Composite material structure reliability analysis method and system based on adaptive hybrid agent model

Abstract

The invention discloses a composite material structure reliability analysis method and system based on an adaptive hybrid agent model, and the method comprises the steps: defining a reliability problem, determining the distribution type and statistical characteristics of basic random variables, and generating an initial design experiment; using the current DOE to train a Kriging model and an artificial neural network model, and constructing an agent model; calculating weights of different agent models on candidate sample points based on an entropy weight method, and realizing adaptive integration of the agent models; a candidate sample set is generated by adopting an MMHDR mechanism, and updating points are searched through a learning function so as to optimize the proxy model; judging whether the accuracy of the integrated model meets the requirement or not, if so, stopping updating, otherwise, continuing to optimize the agent model; outputting a failure probability based on the integrated proxy model; according to the method, by combining the advantages of different agent models and utilizing EWM and MMHDR strategies, the accuracy, efficiency and robustness of reliability analysis of the composite material structure are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of composite material structure reliability analysis, and more specifically, to a method and system for composite material structure reliability analysis based on an adaptive hybrid surrogate model. Background Art

[0002] At present, composite materials play an important role in different engineering problems due to their high strength, high specific modulus, high temperature resistance, good thermal stability, strong designability, simple manufacturing process and low cost.

[0003] However, the mechanical properties of composite material structures are affected by multiple parameters, including ply angle, ply thickness, elastic modulus, shear modulus, Poisson's ratio and density, etc.; in addition, due to manufacturing defects, lack of experience and structural complexity, the mechanical properties of composite material structures in large-scale industrial production often show significant variability; at the same time, the random changes in external load conditions and environmental factors also make it particularly important to evaluate the uncertainty of the mechanical properties of composite material structures.

[0004] However, traditional reliability analysis methods, such as using finite element simulation to evaluate the limit state, have high computational costs; in order to improve the computational efficiency of composite material structure reliability analysis, developing more efficient surrogate models has become the research focus.

[0005] In recent years, some studies have used a single surrogate model or combined different surrogate models to construct an integrated model for time-consuming limit state evaluation; however, these methods often construct the integrated model by calculating the weights of local prediction errors, ignoring the generalization ability of the surrogate model on different dimensional types of data.

[0006] Therefore, how to improve the computational efficiency and accuracy of composite material structure reliability analysis methods is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a method and system for composite material structure reliability analysis based on an adaptive hybrid surrogate model to solve the technical problems mentioned in the background art.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] A method for composite material structure reliability analysis based on an adaptive hybrid surrogate model, comprising the following steps:

[0010] S1. Determine the distribution type and statistical characteristics of the basic random variables of the composite material structure, set the number of sample points of the initial design of experiments (DOE), and calculate the actual limit state function values of the initial DOE;

[0011] S2. Use the current DOE to train a surrogate model including a Kriging model and an artificial neural network ANN model, generate a Monte Carlo candidate sample set MC according to the statistical characteristics of random variables, and use different surrogate models to predict the sample points in the MC set;

[0012] S3. Based on the EWM entropy weight method, combine the prediction errors and entropy values of the surrogate models, calculate the weights of different surrogate models at the candidate sample points, realize the adaptive integration of the surrogate models, and thus obtain a combined model;

[0013] S4. Adopt the MMHDR mechanism to generate a conditional sample point set based on the Monte Carlo candidate sample set MC, search and update points through a learning function, and add the updated points to the initial DOE;

[0014] S5. Use the updated DOE to retrain the surrogate model, and according to the preset threshold, judge whether the accuracy of the current integrated model meets the requirements. Repeat steps S3 to S5 until the convergence criterion is met or the maximum number of iterations is reached;

[0015] S6. After the iteration ends, output the failure probability of the composite material structure based on the integrated surrogate model.

[0016] Preferably, step S1 further includes analyzing the composition, function, and working conditions of the structure according to the data information of the composite material structure to be evaluated, and determining the failure mode of the structure and the corresponding limit state equation.

