A method and platform for identifying impact dynamic parameters of large explosion-resistant structures

Abstract

The present invention relates to a method for identifying impact dynamic parameters of a large-scale explosion-proof structure. The identification method comprises determining the impact dynamic parameters of the structure to be identified; measuring the geometric parameters of the structure to be identified; measuring the dynamic response of the structure to be identified under an explosion impact; establishing a numerical model of the structure to be identified under the explosion impact and constructing an objective function; using an improved MWOA-GPO algorithm to quickly find the optimum in a parameter range; and calculating the mechanical parameters corresponding to the minimum target value after several iterations. In view of the problems of difficulty in obtaining dynamic parameters of explosion-proof structures of large-scale protective engineering and inaccurate measurement results, the present invention proposes a method for optimizing a finite element numerical model by using the MWOA-GPO algorithm and then inverting the dynamic parameters of the structure. At the same time, based on the identification method, an identification platform for obtaining the dynamic parameters of the structure is proposed, which effectively improves the convenience and economy of obtaining the structural parameters.

Description

Technical Field

[0001] The invention relates to an impact dynamic parameter identification method and an identification platform for a large explosion-proof structure, and belongs to the field of safety protection engineering. Background Art

[0002] In recent years, the number of safety accidents caused by chemical explosions, such as those involving explosives, has been increasing. When a sudden explosion occurs, it releases enormous amounts of energy in a short period of time, imposing a sudden surge in impact loads on surrounding buildings and causing extensive damage to the structures in a short period of time. To reduce the impact of the explosion on surrounding buildings, a commonly used impact-resistant strategy in protective engineering is to use explosion-resistant structures with stronger impact resistance. The dynamic mechanical parameters of the protective explosion-resistant structures themselves determine their resistance to explosion shocks. Therefore, in order to more rationally deploy explosion-resistant structures, it is crucial to accurately determine their key dynamic mechanical parameters under the impact of explosions.

[0003] Generally, the dynamic mechanical parameters of materials are obtained by sampling the materials on-site and conducting corresponding impact dynamic tests. The necessary dynamic parameters are obtained based on the test results. The main drawback of this parameter acquisition method is that different materials have inevitable errors during the production stage. Even materials produced in the same batch will have different fluctuations in their dynamic parameters. Therefore, measuring the dynamic parameters of materials through experiments generally requires setting up multiple batches of parallel experiments for comparative verification. Based on the results of multiple experiments, the average value of the mechanical parameters is determined. This parameter acquisition method has high experimental costs and poor economic efficiency. In addition, the service life of protective engineering structures is long. Compared with the initial use stage when the structure is constructed, protective engineering is subject to long-term environmental erosion after a long service period. Therefore, only using the impact dynamic parameters of the initial stage to replace the impact dynamic parameters of the material when it is subjected to explosion impact during different service periods ignores the cyclic effect of the material.

[0004] Unlike typical small and medium-sized structures, small explosion-resistant structures are smaller overall, typically with individual components typically measuring less than 20m, and use fewer component types. Large explosion-resistant structures, on the other hand, typically span approximately 100m and require a greater variety of components. Large explosion-resistant structures often employ a greater variety of components and composite materials, requiring a greater number of material dynamic mechanical parameters to be determined. Continuing to use the method of sampling from components to measure dynamic parameters would incur higher costs and lower economic efficiency. Furthermore, the greatest drawback of single-material mechanical testing is that, due to the objective limitations of impact dynamic testing equipment, only specimens are sampled from the entire structure during single-material impact testing. These specimens are small in size, ignoring the structural connections between components within the complex structure and often failing to truly reflect the structure's overall impact resistance to explosions. For large-scale protective engineering structures, due to the large number of internal components and the significant interaction effects of the components, in view of the limitations of the above-mentioned parameter acquisition methods, a new method for calculating the impact dynamic parameters of protective structures is needed, which can not only effectively identify the impact dynamic parameters of explosion-proof structures during the use phase, but also take into account the overall impact resistance effect of the structure, and has considerable accuracy and is convenient in operation. Summary of the Invention

[0005] The present invention provides a method and platform for identifying the impact dynamic parameters of large-scale explosion-resistant structures. The MWOA-GPO algorithm is used to optimize the finite element numerical model and then invert the dynamic parameters of the structure, effectively improving the convenience and economy of obtaining the structural parameters.

