircraft is extremely vulnerable to impacts from foreign debris and some parts falling off the surface of the carrier fuselage, especially high-speed impacts during flight, which can cause damage or even destruction to the aircraft. These damages may seriously degrade the mechanical properties of the structure, significantly reduce the load-bearing capacity of the structure, and pose a major hidden danger to the safe flight of the aircraft. Therefore, detecting the impact location is the key to ensuring its safety and reliability.

[0003] With the rapid development of research in modern manufacturing technology, mechanics theory, and material technology, especially the development of the current aerospace and military industries, traditional simple structural components can no longer meet the actual needs, and higher requirements are put forward for the shape and performance of modern structural components. Such structural components usually have complex shapes, and traditional shock location methods perform poorly in such complex structures, mainly manifested in: the propagation characteristics of guided waves in such complex structures are complex, and the wave arrival times extracted by time-frequency signal processing methods such as wavelet transform are inaccurate. Therefore, the positioning algorithms based on time difference have large errors; in addition, it is difficult to obtain an accurate mechanical model of the actual engineering structure. Therefore, the method based on system modeling is difficult to apply to the actual structure. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for shock location of complex structures based on a hybrid kernel extreme learning machine, which can improve the shock location accuracy of complex structures.

[0005] To achieve the above purpose, the present invention provides the following solutions:

[0006] The present invention provides a method for shock location of complex structures based on a hybrid kernel extreme learning machine, and the method includes:

[0007] Construct a sample data set; the sample data set includes multiple groups of sample data and corresponding label information; the sample data is the arrival time of the shock wave; the arrival time of the shock wave is extracted from the shock test data by using the AIC function; the label information is the shock location; the shock location is obtained from the shock test data;

[0008] Construct a hybrid kernel extreme learning machine model; the implicit layer feature mapping of the hybrid kernel extreme learning machine model is a hybrid kernel function, and the hybrid kernel function includes a global kernel function and a local kernel function;

[0009] Train the hybrid kernel extreme learning machine model by using the sample data set to obtain a complex structure shock location model;

[0010] Use a complex structure impact location model to perform impact location on the complex structure corresponding to the sample data to be measured, so as to determine the impact position.

[0011] Optionally, training the hybrid kernel extreme learning machine model using the sample data set specifically includes:

[0012] In each round of training, use the bootstrap sampling method to extract multiple groups of sample data from the sample data set, and form a sample data subset with the multiple groups of sample data. One sample data subset is extracted in each round of training;

[0013] Use each of the sample data subsets to train the hybrid kernel extreme learning machine model respectively to obtain multiple complex structure impact location models;

[0014] Perform arithmetic averaging on the output results of the multiple complex structure impact location models to obtain the final impact location result.

[0015] Optionally, the expression of the AIC function is:

[0016] AIC(t) = t log(var(R(1,t))) + (T - t - 1) log(var(R(1 + t,T)));

[0017] Where, T represents the length of the time series corresponding to the impact signal, R(1,t) represents the impact signal segment intercepted from 1 to t, R(1 + t,T) represents the impact signal segment intercepted from (1 + t) to T, and var represents the variance function.

[0018] Optionally, the global kernel function is a polynomial kernel function, and the expression of the polynomial kernel function is:

[0019] K P (x,x i ) = (m(x * x i ) + n) d ;

[0020] The local kernel function is a radial basis kernel function; the expression of the radial basis kernel function is:

[0021] K R (x,x i ) = exp(-||x - x i || 2 / 2σ 2 );

[0022] The expression of the hybrid kernel function is:

[0023] K(x,y) = ρK P(x, y) + (1 - ρ)K R (x, y);

[0024] where K P is the global polynomial kernel function, x i is the input vector; m, n, and d are all polynomial kernel parameters, K R is the local radial basis kernel function, σ is the radial basis kernel parameter, K is the mixed kernel function, and ρ represents the weight of the kernel matrix of the polynomial kernel function.

[0025] Optionally, the expression of the target output F(x) of the complex structure impact location model of the hybrid kernel extreme learning machine is:

[0026]

[0027] where K(x, x n ) is the mixed kernel function, C is the regularization coefficient, Ω ELM is the kernel matrix corresponding to the kernel function, and L is the expected output value of the hidden layer of the model.