[0017] Preferably, in step S2, when constructing the surrogate model, use cross-validation and the jackknife technique to calculate the predicted values and prediction variances of different surrogate models.

[0018] Preferably, the specific content of step S3 is:

[0019] Calculate the entropy weight based on the EWM entropy weight method, specifically:

[0020]

[0021] where E j is the entropy of the jth surrogate model, p ij is the proportion of the ith sample point in the jth surrogate model, k is the number of candidate sample points, and N M is the total number of surrogate models;

[0022] Calculate the coefficient of variation G j =1 - E j , and calculate the entropy weight factor based on the coefficient of variation

[0023] Calculate the weights of different surrogate models at the candidate sample points as:

[0024]

[0025] Among them, e j (x) is the prediction error of the j-th surrogate model at the candidate sample points.

[0026] Preferably, in step S4, the MMHDR mechanism is used to generate a conditional sample point set based on the Monte Carlo candidate sample set MC, and the specific content of searching for updated points through the learning function is as follows:

[0027] S41. Randomly select half of the sample points N 1 in the current MC population and their predicted values as the initial subset, and sort the predicted values in ascending order;

[0028] S42. Select the first P 0 N 1 sample points as the seeds for generating the next subset. P 0 is the seed selection ratio. Use the MMHDR strategy to generate N 1 conditional sample point sets with the probability density function f(x) based on the selected seeds;

[0029] S43. Repeat this process until the stop condition is met. Combine the first penultimate subset and the second penultimate subset to search for the local range L e ;

[0030] S44. Use the U function to filter the sample points within the local range L e and obtain the updated points, and add the updated points to the DOE.

[0031] Preferably, in step S5, the judgment condition for whether the accuracy of the current integrated model meets the requirements is:

[0032]

[0033] Among them, P f,j is the failure probability estimated by the j-th surrogate model, P f is the failure probability calculated by the integrated model, and ε t is the preset threshold;

[0034] If the condition is met, stop the update; otherwise, return to step S2 to continue updating the surrogate model.

[0035] A composite material structure reliability analysis system based on an adaptive hybrid surrogate model, based on the described composite material structure reliability analysis method based on an adaptive hybrid surrogate model, includes: a reliability problem definition module, an initial surrogate model construction module, a weight calculation and model integration module, a model optimization module, and a reliability result output module;

[0036] A reliability problem definition module, which is used to determine the distribution types and statistical characteristics of the basic random variables of the composite material structure, set the number of sample points of the initial design of experiments (DOE), and calculate the actual limit state function values of the initial DOE;

[0037] An initial surrogate model construction module, which is used to train a surrogate model including a Kriging model and an artificial neural network (ANN) model using the current DOE, generate a Monte Carlo candidate sample set (MC) according to the statistical characteristics of the random variables, and predict the sample points in the MC set using different surrogate models;

[0038] A weight calculation and model integration module, which is used to calculate the weights of different surrogate models on the candidate sample points based on the EWM entropy weight method, combining the prediction errors and entropy values of the surrogate models, to achieve the adaptive integration of the surrogate models, thereby obtaining a combined model;

[0039] A model optimization module, which is used to generate a conditional sample point set based on the Monte Carlo candidate sample set (MC) using the MMHDR mechanism, search for updated points through a learning function, and add the updated points to the initial DOE; retrain the surrogate model using the updated DOE, and judge whether the accuracy of the current integrated model meets the requirements according to a preset threshold until the convergence criterion is met or the maximum number of iterations is reached;

[0040] A reliability result output module, which is used to output the failure probability of the composite material structure based on the integrated surrogate model after the iteration ends

[0041] A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the described reliability analysis method for a composite material structure based on an adaptive hybrid surrogate model.

[0042] A processing terminal, including a memory and a processor, where a computer program that can run on the processor is stored in the memory, and when the processor executes the computer program, it implements the described reliability analysis method for a composite material structure based on an adaptive hybrid surrogate model.