[0006] The technical solution adopted by the present invention to solve its technical problem is:

[0007] A method for identifying impact dynamic parameters of a large explosion-resistant structure comprises the following steps:

[0008] Step S1: determining the impact dynamic parameters of the structure to be identified;

[0009] Step S2: Determine the geometric parameters of the structure to be identified;

[0010] Step S3: measuring the dynamic response of the structure to be identified under the impact of the explosion;

[0011] Step S4: establishing a numerical model of the structure to be identified under the impact of the explosion and constructing the objective function;

[0012] Step S5: Using the improved MWOA-GPO algorithm to quickly optimize in the parameter interval, that is, using the improved gray wolf algorithm to optimize in the global space. When the number of global optimization reaches the preset number of iterations, the optimization in the global space is transformed into the optimization in the local space. At this stage, the GPO local optimization technology is started;

[0013] Step S6: Repeated iteration of global optimization and local optimization until the convergence iteration criterion meets the adaptation accuracy and the maximum allowed number of iterations, calculates the mechanical parameters corresponding to the minimum target value, exits the optimization process, and outputs the optimal structural dynamic parameters;

[0014] As a further preferred embodiment of the present invention, the impact dynamic parameters of the structure to be identified determined in step S1 include the dynamic mechanical parameters of a single material if the structure is a single-medium structure, and include the dynamic mechanical parameters of the materials of each component if the structure is a multi-medium combination structure;

[0015] The aforementioned material dynamic mechanical parameters include determining the dynamic parameters required in actual engineering design, the initial values of unknown parameters, and the upper and lower limits of the preset parameter search range;

[0016] As a further preferred embodiment of the present invention, in step S2, the geometric parameters of the structure to be identified are measured by a contact or non-contact measurement method;

[0017] As a further preferred embodiment of the present invention, in step S3, the dynamic response of the structure to be identified under the impact of the explosion is the measured dynamic response of the key points of the structure, including the impact damage deformation value W of the structure. kp1 、W kp2 ,…,W kpn , the impact overpressure value P of the structure kp1 、P kp2 ,…,P kpn And the acceleration value A of the structure kp1 、A kp2 ,…,A kpn ;

[0018] As a further preferred embodiment of the present invention, in step S4, the geometric parameters of the structure to be identified obtained in step S2 are combined with the initial values of the position parameters in step S1 to establish an initial numerical model of the structure;

[0019] The constructed objective function is Among them, x is a set of material dynamic mechanical parameters, S i is the dynamic response value of the i-th key monitoring point on the structure after the explosion, is the calculated dynamic response value of the finite element at the i-th displacement monitoring point under this set of mechanical parameters;

[0020] When the value of the objective function satisfies the convergence condition of the objective function min f(x), the dynamic parameter values in the numerical model are the dynamic parameters in the real structure;

[0021] As a further preference of the present invention, the specific steps of step S5 are:

[0022] Step S5-1, performing iterative optimization of the improved grey wolf algorithm in the global space;

[0023] Step S5-2, dynamic acceleration of GPO in local space;

[0024] Step S5-1 is specifically as follows:

[0025] Step S5-1-1, set the parameters of the MWOA algorithm, including the number of gray wolf populations to be searched NP, the dimension of the algorithm D, and the number of local optimization times and the start time Iter max , when the MWOA algorithm reaches the fitness evaluation times t in the single cycle stage, max When , it enters the GPO local acceleration stage;

[0026] Step S5-1-2, before each iterative optimization of the MWOA algorithm, calculate the fitness value of each wolf in the search population, and save the position information of the wolf with the best fitness value in the population as α, the position information of the wolf with the second best fitness value in the population as β, the position information of the gray wolf with the third best fitness value in the population as δ, and the position information of the remaining wolves as ω; when the four classes in the wolf pack are determined, start the optimization in the global space, and update the position of the next generation of wolf pack particles according to the iterative formula of the improved gray wolf algorithm. When the wolf population calculation is updated in each iterative update, the number of fitness calculations calculated numerically at this time is t=t+1;

[0027] Step S5-1-3, in the optimization process, the potential optimal particle swarm needs to be brought into the real fitness function to calculate the real fitness value, that is, according to the position I of the optimal individual of the current gray wolf population ij In the numerical model established in the dynamic update step S4, the current parameter I is calculated according to the updated numerical model. ij The fitness value FE(I ij ), where the population composed of particles with the best fitness in each group is recorded as QBE(i), and the best individual corresponding to the minimum fitness value FE among all particles is recorded as the current global best individual PB gj , the corresponding objective function value BFE(PB gj ) is the current global optimal solution; the number of calculations of the numerical model reaches t = Iter max After that, it enters the GPO acceleration phase;

[0028] Step S5-2 is specifically as follows:

[0029] Step S5-2-1, establish a Gaussian process agent model, and find the global optimal individual PB in step S5-1 gj Nearby, according to distance PB gj The number of points closest to the target is NP×D, and a new learning sample set (X sample , Y sample ), generate a local proxy model;

[0030] Step S5-2-2: Based on the established GPR model, the GPO local acceleration technology is used to quickly find the current local optimal point BL. ij , and BL ij The position information of is input into the numerical model and its fitness FE(BL ij );

[0031] Step S5-2-3, based on the BL found in step S5-2-2 ij The fitness FE(BL ij ) and the current optimal point PB found in step S5-1-3 gj The fitness FE(PB gj ) are compared, and according to the evaluation results, if FE(BL ij )>FE(PB gj ), prove that the acceleration is successful, then let the global optimal point PB gj =BL ij ; If FE(BL ij ) <FE(PB gj ), then the acceleration is unsuccessful, then PB gj After the judgment is completed, the global optimization phase of the improved grey wolf algorithm MWOA is returned to step S5-1-2. In step S5-1-2, the optimal point PB is selected. gj Continue the iterative process of global optimization and local acceleration until the convergence condition is met;

[0032] As a further preferred embodiment of the present invention, the optimization process of the MWOA algorithm in the global space in step S5-1-2 includes:

[0033] Step S5-1-2-1, wolf pack initial position information P ij =I ij , the first generation population P ij Substitute into the numerical model built in step S4 to perform fitness evaluation and obtain the first generation fitness value FE of all individuals 1st (P ij ), the particle P with the smallest fitness value in the current setxj is the current global optimal grasshopper, and its corresponding objective function value FE 1st (P xj ) is the current first-generation global optimal solution;

[0034] Step S5-1-2-2, update the search particle position according to the improved wolf pack algorithm MWOA algorithm optimization formula to obtain the next search population position P nextij ;

[0035] Step S5-1-2-3, update the current wolf pack position P nextij Bring it into step S5-1-3 to perform fitness evaluation and obtain the objective function value FE (P nextij ); Current P nextij The wolf individual N with the smallest fitness function value mingj is the current global optimal wolf pack individual, and its corresponding objective function value FE(N mingj ) is the current global optimal solution;

[0036] Step S5-1-2-4, based on the current wolf pack position P nextij , repeat the above steps, at this time the number of iterations t=t+1, after several iterations, the number of global optimization iterations t=Iter max , entering the GPO acceleration phase;

[0037] As a further preferred embodiment of the present invention, the specific optimization steps of the improved grey wolf algorithm using the positions of α, β and δ are as follows:

[0038] Step a-1: Surround the prey

[0039] Formula (1) represents the distance between gray wolf individuals. The update method of offspring gray wolf individuals is expressed as formula (2):

[0040] D=|C·Y p (t c )-Y(t c )| (20)

[0041] Y(t c +1)=Y p (t c )-A·Dist (21)

[0042] Where Dist is the distance between the hunting wolf pack and the target, X p (t c ) is the prey position, t c is the current iteration number, X(t c) is the current position of the wolf pack, A and C are coefficient control vectors, and the control parameter C is mainly responsible for controlling the gray wolf algorithm's ability to explore the unknown space. The control parameter C is not a fixed random value, but changes dynamically with the number of iterations. The control parameter C modified by MWOA is calculated according to the following formula:

[0043] A=2×au×r1-au (22)

[0044] C=2×r2-au / 2 (23)

[0045] During the iteration process, the value of au decreases linearly from 2 to 0 as the number of iterations increases. r1 and r2 are random variables and are in the range of [0, 1].

[0046] Step a-2: Capturing the prey

[0047] The position of ω is updated based on the positions of α, β, and δ as follows:

[0048] Dist α =|C1·Y α (t c )-Y(t c )| (24)

[0049] Dist β =|C2·Y β (t c )-Y(t c )| (25)

[0050] Dist δ =|C3·Y δ (t c )-Y(t c )| (26)

[0051] Then, the ω wolf pack updates towards the positions of α, β and δ as follows:

[0052] Y1(t c +1)=Y α (t c )-A1·Dist α (27)

[0053] Y2(t c +1)=Y β (t c )-A2·Dist β (28)

[0054] Y3(t c +1)=Y δ (t c )-A3·Dist δ (29)

[0055] The final position of the offspring wolf pack is:

[0056] Y3=(Y1+Y2+Y3) / 3 (30);

[0057] The technical steps for GPO acceleration are as follows:

[0058] Step b-1: Generation process of Gaussian process model:

[0059] The statistical characteristics of the GPR regression model are completely determined by its mean m(X) and covariance function k(X,X / ) to determine the definition, the definition is:

[0060] f(X)~GP(m(X),k(X,X / )) (31)

[0061] Assume there is a training set of n observation data D={(x i ,y i )|i=1,2,,,n},x i For the d-dimensional input vector, the target value y i ∈R, if the matrix X represents the d×n-dimensional input matrix and the vector y represents the output vector, then the training set is abbreviated as D=(X,y); in the actual process, the noise ε is an independent random variable, that is:

[0062] ε~N(0,σ 2 n ) (32)

[0063] Then, with the addition of Gaussian noise to the standard regression model, the prior distribution of the input vector y is:

[0064] y~N(0,K+σ 2 n I) (33)

[0065] Where: K = K(X,X) is an m×m order symmetric positive definite covariance matrix, and any item k in the K matrix ij Measured x i and x j (i, j = 1, 2, ... m) correlation, I is the m × m unit matrix;

[0066] Step b-2: Gaussian process model fitting new sample points:

[0067] For a new test input matrix X * , GP model predicts the same * The corresponding output vector y * , then the output vector y of the training sample and the test output vector f* The resulting joint distribution Gaussian prior distribution is:

[0068]

[0069] Where: K=K(X,X * ) is the test input matrix X * The m×1 order covariance matrix with the input training sample matrix X can be abbreviated as K(X * ), K(X * ,X * ) is the test point input matrix X * The covariance matrix of itself can be abbreviated as k(X * );

[0070] The mean and variance of the function's predicted values are:

[0071]

[0072]

[0073] In the above formula, It is the fitting regression model based on the training sample points, where the individual position coordinates X are implicitly included. * With fitness f * The corresponding relationship is used to replace the mapping function curve between the dynamic parameters and the target adaptation function;

[0074] Step b-3: Local accelerated optimization process based on the agent model:

[0075] Based on the fitting model established in formula (17), new samples can be obtained by the extreme value of the convergence function, which is referred to as the LCB criterion:

[0076]

[0077] in: kσ are the mean and variance of the function prediction value, and the calculation formula of the constant k expressing the confidence range is

[0078] k=3-2.9×i / M (38)

[0079] Among them, i is the current iteration number iter, M is the maximum iteration number Maxiter;

[0080] As a further preferred embodiment of the present invention, the convergence iteration criterion includes the maximum total number of times the algorithm fitness function is calculated ITR max , which is also the maximum number of finite element calls; the convergence value error of the algorithm, the calculated objective function min f(x) satisfies the convergence value, set to 1e -3 .

[0081] An identification platform for the method for identifying impact dynamic parameters of large explosion-resistant structures, comprising a geometric model M1, a dynamic response monitoring module M2, a numerical model module M3, a parameter calculation module M4, a storage and transmission module M5, and a control and display module M6, which are electrically connected to each other;

[0082] The dynamic response monitoring module M2 includes a structural damage and deformation measurement unit M2-1 for measuring the damage and deformation displacement value of key points of the large structure under the impact load of the explosion, a structural overpressure measurement unit M2-2 for measuring the overpressure value of key points of the large structure under the impact load of the explosion, and a structural acceleration measurement unit M2-3 for measuring the vibration acceleration value of key points of the large structure under the impact load of the explosion;

[0083] The numerical model module M3 includes an initial model unit M3-1 and a model iterative update unit M3-2;

[0084] The parameter calculation module M4 includes a structural parameter initialization unit M4-1 for setting the initial solution interval of the structure, a MWOA calculation unit M4-2 for solving the dynamic parameters of the structure in the global space using the MWOA algorithm, and a GPR acceleration unit M4-3 for accelerating the solution speed of the MWOA in the local space using the GPR model;

[0085] The storage and transmission module M5 includes a local storage unit M5-1 for storing the calculated data in a local data center centralized device, a backup storage unit M5-2 for backing up the data and storing it in the cloud, and a transmission unit M5-3 for sending the calculated data to a receiving platform or a corresponding client via a universal wireless, wired, or selective transmission method;

[0086] The control and display module M6 includes an information display unit M6-1 for displaying various types of information and a time control unit M6-2 that can enter any module to direct the progress under the administrator's instruction.