[0028] Optionally, when training the hybrid kernel extreme learning machine model with the sample data set, an immune particle swarm optimization algorithm is used to optimize the hyperparameters of the hybrid kernel extreme learning machine model; the hyperparameters include the radial basis kernel parameter σ, the polynomial kernel parameters m and n, the weight ρ of the kernel matrix of the polynomial kernel function, and the regularization coefficient C.

[0029] To achieve the above object, the present invention also provides a complex structure impact location system based on a hybrid kernel extreme learning machine, the system includes:

[0030] A data set construction unit for constructing a sample data set; the sample data set includes multiple groups of sample data and corresponding label information; the sample data is the arrival time of the shock wave; the arrival time of the shock wave is extracted from the shock test data by using the AIC function; the label information is the impact location; the impact location is obtained from the shock test data;

[0031] A hybrid kernel extreme learning machine model construction unit for constructing a hybrid kernel extreme learning machine model; the hidden layer feature mapping of the hybrid kernel extreme learning machine model is a hybrid kernel function, and the hybrid kernel function includes a global kernel function and a local kernel function;

[0032] A complex structure impact location model determination unit for training the hybrid kernel extreme learning machine model with the sample data set to obtain a complex structure impact location model;

[0033] An impact location determination unit for performing impact location on the complex structure corresponding to the sample data to be measured through the complex structure impact location model to determine the impact location.

[0034] Optionally, the complex structure impact location model determination unit specifically includes:

[0035] A sample data subset acquisition module, which is used to extract multiple groups of sample data from the sample data set by the bootstrap sampling method in each round of training, and form a sample data subset with the multiple groups of sample data. One sample data subset is extracted in each round of training;

[0036] A complex structure impact location model determination module, which is used to train the hybrid kernel extreme learning machine model with each of the sample data subsets to obtain multiple complex structure impact location models;

[0037] An impact location result determination module, which is used to perform arithmetic averaging on the output results of the multiple complex structure impact location models to obtain the final impact location result.

[0038] Optionally, the expression of the AIC function is:

[0039] AIC(t) = t log(var(R(1,t))) + (T - t - 1) log(var(R(1 + t,T)));

[0040] Where, T represents the length of the time series corresponding to the impact signal, R(1,t) represents the impact signal segment intercepted from 1 to t, R(1 + t,T) represents the impact signal segment intercepted from (1 + t) to T, and var represents the variance function.

[0041] Optionally, the global kernel function is a polynomial kernel function, and the expression of the polynomial kernel function is:

[0042] K P (x,x i ) = (m(x * x i ) + n) d ;

[0043] The local kernel function is a radial basis kernel function; the expression of the radial basis kernel function is:

[0044] K R (x,x i ) = exp(-||x - x i || 2 / 2σ 2 )

[0045] The expression of the hybrid kernel function is:

[0046] K(x,y) = ρK P (x,y) + (1 - ρ)K R (x,y);

[0047] Among them, K P is the global polynomial kernel function, and x i is the input vector; m, n, and d are all polynomial kernel parameters, and K R is the local radial basis kernel function, σ is the radial basis kernel parameter, K is the mixed kernel function, and ρ represents the weight of the kernel matrix of the polynomial kernel function.

[0048] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention:

[0049] The present invention provides a complex structure impact location method and system based on a mixed kernel extreme learning machine. The method includes: constructing a sample data set; the sample data set includes multiple groups of sample data and corresponding label information, and the sample data is the arrival time of the shock wave; constructing a mixed kernel extreme learning machine model, and the hidden layer feature mapping of the mixed kernel extreme learning machine model is a mixed kernel function, and the mixed kernel function includes a global kernel function and a local kernel function; training the mixed kernel extreme learning machine model with the sample data set to obtain a complex structure impact location model; and finally, performing impact location on the complex structure corresponding to the sample data to be measured through the complex structure impact location model to determine the impact position. The present invention can improve the impact location accuracy of complex structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0051] Figure 1 is the flowchart of the complex structure impact location method based on the mixed kernel extreme learning machine of the present invention;

[0052] Figure 2 is the flowchart of the immune particle swarm optimization algorithm for optimizing the mixed kernel extreme learning machine of the present invention;

[0053] Figure 3 is the structural schematic diagram of the complex structure impact location system based on the mixed kernel extreme learning machine of the present invention;

[0054] Figure 4 is the schematic diagram of the specimen size, sensor position, and impact test points in the specific embodiment of the present invention;