[0043] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and system for reliability analysis of composite material structures based on an adaptive hybrid surrogate model. By combining the advantages of different surrogate models, the integrated surrogate model can more accurately predict the limit state function, improving the accuracy of reliability analysis; adopting a local learning strategy and an MMHDR mechanism effectively reduces the size of the candidate sample set, reduces the computational cost, and improves the computational efficiency; adopting an EWM strategy takes into account the overall stability of the surrogate model, enabling the integrated surrogate model to exhibit stronger generalization ability on different dimensional types of data and enhancing the robustness of reliability analysis; the present invention is particularly suitable for reliability analysis of complex composite material structures with low failure probabilities, ensuring both the accuracy of analysis and the computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] 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 for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on the provided drawings.

[0045] Figure 1 Flowchart of the method for reliability analysis of composite material structures based on an adaptive hybrid surrogate model provided by the present invention;

[0046] Figure 2 Schematic diagram of the scenario of the method for reliability analysis of composite material structures based on an adaptive hybrid surrogate model provided by the present invention;

[0047] Figure 3 Schematic diagram of the entropy weight calculation process provided by the present invention;

[0048] Figure 4 Distribution of sample points under different sampling strategies provided by the present invention;

[0049] Figure 5 Comparison chart of the accuracy and efficiency of different reliability analysis methods in the reliability analysis of steel columns provided by the embodiments of the present invention;

[0050] Figure 6 Comparison of the accuracy and efficiency of different reliability analysis methods in the reliability analysis of an inflatable offshore anti-collision fence structure provided by the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] 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. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0052] Embodiment 1

[0053] An embodiment of the present invention discloses a method for analyzing the reliability of a composite material structure based on an adaptive hybrid surrogate model, as Figure 1 and Figure 2 , including the following steps:

[0054] S1. Define the reliability problem: Determine the distribution type and statistical characteristics of the basic random variables of the composite material structure, set the number of sample points N of the initial design of experiment DOE (Design of Experiment), 0 , and calculate the actual limit state function (Limit State Function, LSF) values of the initial DOE;

[0055] S2. Construct a surrogate model: Use the current DOE to train a surrogate model including a Kriging model and an artificial neural network ANN model, generate a Monte Carlo candidate sample set MC according to the statistical characteristics of the random variables, including N c sample points, and use different surrogate models to predict the sample points in the MC set;

[0056] S3. Calculate the weights of the surrogate models: Based on the EWM entropy weight method, combine the prediction errors and entropy values of the surrogate models, calculate the weights of different surrogate models on the candidate sample points, and realize the adaptive integration of the surrogate models, so as to obtain a combined model;

[0057] S4. Establish a candidate sample set and search for updated points: Use the MMHDR mechanism (Modified Metropolis-Hastings with Delayed Rejection, that is, an improved Metropolis-Hastings algorithm with delayed rejection) to generate a conditional sample point set based on the Monte Carlo candidate sample set MC, and search for updated points through a learning function, and add the updated points to the initial DOE;

[0058] S5. Judge the model update stop condition: Retrain the surrogate model using the updated DOE, and judge whether the accuracy of the current integrated model meets the requirements according to the preset threshold, and repeat steps S3 to S5 until the convergence criterion is met or the maximum number of iterations is reached;

[0059] S6. Output the reliability analysis results: After the iteration ends, output the failure probability of the composite material structure based on the integrated surrogate model.

[0060] In this embodiment, when step S6 outputs the failure probability based on the integrated surrogate model, the reliability index, such as the reliability index β, can also be output simultaneously to provide a more comprehensive reliability analysis result.

[0061] To further implement the above technical solution, step S1 further includes analyzing the composition, function, and working condition of the structure according to the data information of the composite material structure to be evaluated, including but not limited to information such as specifications, design standards, expert opinions, and historical data, and determining the failure mode of the structure and the corresponding limit state equation.

[0062] To further implement the above technical solution, in step S2, when constructing the surrogate model, the cross-validation and jackknife techniques are used to calculate the predicted values and prediction variances of different surrogate models.