[0087] Through the above technical solution, compared with the existing technology, the present invention has the following beneficial effects:

[0088] 1. The identification method provided by the present invention adopts the means of numerical model and optimization algorithm fitting, eliminating the need for single material mechanics experiments using existing materials. It transfers the method of obtaining parameters of explosion-proof protective structures from traditional mechanical calculations and mechanical tests to a computer computing platform. The required dynamic parameters are obtained through fast and convenient numerical calculations, reducing the cost of obtaining material parameters, and the obtained structural dynamic parameters have good economic efficiency.

[0089] 2. The identification method provided by the present invention innovatively applies the improved MWOA-GPO algorithm to the optimization problem of the impact dynamic parameters of large explosion-resistant structures. In the process of global optimization of the MWOA algorithm, the local GPR proxy model is embedded in the local optimization of the optimization algorithm, and the proxy model is used to replace the real numerical calculation. The local optimization method based on the proxy model finds the potential optimal point, which solves the dilemma of the MWOA algorithm falling into the local optimal value in the local space. It gives full play to the characteristics and advantages of the MWOA algorithm, GPO machine learning method and numerical calculation model technology, and reduces the number of numerical re-analysis under the premise of obtaining the global optimal solution, effectively reduces the high computational cost problem caused by optimizing the numerical calculation target parameters, and realizes the rapid acquisition of the structural parameters of the protective engineering.

[0090] 3. The identification platform provided by the present invention realizes the integrated process of monitoring, modeling, correction, attribute parameter setting, optimization, and feedback information, effectively improving the intelligence level of anti-explosion dynamic parameter acquisition, and timely performing information correction and attribute setting, providing a substantial solution to improve the deficiencies of traditional structural dynamic parameters. At the same time, it avoids the computational waste caused by overly deep searches, reduces the computational cost, and provides a commercial solution for the acquisition of dynamic parameters of complex structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] The present invention will be further described below with reference to the accompanying drawings and examples.

[0092] Figure 1 This is a flow chart of the method for identifying impact dynamic parameters of large explosion-proof structures provided by the present invention;

[0093] Figure 2 is a flow chart of the MWOA-GPO algorithm provided by the present invention;

[0094] Figure 3 This is a schematic diagram of the identification device and each unit in the identification device provided by the present invention;

[0095] Figure 4 It is a geometric dimension diagram of an implementation case provided by the present invention;

[0096] Figure 5 It is the dynamic response monitoring point of the implementation case provided by the present invention;

[0097] Figure 6 It is a numerical model of a structure of an embodiment provided by the present invention;

[0098] Figure 7 This is a layout diagram of the identification device provided by the present invention being set in a structure. DETAILED DESCRIPTION

[0099] The present invention will now be described in further detail with reference to the accompanying drawings. In the description of this application, it should be understood that the terms "left side", "right side", "upper", "lower", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are intended only to facilitate the description of the present invention and simplify the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. "First", "second", etc. do not indicate the importance of the components and therefore should not be understood as limiting the present invention. The specific dimensions used in this embodiment are only for illustrative purposes only and do not limit the scope of protection of the present invention.

[0100] As described in the background art, existing methods for obtaining dynamic mechanical parameters of materials often have poor accuracy in the parameters obtained, high calculation costs, and poor economic efficiency. Therefore, this application aims to propose a method for identifying the impact dynamic parameters of large explosion-proof structures. Based on the explosion experiment and the measured structural dynamic response, the improved optimization strategy MWOA-GPO algorithm and numerical model technology are used to quickly find the optimal solution within the parameter range and solve the structural impact dynamic parameters that best match the field experimental results. In other words, the calculation principle of the method for calculating the impact dynamic parameters of the corresponding components of the explosion-proof structure proposed in this application is to convert the original parameter test problem into a constrained optimization problem of the target parameters. Based on the actual structural dynamic response of the actual structure under the field explosion load, the target dynamic parameters are used as the optimization variables, and the optimization power of the optimization algorithm is used to find a set of dynamic parameters that best match the actual explosion of the structure, that is, the dynamic parameter value of the structure.

[0101] It should be noted here that when using the inversion method to calculate the impact dynamic parameters of large structures, due to the large size of large structures and the more complex shapes of components, it is difficult to establish a target fitness function that is displayed between the target parameters and the structural dynamic response. In order to more clearly express the relationship between the target parameters and the structural dynamic response, an implicit numerical relationship between the optimization parameters and the structural dynamic response is often established on a finite element numerical calculation platform. However, due to the complexity of large structures, excessive finite element calculations will result in a situation where high computational time is consumed, i.e., the so-called expensive computation problem. In order to be able to use finite element technology to invert the target parameters of the structure and to further reduce the computational time and cost when calculating the target dynamic parameters of the structure using finite element methods. Therefore, in response to the high computational problem that occurs in the numerical optimization calculation of the optimization algorithm, this application proposes a joint optimization strategy based on an improved grey wolf algorithm (MWOA) and Gaussian process optimization (GPO), referred to as (MWOA-GPO). The MWOA-GPO algorithm integrates the Gaussian process local optimization algorithm (GPO) into the MWOA algorithm to reduce the number of numerical calculations as much as possible, thereby reducing the calculation time and improving the efficiency of inversion.