[0055] Figure 5 is the comparison diagram of the impact signal and the AIC transformed signal in the specific embodiment of the present invention;

[0056] Figure 6 This is a comparison graph of the correlation coefficients of the shock wave arrival time and distance extracted by the AIC and continuous wavelet transform in the specific embodiment of the present invention;

[0057] Figure 7 This is the fitness evolution curve of the immune particle swarm optimization algorithm in the specific embodiment of the present invention;

[0058] Figure 8 This is a schematic diagram of the actual impact position and the predicted position in the specific embodiment of the present invention;

[0059] Figure 9 This is a comparison graph of the predicted average errors of the test set under multiple different types of models in the specific embodiment of the present invention.

[0060] Symbol description:

[0061] Data set construction unit 1, hybrid kernel extreme learning machine model construction unit 2, complex structure impact location model determination unit 3, and impact location determination unit 4. Specific implementation manners

[0062] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0063] The purpose of the present invention is to provide a complex structure impact location method based on a hybrid kernel extreme learning machine, which can improve the impact location accuracy of complex structures.

[0064] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0065] As Figure 1 shown, the present invention provides a complex structure impact location method based on a hybrid kernel extreme learning machine, and the method includes the following steps:

[0066] S1: Construct a sample data set; the sample data set includes multiple groups of sample data and corresponding label information; the sample data is the arrival time of the shock wave; the arrival time of the shock wave is extracted from the shock test data by using the Akaike information criterion (AIC) function; the label information is the impact location; the impact location is obtained from the shock test data.

[0067] S2: Construct a hybrid kernel extreme learning machine model; the hidden layer feature mapping of the hybrid kernel extreme learning machine model is a hybrid kernel function, and the hybrid kernel function includes a global kernel function and a local kernel function. Combining the kernel function with the extreme learning machine (ELM) and replacing the feature mapping of the ELM hidden layer with the kernel function can divide the data more precisely.

[0068] S3: Use the sample data set to train the hybrid kernel extreme learning machine model to obtain a complex structure impact location model.

[0069] S4: Locate the impact on the complex structure corresponding to the sample data to be measured through the complex structure impact location model to determine the impact position.

[0070] Furthermore, in step S1, an impact test is performed on the structure surface to construct a training set, a validation set, and a test set. The training set and the validation set are used to train and select parameters, and the test set is used to test the prediction and generalization performance of the model. The arrival time of the shock wave is selected as the input feature of the impact, and the AIC function is used to extract the arrival time of the shock wave. AIC indicates that a time series can be divided into locally stationary segments, and each segment can be modeled as an autoregressive process. The impact signal can be divided into a time series. The signal before the impact is a noise segment, and the signal after the impact is an impact oscillation signal. The point where the AIC is minimized is the separation point of these two time series, which is the arrival time of the shock wave. The expression of the AIC function is:

[0071] AIC(t) = t log(var(R(1,t))) + (T - t - 1) log(var(R(1 + t,T)));

[0072] where T represents the length of the time series corresponding to the impact signal, R(1,t) represents the impact signal segment intercepted from 1 to t, R(1 + t,T) represents the impact signal segment intercepted from (1 + t) to T, var represents the variance function, t is the independent variable, and its range is 1 to T. The t value corresponding to the minimum AIC represents the moment when the shock wave arrives. Among them, the definition of the variance function var is as follows:

[0073]

[0074] where N represents the length of the signal sequence, represents the average value of the signal in the time series.

[0075] Kernel functions can be divided into global kernels and local kernels. Global kernel functions have strong generalization ability, and local kernel functions have strong learning ability. The hybrid kernel function formed by combining the two kernel functions has both strong learning ability and good generalization ability. The present invention combines the global polynomial (Poly) kernel function and the local radial basis function (RBF) kernel function.

[0076] Further, in step S2, the global kernel function is a polynomial kernel function, and the expression of the polynomial kernel function is:

[0077] K P (x,x i )=(m(x*x i )+n) d ;

[0078] The local kernel function is a radial basis kernel function; the expression of the radial basis kernel function is:

[0079] K R (x,x i )=exp(-||x-x i || 2 / 2σ 2 );

[0080] The expression of the hybrid kernel function is:

[0081] K(x,y)=ρK P (x,y)+(1-ρ)K R (x,y);

[0082] Where, K P is the global polynomial kernel function, x i is the input vector; m, n, and d are all polynomial kernel parameters, K R is the local radial basis kernel function, σ is the radial basis kernel parameter, K is the hybrid kernel function, and ρ represents the weight of the kernel matrix of the polynomial kernel function. d affects the mapping ability of the kernel function. The larger d is, the higher the mapping dimension is, and the stronger the learning ability is. According to the empirical value, d is taken as 2.