[0063] To further implement the above technical solution, as Figure 3 , the specific content of step S3 is:

[0064] Calculate the entropy weight based on the EWM entropy weight method, specifically:

[0065]

[0066] where E j is the entropy of the j-th surrogate model, p ij is the proportion of the i-th sample point in the j-th surrogate model, k is the number of candidate sample points, and N M is the total number of surrogate models;

[0067] Calculate the coefficient of variation G j = 1 - E j , and calculate the entropy weight factor based on the coefficient of variation

[0068] Calculate the weights of different surrogate models at the candidate sample points as:

[0069]

[0070] where e j (x) is the prediction error of the j-th surrogate model at the candidate sample point;

[0071]

[0072] where, is the local prediction result of the combined model, is the mean of the global prediction results of the combined model.

[0073] In this embodiment, the combined model obtained in step S3 is specifically as follows:

[0074]

[0075] where N M represents the total number of combined models, represents the predicted value of the i-th surrogate model, and w i (x) represents the weight of the surrogate model in the i-th combined model. To further implement the above technical solution, in step S4, as Figure 4 , the MMHDR mechanism is used to generate a conditional sample point set based on the Monte Carlo candidate sample set MC, and the specific content of searching and updating points through the learning function is as follows:

[0076] S41. Randomly select half of the sample points (N 1 = 0.5N c ) in the current MC population and their predicted values as the initial subset, and sort the predicted values in ascending order;

[0077] S42. Select the first P 0 N 1 sample points as the seeds for generating the next subset. P 0 is the seed selection ratio. Use the MMHDR strategy to generate N 0 sample points with a probability density function f(x) based on the selected seeds (the initial state points of P 1 Markov chains); 1

[0078] S43. Repeat S42 until the stopping condition is met (until the performance function values of the first P 0 N 1 are not greater than 0). Combine the first reciprocal subset and the second reciprocal subset to search for the local range L e of the surrogate model update points;

[0079] S44. Use the U function to filter the sample points within the local range L e , and obtain the update points, and add the update points to the DOE.

[0080] To further implement the above technical solution, in step S5, the judgment condition for whether the accuracy of the current integrated model meets the requirements is:

[0081]

[0082] where P f,j is the failure probability estimated by the j-th surrogate model, P f is the failure probability calculated by the integrated model, and ε t ​is a preset threshold value, which can be adjusted according to the requirements of specific engineering problems to ensure the accuracy and efficiency of reliability analysis;

[0083] If the condition is satisfied, stop the update; otherwise, return to step S2 to continue updating the surrogate model.

[0084] The method for calculating the failure probability is:

[0085]

[0086] where N MC represents the number of samples in the sample set MC established by the Monte Carlo method, represents marking the number of sample points in the safe state, and its value is determined according to the output of the combined model, which can be expressed by the following formula:

[0087]

[0088] Embodiment 2

[0089] This embodiment takes the reliability analysis of a steel column as an example;

[0090] (1) Analyze the composition, function, and working conditions of the structure based on information such as the instruction manual, design standards, expert opinions, and historical data of the steel column structure to be evaluated, determine the failure mode of the steel column structure and the corresponding limit state equation, and obtain the random variables affecting the structural performance function and their distribution information;

[0091] The limit state equation in this embodiment is expressed as:

[0092]

[0093] where P is the axial force of the steel column and Pcr is the yield stress;

[0094] Pcr is calculated by the following formula:

[0095]

[0096] where λ represents the slenderness ratio control, F y represents the yield stress, and A s represents the cross-section of the column;

[0097] In this embodiment, the statistical characteristics of each random variable are shown in Table 1;

[0098] Table 1 Statistical properties of random variables:

[0099] Variable Name Distribution Mean Cov P(kips) Gumbel 200 0.12 λ Lognormal 1.35 0.12 <![CDATA[F y > Normal 36 0.05 <![CDATA[A s > Normal 21.8 0.05

[0100] (2) Use LHS to generate N 0= An initial DOE of 100 sample points was used, and the Kriging model and artificial neural network model were trained. A three-layer structure was used, including 10 hidden layer nodes, and the ReLU activation function was used. The sample points in the initial DOE and their corresponding limit state function values ​​were used as training data.