[0102] Next, we will elaborate on the identification method. Figure 1-Figure 2 As shown, the following steps are included:

[0103] Step S1: Determine the impact dynamic parameters of the structure to be identified; the impact dynamic parameters of the structure to be identified include the dynamic mechanical parameters of a single material if the structure is a single-medium structure, and the dynamic mechanical parameters of the materials of each component if the structure is a multi-medium combination structure; the aforementioned dynamic mechanical parameters of the materials include the dynamic parameters required in actual engineering design, such as: dynamic elastic model, dynamic strength, dynamic amplification factor and other parameters; the initial values of the unknown parameters, and the upper and lower limits of the given search interval of the parameters, and the subsequent optimization process steps of the algorithm will be executed within this interval.

[0104] Step S2: Determine the geometric parameters of the structure to be identified; use a contact or non-contact measurement method to determine the geometric parameters of the structure to be identified.

[0105] Step S3: Determine the dynamic response of the structure to be identified under the impact of the explosion, which is mainly used to monitor the dynamic deformation of the structure under the real explosion impact load. The monitored dynamic response values mainly include the consistent numerical model at the key points of the structure, the impact dynamic response of the structure under the explosion load; including the impact damage deformation value W of the structure kp1 、W kp2 ,…,W kpn , the impact overpressure value P of the structure kp1 、P kp2 ,…,P kpn And the acceleration value A of the structure kp1 、A kp2 ,…,A kpn .

[0106] Step S4: Establish a numerical model of the structure to be identified under the impact of the explosion and construct the objective function; specifically, the geometric parameters of the structure to be identified obtained in step S2 are combined with the initial values of the position parameters in step S1 to establish an initial numerical model of the structure; then, the objective function of the inversion problem is constructed as Among them, x is a set of material dynamic mechanical parameters, S i is the dynamic response value of the i-th key monitoring point on the structure after the explosion, is the calculated dynamic response value of the finite element at the i-th displacement monitoring point under this set of mechanical parameters; the meaning of the objective function min f(x) can be understood as the error between the numerical model and the real structure. The calculation goal is to find a set of dynamic parameters that minimizes the objective function value min f(x) after several iterations. When the value of the objective function meets the convergence condition of the objective function min f(x), it can be considered that the numerical model and the real structure have good consistency. In this case, the dynamic parameter values in the numerical model are the dynamic parameters in the real structure.

[0107] Step S5: Using the improved MWOA-GPO algorithm to quickly optimize in the parameter interval, that is, using the improved gray wolf algorithm to optimize in the global space. When the number of global optimization reaches the preset number of iterations, the optimization in the global space is transformed into the optimization in the local space. At this stage, the GPO local optimization technology is started;

[0108] The idea behind this step is that, in the initial stage, a modified Grey Wolf Algorithm (MWOA) is used to perform optimization in the global space. When the global optimization reaches a certain number of iterations, in order to significantly reduce the number of numerical model calculations, the global optimization is switched to the local optimization. At this stage, the GPO local optimization technique is activated. According to the GPO optimization steps, a GPR proxy model is first generated near the current optimal point. Then, the local optimal point is found according to the GPO optimization criteria. After repeated iterations of global and local optimization, the algorithm optimization process is exited until the convergence criteria meet the fitness accuracy and the maximum allowed number of iterations. The calculated result is the optimal set of structural dynamic parameters. By using the local GPO-based acceleration technique, the global optimization speed of the algorithm is greatly improved and the computational time cost is reduced. Step S5 comprises: step S5-1, optimization of the MWOA algorithm in the global space; step S5-2, dynamic acceleration of the GPO in the local space. The implementation of step S5 relies on the interaction between steps S5-1 and S5-2.