[0083] Introduce the regularization coefficient C and the identity matrix I in ELM, then the least squares solution of the output weight is

[0084]

[0085] Introduce the kernel function, and the corresponding kernel matrix Ω ELM is

[0086] Ω ELM =HH F =h(x i )h(x j )=K(xi , x j );

[0087] In the formula, β is the output weight, h(x) is the output function of the hidden layer, H is the output matrix of the hidden layer, F is the target output of the model, C is the regularization coefficient, L is the expected output of the model, x i , x j is the test input vector, and K is the kernel function.

[0088] Furthermore, the expression of the target output F(x) of the complex structure impact location model of the hybrid kernel extreme learning machine is:

[0089]

[0090] where K(x, x n ) is the hybrid kernel function, C is the regularization coefficient, Ω ELM is the kernel matrix corresponding to the kernel function, and L is the expected output value of the model.

[0091] The initial parameters of the hybrid kernel extreme learning machine have a great influence on the performance of the model. Furthermore, in step S3, when training the hybrid kernel extreme learning machine model with the sample data set, the immune particle swarm optimization (PSO) algorithm is used to optimize the hyperparameters of the hybrid kernel extreme learning machine model; the hyperparameters include the radial basis kernel parameter σ, the polynomial kernel parameters m and n, the weight ρ of the kernel matrix of the polynomial kernel function, and the regularization coefficient C, where d takes 2.

[0092] As Figure 2 shown, using the immune particle swarm optimization algorithm to optimize the hyperparameters of the hybrid kernel extreme learning machine model specifically includes:

[0093] (1) Set σ, m, n, ρ, and C as particles, and randomly initialize the positions and velocities of the particles.

[0094] (2) Calculate the output weight matrix of the hidden layer.

[0095] (3) Calculate the output of the hidden layer nodes, use the mean square error mse as the fitness function, and calculate the individual extreme value and the global extreme value.

[0096] (4) Update the particle positions and velocities, and the update formulas are as follows:

[0097]

[0098]

[0099] Among them, k is the number of iterations, w is the inertia factor, c1 and c2 are learning factors, r1 and r2 are random numbers between 0 and 1, pbest is the individual optimal value, and gbest is the global optimal value.

[0100] (5) Calculate the concentration and excitation degree of the particles, and select several particles with a larger excitation degree for immune treatment; the concentration characterizes the quality of the diversity of the particle population, and a high concentration means that a large number of similar populations exist; the excitation degree is calculated from the fitness and the concentration, and a part of the particles with a large excitation degree are selected for immune treatment (only mutation treatment is performed here), and random numbers are generated to replace this part of the particles. Specifically, the calculation formula for the excitation degree MSLL of the particles is as follows:

[0101] MSLL = α * FIT + β * ND;

[0102] Among them, ND represents the particle concentration, and FIT represents the particle fitness.

[0103] (6) Calculate the fitness value of the updated particles, update the individual extreme value and the global extreme value. If the maximum number of iterations is not satisfied, return to step (6), otherwise execute step (7).

[0104] (7) Save the particles corresponding to the optimal fitness of the population, that is, the optimal mixed kernel function parameters, substitute the parameters into the mixed kernel extreme learning machine model, and test with the test set.

[0105] Furthermore, the Bagging algorithm is used to improve the accuracy and stability of the model and avoid overfitting. In step S3, the training of the mixed kernel extreme learning machine model using the sample data set specifically includes:

[0106] In each round of training, multiple groups of sample data are drawn from the sample data set using the bootstrap sampling method (sampling with replacement), and multiple groups of sample data form a sample data subset. One sample data subset is drawn in each round of training, and the multiple sample data subsets are independent of each other.

[0107] Each of the sample data subsets is used to train the mixed kernel extreme learning machine model to obtain multiple complex structure impact location models.

[0108] Arithmetic means are taken for the output results of the multiple complex structure impact location models to obtain the final impact location result.