[0101] (3) Calculate the prediction errors of different proxy models on the candidate sample points, calculate the entropy values ​​of different proxy models, set the initial weight factor to 0.5, and then determine the weight of each proxy model;

[0102] (4) Using the MMHDR strategy to search for update points in the candidate sample set, in this embodiment, the seed selection ratio P 0 is 0.1;

[0103] (5) Add the updated points to the DOE, use the updated DOE to retrain the proxy model, and repeat steps (3) to (5) until the convergence criterion ε is met. t = 0.01 or the maximum number of iterations reaches 100;

[0104] (6) The reliability index β of the steel column is calculated using the final proxy model, and the entire calculation process is completed.

[0105] The method proposed by the present invention is compared with other reliability analysis methods, and the results are shown in Table 2;

[0106] Table 2 Calculation results of steel column example:

[0107]

[0108]

[0109] It can be seen from Table 2 that the method proposed in the present invention has obvious advantages in both computational efficiency and accuracy. Figure 5 The iterative process of various reliability methods and the weight changes of individual surrogate models of the two proposed reliability methods are shown. It can be seen from the figure that the proposed method converges faster. When comparing the weights of different surrogate models, it can be observed that in the combined surrogate model, the contribution of ANN is relatively small, while the Kriging model plays a more important role in accurately estimating the failure probability. As DOE continues to expand, the Kriging model maintains its dominant position.

[0110] Embodiment 3

[0111] This embodiment takes the reliability analysis of an inflatable marine crash barrier structure as an example.

[0112] (1) Analyze the composition, function, and working conditions of the structure based on information such as the instruction manual, design standards, expert opinions, and historical data of the inflatable offshore anti-collision barrier structure to be evaluated. Determine the failure modes of the structure and the corresponding limit state equations, and obtain the random variables affecting the structural performance function and their distribution information.

[0113] The limit state equation in this embodiment is expressed as:

[0114]

[0115] Among them, x 1 is the thickness of the inflatable offshore anti-collision barrier structure, x 2 is the first modulus of the impact load response, x 3 is the second modulus of the impact load response, x 4 is the HSMST full impact stress, x 5 is the HSMST full impact strain; in this embodiment, each variable follows a normal distribution, and the statistical characteristics are shown in Table 3.

[0116] Table 3 Statistical properties of random variables:

[0117]

[0118]

[0119] (2) Use LHS to generate an initial DOE containing N 0 = 200 sample points; train the Kriging model and the artificial neural network model, which contains 20 hidden layer nodes, use the Sigmoid activation function, and use the sample points in the initial DOE and their corresponding limit state function values as training data.

[0120] (3) Calculate the prediction errors of different surrogate models at the candidate sample points, calculate the entropy values of different surrogate models, set the initial weight factor to 0.5, and then determine the weights of each surrogate model.

[0121] (4) Use the MMHDR strategy to search for updated points in the candidate sample set. In this embodiment, the seed selection ratio P 0 is 0.1.

[0122] (5) Add the updated points to the DOE, and retrain the surrogate model using the updated DOE. Repeat steps (3) to (5) until the convergence criterion ε t = 0.01 or the maximum number of iterations 100 is reached.

[0123] (6) Use the final surrogate model to calculate the reliability index β of the inflatable offshore anti-collision barrier structure, and the entire calculation process ends.