[0109] Step S5-1 is specifically as follows:

[0110] Step S5-1-1, set the parameters of the MWOA algorithm, including the number of gray wolf populations to be searched NP, the dimension of the algorithm D, and the number of local optimization times and the start time Iter max , when the MWOA algorithm reaches the fitness evaluation times t in the single cycle stage, max When the GPO local acceleration phase begins, the current number of iterations t>Iter max The convergence criteria of the MWOA-GPO algorithm are: objective function convergence condition ε, global optimization number T max ; Initialize the random distribution search population position I ij, where i is the population size and j is the number of dynamic parameters of the structure to be inverted;

[0111] Step S5-1-2, before each iterative optimization of the MWOA algorithm, calculate the fitness value of each wolf in the search population, and save the position information of the wolf with the best fitness value in the population as α, the position information of the wolf with the second best fitness value in the population as β, the position information of the gray wolf with the third best fitness value in the population as δ, and the position information of the remaining wolves as ω; when the four classes in the wolf pack are determined, start the optimization in the global space, and update the position of the next generation of wolf pack particles according to the iterative formula of the improved gray wolf algorithm. When the wolf population calculation is updated in each iterative update, the number of fitness calculations calculated numerically at this time is t=t+1;

[0112] Step S5-1-2 The optimization process of the MWOA algorithm in the global space includes:

[0113] Step S5-1-2-1, wolf pack initial position information P ij =I ij , the first generation population P ij Substitute into the numerical model built in step S4 to perform fitness evaluation and obtain the first generation fitness value FE of all individuals 1st (P ij ), the particle P with the smallest fitness value in the current set xj is the current global optimal grasshopper, and its corresponding objective function value FE 1st (P xj ) is the current first-generation global optimal solution;

[0114] Step S5-1-2-2, update the search particle position according to the improved wolf pack algorithm MWOA algorithm optimization formula to obtain the next search population position P nextij ;

[0115] Step S5-1-2-3, update the current wolf pack position P nextij Bring it into step S5-1-3 to perform fitness evaluation and obtain the objective function value FE (P nextij ); Current P nextij The wolf individual N with the smallest fitness function value mingj is the current global optimal wolf pack individual, and its corresponding objective function value FE(N mingj ) is the current global optimal solution;

[0116] Step S5-1-2-4, based on the current wolf pack position P nextij , repeat the above steps, at this time the number of iterations t=t+1, after several iterations, the number of global optimization iterations t=Iter max , entering the GPO acceleration phase.

[0117] Step S5-1-3, in the optimization process, the potential optimal particle swarm needs to be brought into the real fitness function to calculate the real fitness value, that is, according to the position I of the optimal individual of the current gray wolf population ij In the numerical model established in the dynamic update step S4, the current parameter I is calculated according to the updated numerical model. ij The fitness value FE(I ij ), where the population composed of particles with the best fitness in each group is recorded as QBE(i), and the best individual corresponding to the minimum fitness value FE among all particles is recorded as the current global best individual PB gj , the corresponding objective function value BFE(PB gj ) is the current global optimal solution; the number of calculations of the numerical model reaches t = Iter max After that, it enters the GPO acceleration phase;

[0118] The modified grey wolf optimization algorithm (MWOA) used in this application is a new and improved group intelligence optimization algorithm proposed based on the original wolf pack algorithm. The basic optimization principle of the improved grey wolf algorithm (MWOA) is as follows: the grey wolf population mainly relies on four groups for optimization: α, β, δ and ω. The α wolf is the optimal solution of the wolf pack and is the leader of the wolf pack. They are responsible for making decisions during the hunting process. The β wolf is the second echelon. In the wolf pack optimization group, when the α wolf dies or becomes old, the β wolf becomes the α wolf. The δ wolf pack controls the ω wolf pack and provides information to the α and β wolf packs. The lowest level in the wolf pack hierarchy is the ω wolf pack. During the iterative process, the movement direction of ω is guided by α, β and δ to achieve global optimization. Global optimization can be summarized as: surrounding prey and attacking prey.

[0119] Specifically, the specific optimization steps of the improved grey wolf algorithm (MWOA) using the positions of α, β and δ are as follows:

[0120] Step a-1: Surround the prey

[0121] Formula (1) represents the distance between gray wolf individuals. The update method of offspring gray wolf individuals is expressed as formula (2):

[0122] D=|C·Y p (t c )-Y(t c )| (39)

[0123] Y(t c +1)=Y p (t c)-A·Dist (40)

[0124] Where Dist is the distance between the hunting wolf pack and the target, X p (t c ) is the prey position, t c is the current iteration number, X(t c ) is the current position of the wolf pack, A and C are coefficient control vectors, and the control parameter C is mainly responsible for controlling the gray wolf algorithm's ability to explore unknown space. In the original gray wolf algorithm, the control parameter C is generally set to a multiple of the random parameter, which mainly represents the distance weight from the prey. In order to further enhance the gray wolf algorithm's ability to extrapolate to unknown areas, the improved gray wolf algorithm used in this application, the control parameter C is no longer a fixed random value, but changes dynamically with the number of iterations, emphasizing the favorable effect of C greater than 1 on the distance weight of the prey. Based on this fact, the control parameter C modified by MWOA is calculated according to the following formula:

[0125] A=2×au×r1-au (41)

[0126] C=2×r2-au / 2 (42)

[0127] During the iteration process, the value of au decreases linearly from 2 to 0 as the number of iterations increases. r1 and r2 are random variables and are in the range of [0, 1].