[0109] Furthermore, it also includes step S5: comparing the prediction performances of several algorithms such as BP, ELM, HKELM, IPSO - HKELM, and the ensemble learning (EL) based on IPSO - HKELM, and comparing the average impact location errors of the test set.

[0110] To achieve the above - mentioned purpose, asFigure 3 As shown in the figure, the present invention also provides a complex structure impact location system based on a hybrid kernel extreme learning machine. The system includes: a dataset construction unit 1, a hybrid kernel extreme learning machine model construction unit 2, a complex structure impact location model determination unit 3, and an impact location determination unit 4.

[0111] The dataset construction unit 1 is used to construct a sample dataset. The sample dataset includes multiple groups of sample data and corresponding label information. The sample data is the arrival time of the shock wave. The arrival time of the shock wave is extracted from the shock test data using the AIC function. The label information is the impact location, which is obtained from the shock test data.

[0112] The hybrid kernel extreme learning machine model construction unit 2 is used to construct a hybrid kernel extreme learning machine model. The implicit layer feature mapping of the hybrid kernel extreme learning machine model is a hybrid kernel function, and the hybrid kernel function includes a global kernel function and a local kernel function.

[0113] The complex structure impact location model determination unit 3 is used to train the hybrid kernel extreme learning machine model using the sample dataset to obtain a complex structure impact location model.

[0114] The impact location determination unit 4 is used to perform impact location on the complex structure corresponding to the sample data to be measured through the complex structure impact location model to determine the impact location.

[0115] Furthermore, the complex structure impact location model determination unit 3 specifically includes:

[0116] The sample data subset acquisition module is used to extract multiple groups of sample data from the sample dataset using the bootstrap sampling method in each round of training, and form a sample data subset with the multiple groups of sample data. A sample data subset is extracted in each round of training.

[0117] The complex structure impact location model determination module is used to train the hybrid kernel extreme learning machine model using each of the sample data subsets to obtain multiple complex structure impact location models.

[0118] The impact location result determination module is used to perform arithmetic averaging on the output results of the multiple complex structure impact location models to obtain the final impact location result.

[0119] Furthermore, the expression of the AIC function is:

[0120] AIC(t) = t log(var(R(1,t))) + (T - t - 1) log(var(R(1 + t,T)));

[0121] Wherein, T represents the length of the time series corresponding to the impact signal, R(1,t) represents the impact signal segment intercepted from 1 to t, R(1 + t,T) represents the impact signal segment intercepted from (1 + t) to T, and var represents the variance function.

[0122] Further, the global kernel function is a polynomial kernel function, and the expression of the polynomial kernel function is:

[0123] K P (x,x i )=(m(x*x i )+n) d ;

[0124] The local kernel function is a radial basis kernel function; the expression of the radial basis kernel function is:

[0125] K R (x,x i )=exp(-||x - x i || 2 / 2σ 2 );

[0126] The expression of the hybrid kernel function is:

[0127] K(x,y)=ρK P (x,y)+(1 - ρ)K R (x,y);

[0128] Wherein, K P is the global polynomial kernel function, x i is the input vector; m, n, and d are all polynomial kernel parameters, K R is the local radial basis kernel function, σ is the radial basis kernel parameter, K is the hybrid kernel function, and ρ represents the weight of the kernel matrix of the polynomial kernel function.

[0129] Next, in conjunction with the accompanying drawings, taking a variable - thickness curved - surface thermal protection structure as an example, the technical solution of the present invention will be described in detail.

[0130] Taking the variable - thickness curved - surface thermal protection structure as an example, 16 piezoelectric sensors are arranged on the back of the specimen, 90 points are selected on the surface of the specimen for impact tests, and each position is impacted 3 times, resulting in a total of 270 groups of impact test data. In addition, impact tests are carried out at another 9 different positions, and the obtained data is used as the test set. The specimen size, sensor positions, and impact test points are as Figure 4 shown. The signal sampling rate is 1 MHz, and the number of sampling points is 5000 points.

[0131] The operation process mainly includes the following steps:

[0132] Step 1: Read the impact signal and use the AIC function to extract the arrival time of the shock wave, as Figure 5 shown.