[0124] The method proposed in the present invention is compared with other reliability analysis methods, and the results are shown in Table 4;

[0125] Table 4 Calculation results of the sea inflatable fender sea blocking structure example:

[0126]

[0127]

[0128] It can be seen from the table that the method proposed in the present invention shows obvious advantages in both calculation efficiency and accuracy; further, Figure 6 The iterative processes of various reliability methods and the weight changes of the single surrogate models of the two proposed reliability methods are presented. It can be seen from the figure that the EWM-MMHDR method converges faster, and the region of the LSF is quickly identified by the MMHDR method, significantly improving the efficiency of the reliability method based on the agent model; when comparing the weight changes of different surrogate models in the proposed methods, it is noted that the Kriging model initially receives a larger weight at the beginning of the iteration. However, as the DOE becomes more comprehensive, the artificial neural network model plays a more substantial role in the combined model.

[0129] Example 4

[0130] This example proposes a composite material structure reliability analysis system based on an adaptive hybrid surrogate model, based on a composite material structure reliability analysis method based on an adaptive hybrid surrogate model, including: a reliability problem definition module, an initial surrogate model construction module, a weight calculation and model integration module, a model optimization module, and a reliability result output module;

[0131] The reliability problem definition module is used to determine the distribution type and statistical characteristics of the basic random variables of the composite material structure, set the number of sample points of the initial design of experiments DOE, and calculate the actual limit state function values of the initial DOE;

[0132] The initial surrogate model construction module is used to train surrogate models including the Kriging model and the artificial neural network ANN model using the current DOE, generate a Monte Carlo candidate sample set MC according to the statistical characteristics of the random variables, and predict the sample points in the MC set using different surrogate models;

[0133] The weight calculation and model integration module is used to calculate the weights of different surrogate models at the candidate sample points based on the EWM entropy weight method, combining the prediction errors and entropy values of the surrogate models, and realize the adaptive integration of the surrogate models, so as to obtain a combined model;

[0134] The model optimization module is used to generate a conditional sample point set based on the Monte Carlo candidate sample set MC by adopting the MMHDR mechanism, search for updated points through a learning function, and add the updated points to the initial DOE; re-train the surrogate model using the updated DOE, and judge whether the accuracy of the current integrated model meets the requirements according to a preset threshold until the convergence criterion is met or the maximum number of iterations is reached;

[0135] The reliability result output module is used to output the failure probability of the composite material structure based on the integrated surrogate model after the iteration ends.

[0136] A computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, it implements a reliability analysis method for a composite material structure based on an adaptive hybrid surrogate model.

[0137] A processing terminal includes a memory and a processor. A computer program that can run on the processor is stored in the memory. When the processor executes the computer program, it implements a reliability analysis method for a composite material structure based on an adaptive hybrid surrogate model.

[0138] 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. The same or similar parts among the embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0139] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A composite material structure reliability analysis method based on an adaptive hybrid surrogate model, characterized in that: The following steps are involved: S1. Determine the distribution type and statistical characteristics of the basic random variables of the composite structure, set the number of sample points of the initial design experiment DOE, and calculate the actual limit state function value of the initial DOE; S2, using the current DOE to train the proxy models including the Kriging model and the artificial neural network ANN model, generating the Monte Carlo candidate sample set MC according to the statistical characteristics of the random variables, and using different proxy models to predict the sample points in the MC set; S3, based on the EWM entropy weight method, combined with the prediction error and entropy value of the proxy model, the weights of different proxy models on the candidate sample points are calculated to achieve adaptive integration of the proxy models, thereby obtaining a combined model; S4, using the MMHDR mechanism to generate a set of conditional sample points based on the Monte Carlo candidate sample set MC, and searching for update points through the learning function, and adding the update points to the initial DOE; S5, retraining the proxy model using the updated DOE, judging whether the accuracy of the current integrated model meets the requirements according to the preset threshold, and repeating steps S3 to S5 until the convergence criterion is met or the maximum number of iterations is reached; S6. After the iteration, the failure probability of the composite structure based on the integrated agent model is output.

2. A composite material structure reliability analysis method based on an adaptive hybrid agent model according to claim 1, characterized in that: Step S1 also includes analyzing the composition, function and working conditions of the structure according to the data information of the composite material structure to be evaluated, and determining the failure mode of the structure and the corresponding limit state equation.