[0133] Step 2: Compare the correlation between the arrival time of the shock wave extracted by the AIC function and the continuous wavelet transform (CWT) and the distance from the impact to the sensor. As Figure 6 shown, there is a significant correlation between the arrival time of the wave extracted by the AIC function and the distance. Therefore, it is reasonable to use the arrival time of the wave extracted by AIC to predict the impact position.

[0134] Step 3: Use the arrival time of the shock wave as the network input and the impact position as the network output to establish a sample set.

[0135] Step 4: Use the bootstrap sampling method to extract the sample set for 8 rounds, establish 8 sample subsets, and each sample subset includes 230 groups of training data and 40 groups of validation data.

[0136] Step 5: Combine the polynomial (Poly) kernel function and the radial basis function (RBF) kernel function to obtain a hybrid kernel function, and combine it with the extreme learning machine (ELM) to obtain a hybrid kernel extreme learning machine (HKELM). Input the data of the 8 sample subsets into the 8 base models.

[0137] Step 6: Use the immune particle swarm optimization algorithm to optimize 5 hyperparameters of the base model. The fitness evolution curve of the algorithm is as Figure 7 shown. Substitute the test data to obtain the positioning results of the 8 base models, perform arithmetic averaging on the results to obtain the final predicted output, calculate the error between the predicted value and the actual value, and the actual impact position and the predicted position are as Figure 8 shown.

[0138] Step 7: Compare the prediction errors of several algorithms such as BP, ELM, HKELM, IPSO-HKELM, and IPSO-HKELM(EL), as Figure 9 shown.

[0139] Technical effects of the present invention:

[0140] Aiming at the difficulty of impact positioning based on guided waves in complex structures, the present invention provides an impact positioning method and system based on a hybrid kernel extreme learning machine (IPSO-HKELM). By mining the impact response characteristic data of complex structures, a high-precision impact position prediction model is constructed. It has the following advantages:

[0141] 1) Compared with the arrival time of the wave extracted by the time-frequency analysis method, the AIC extracts the shock wave with a higher correlation between the arrival time and the distance, and the effect is better.

[0142] 2) Compared with other network models, the error of the prediction result of IPSO-HKELM is the smallest.

[0143] 3) Combining with ensemble learning further improves the prediction accuracy of the model and avoids the phenomenon of overfitting of the model.

[0144] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For the relevant parts, reference can be made to the description in the method part.

[0145] In this article, specific examples are used to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention. At the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A complex structure impact location method based on a hybrid kernel extreme learning machine, characterized in that The method includes: Constructing a sample data set; the sample data set includes multiple groups of sample data and corresponding label information; the sample data is the arrival time of the shock wave; the arrival time of the shock wave is extracted from the shock test data using the AIC function; the label information is the shock position; the shock position is obtained from the shock test data; Constructing a hybrid kernel extreme learning machine model; the hidden layer feature mapping of the hybrid kernel extreme learning machine model is a hybrid kernel function, and the hybrid kernel function includes a global kernel function and a local kernel function; Training the hybrid kernel extreme learning machine model using the sample data set to obtain a complex structure shock location model; Performing shock location on the complex structure corresponding to the sample data to be measured through the complex structure shock location model to determine the shock position.

2. A complex structure impact location method based on a hybrid kernel extreme learning machine according to claim 1, characterized in that The training of the hybrid kernel extreme learning machine model using the sample data set specifically includes: In each round of training, multiple groups of sample data are extracted from the sample data set using the bootstrap sampling method, and the multiple groups of sample data form a sample data subset, and a sample data subset is extracted in each round of training; Training the hybrid kernel extreme learning machine model using each of the sample data subsets to obtain multiple complex structure shock location models; Performing arithmetic averaging on the output results of the multiple complex structure shock location models to obtain the final shock location result.

3. A complex structure impact location method based on a hybrid kernel extreme learning machine according to claim 1, characterized in that, The expression of the AIC function is: AIC(t) = tlog(var(R(1,t))) + (T - t - 1)log(var(R(1 + t,T))); where, T represents the length of the time series corresponding to the shock signal, R(1,t) represents the shock signal segment intercepted from 1 to t, R(1 + t,T) represents the shock signal segment intercepted from (1 + t) to T, t is the independent variable, and var represents the variance function.

4. A complex structure impact location method based on a hybrid kernel extreme learning machine according to claim 1, characterized in that The global kernel function is a polynomial kernel function, and the expression of the polynomial kernel function is: K P (x, x i ) = (m(x * x i ) + n) d ; The local kernel function is a radial basis kernel function; the expression of the radial basis kernel function is: K R (x, x i ) = exp(-||x - x i ||² / 2σ 2 ); The expression of the hybrid kernel function is: K(x,y) = ρK P (x,y) + (1 - ρ)K R (x,y); Among them, K P is the global polynomial kernel function, and x i is the input vector; m, n, and d are all polynomial kernel parameters, and K R is the local radial basis kernel function, σ is the radial basis kernel parameter, K is the mixed kernel function, and ρ represents the weight of the kernel matrix of the polynomial kernel function.

5. A complex structure impact location method based on a hybrid kernel extreme learning machine according to claim 4, characterized in that The expression of the target output F(x) of the complex structure shock location model is: Among them, K(x, x n ) is the mixed kernel function, I is the identity matrix, C is the regularization coefficient, Ω ELM is the kernel matrix corresponding to the kernel function, and L is the expected output value of the model.

6. A complex structure impact location method based on a hybrid kernel extreme learning machine according to claim 5, characterized in that When training the hybrid kernel extreme learning machine model using the sample data set, an immune particle swarm optimization algorithm is used to optimize the hyperparameters of the hybrid kernel extreme learning machine model; the hyperparameters include the radial basis kernel parameter σ, the polynomial kernel parameters m and n, the weight ρ of the kernel matrix of the polynomial kernel function, and the regularization coefficient C.

7. A complex structure impact location system based on a hybrid kernel extreme learning machine, characterized in that The system includes: A data set construction unit for constructing a sample data set; the sample data set includes multiple groups of sample data and corresponding label information; the sample data is the arrival time of the shock wave; the arrival time of the shock wave is extracted from the shock test data using the AIC function; the label information is the shock position; the shock position is obtained from the shock test data; A hybrid kernel extreme learning machine model construction unit for constructing a hybrid kernel extreme learning machine model; the hidden layer feature mapping of the hybrid kernel extreme learning machine model is a hybrid kernel function, and the hybrid kernel function includes a global kernel function and a local kernel function; A complex structure impact location model determination unit is used to train the hybrid kernel extreme learning machine model with the sample data set to obtain a complex structure impact location model; An impact location determination unit is used to perform impact location on the complex structure corresponding to the sample data to be measured through the complex structure impact location model to determine the impact location.

8. A complex structure impact location system based on a hybrid kernel extreme learning machine according to claim 7, characterized in that The complex structure impact location model determination unit specifically includes: A sample data subset acquisition module is used to extract multiple groups of sample data from the sample data set by the bootstrap sampling method in each round of training, and form a sample data subset with the multiple groups of sample data. One sample data subset is extracted in each round of training; A complex structure impact location model determination module is used to train the hybrid kernel extreme learning machine model with each of the sample data subsets to obtain multiple complex structure impact location models; An impact location result determination module is used to perform arithmetic averaging on the output results of the multiple complex structure impact location models to obtain the final impact location result.

9. A complex structure impact location system based on a hybrid kernel extreme learning machine according to claim 7, characterized in that The expression of the AIC function is: AIC(t) = t log(var(R(1,t))) + (T - t - 1) log(var(R(1 + t,T))); Where, T represents the length of the time series corresponding to the impact signal, R(1,t) represents the impact signal segment intercepted from 1 to t, R(1 + t,T) represents the impact signal segment intercepted from (1 + t) to T, t is the independent variable, and var represents the variance function.

10. A complex structure impact location system based on a hybrid kernel extreme learning machine according to claim 7, characterized in that, The global kernel function is a polynomial kernel function, and the expression of the polynomial kernel function is: K P (x, x i ) = (m(x * x i ) + n) d ; The local kernel function is a radial basis kernel function; the expression of the radial basis kernel function is: K R (x,x i ) = exp(-||x - x i || 2 / 2σ 2 ); The expression of the hybrid kernel function is: K(x,y) = ρK P (x,y) + (1 - ρ)K R (x,y); Among them, K P is the global polynomial kernel function, and x i is the input vector; m, n, and d are all polynomial kernel parameters, and K R is the local radial basis kernel function, σ is the radial basis kernel parameter, K is the mixed kernel function, and ρ represents the weight of the kernel matrix of the polynomial kernel function.

Images