3. The composite material structure reliability analysis method based on the adaptive hybrid agent model according to claim 1 is characterized in that: In step S2, when constructing the surrogate model, the predicted values ​​and prediction variances of different surrogate models are calculated using cross-validation and jackknife techniques.

4. The composite material structure reliability analysis method based on the adaptive hybrid agent model according to claim 1 is characterized in that: The specific content of step S3 is: The entropy weight is calculated based on the EWM entropy weight method, specifically: Among them, E j is the entropy of the jth surrogate model, p ij is the proportion of the i-th sample point in the j-th proxy model, k is the number of candidate sample points, N M is the total number of proxy models; Calculate the coefficient of variation G for each model j =1-E j , and calculate the entropy weight factor based on the difference coefficient The weights of different proxy models on candidate sample points are calculated as: Among them, e j (x) is the prediction error of the j-th proxy model on the candidate sample point.

5. The composite material structure reliability analysis method based on the adaptive hybrid agent model according to claim 1 is characterized in that: In step S4, the MMHDR mechanism is used to generate a conditional sample point set based on the Monte Carlo candidate sample set MC, and the specific content of searching for update points through the learning function is: S41. Randomly select half of the sample points N1 in the current MC population and their predicted values ​​as the initial subset, and sort the predicted values ​​in ascending order; S42. Select the first P0N1 samples as the seeds for generating the next subset, P0 is the seed selection ratio, and use the MMHDR strategy to generate N1 conditional sample point sets with a probability density function f(x) based on the selected seeds; S43. Repeat this process until the stop condition is met, and combine the first reciprocal subset with the second reciprocal subset to search the local range L of the proxy model update point e ; S44. Use U function to filter the local range L e , and obtain the updated points, and add the updated points to the DOE.

6. The composite material structure reliability analysis method based on the adaptive hybrid agent model according to claim 1 is characterized in that: In step S5, the judgment condition of whether the accuracy of the current integrated model meets the requirements is: Among them, P f,j is the failure probability estimated by the jth surrogate model, P f is the failure probability calculated by the integrated model, ε t is the preset threshold; If the conditions are met, the updating is stopped; otherwise, the process returns to step S2 to continue updating the proxy model.

7. A composite material structure reliability analysis system based on an adaptive hybrid agent model, characterized in that: A composite material structure reliability analysis method based on an adaptive hybrid agent model according to any one of claims 1 to 6, comprising: a reliability problem definition module, an initial agent model construction module, a weight calculation and model integration module, a model optimization module and a reliability result output module; Reliability problem definition module, which is used to determine the distribution type and statistical characteristics of the basic random variables of the composite structure, set the number of sample points of the initial design experiment DOE, and calculate the actual limit state function value of the initial DOE; The initial proxy model building module is used to use the current DOE to train the proxy models including the Kriging model and the artificial neural network ANN model, generate the Monte Carlo candidate sample set MC according to the statistical characteristics of the random variables, and use different proxy models to predict the sample points in the MC set; The weight calculation and model integration module is used to calculate the weights of different proxy models on candidate sample points based on the EWM entropy weight method, combined with the prediction error and entropy value of the proxy model, to achieve adaptive integration of the proxy model, thereby obtaining a combined model; The model optimization module is used to generate a set of conditional sample points based on the Monte Carlo candidate sample set MC using the MMHDR mechanism, and search for update points through the learning function, and add the update points to the initial DOE; the updated DOE is used to retrain the proxy model, and according to the preset threshold, whether the accuracy of the current integrated model meets the requirements, until the convergence criterion is met or the maximum number of iterations is reached; The reliability result output module is used to output the failure probability of the composite material structure based on the integrated agent model after the iteration is completed.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for analyzing the reliability of a composite material structure based on an adaptive hybrid agent model as described in any one of claims 1 to 6 is implemented.

9. A processing terminal, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, a composite material structure reliability analysis method based on an adaptive hybrid agent model as described in any one of claims 1 to 6 is implemented.

Images