[0128] Step a-2: Capturing the prey

[0129] The position of ω is updated based on the positions of α, β, and δ as follows:

[0130] Dist α =|C1·Y α (t c )-Y(t c )| (43)

[0131] Dist β =|C2·Y β (t c )-Y(t c )| (44)

[0132] Dist δ =|C3·Y δ (t c )-Y(t c )| (45)

[0133] Then, the ω wolf pack updates towards the positions of α, β and δ as follows:

[0134] Y1(t c +1)=Y α(t c )-A1·Dist α (46)

[0135] Y2(t c +1)=Y β (t c )-A2·Dist β (47)

[0136] Y3(t c +1)=Y δ (t c )-A3·Dist δ (48)

[0137] The final position of the offspring wolf pack is:

[0138] Y3=(Y1+Y2+Y3) / 3 (49);

[0139] Swarm intelligence algorithms are widely used to find optimal solutions to complex functions. However, when the fitness function being solved is too complex, a single fitness function evaluation can often take dozens of hours. During the swarm intelligence algorithm's iterative optimization process, a large number of optimizing individuals participate in the fitness evaluation, requiring extensive numerical reanalysis. This can cause the entire solution process to take hundreds of hours, or even days, resulting in the so-called "high computational cost problem." To further reduce the high cost of swarm intelligence solutions, machine learning techniques are often used in conjunction with optimization algorithms for optimization simulations.

[0140] Machine learning is a branch of artificial intelligence, similar to the concept of artificial intelligence. It summarizes and learns from known events to fit and predict unknown events. Compared with traditional polynomial fitting analysis methods, machine learning analysis has a stronger ability to integrate information and is more suitable for deriving appropriate regression equations from large amounts of data. Currently, common machine learning methods include artificial neural networks (ANNs), represented by artificial neurons, and support vector machines (SVMs), represented by kernel learning. However, these two machine learning theories still have some shortcomings. For example, ANNs face challenges in determining the optimal network topology and hyperparameters, the risk of over- and under-learning, and poor generalization to small samples. Furthermore, there is no rigorous theoretical basis for determining the optimal hyperparameters for SVMs, making it difficult to guarantee reliable predictions.

[0141] Based on the above problems, the present application has come up with Gaussian process optimization (GPO), which is an optimization algorithm in local space based on the Gaussian process proxy model (GPR). By constructing a local GPR model, the local optimal solution is then determined from it. Among them, the Gaussian process model (GPR) is a machine learning method based on Bayesian statistics. This method is based on Gaussian random process probability theory, Bayesian statistical theory, kernel function and feature space mapping theory and maximum likelihood estimation theory to determine the mapping function. It can be seen from the specific theoretical basis that compared with the complex and multiple neuron layers of human neural networks, the kernel function machine learning method of Gaussian process can be more conducive to solving and is more suitable for processing complex problems such as high dimension, small sample, and nonlinearity. In addition, in order to solve the problem that hyperparameters in kernel functions are difficult to determine, the Gaussian process machine learning method not only has the advantage of using the variance of the predicted value to measure the uncertainty of the prediction results, but can also adaptively obtain the optimal hyperparameters based on the maximum likelihood method during training. That is, by establishing the partial derivative equation of the log-likelihood function of the conditional probability of the training sample with respect to the hyperparameters, and then using optimization methods such as conjugate gradient to search for the optimal solution of the hyperparameters, a method for adaptively obtaining hyperparameters in calculations is realized, which has strong applicability to highly nonlinear regression problems.

[0142] In this application, after the MWOA algorithm enters the local optimization (GPO) state from the global optimization state, it constructs an implicit proxy model of the Gaussian process (GPR) around the current optimal point, establishes a mapping relationship between unknown parameters and the target fitness function, and predicts the potential optimal point through the optimization calculation formula, which greatly reduces the number of calculations of the fitness function, that is, step S5-2 around the current global optimal point PB gj Accelerate the local space, specifically:

[0143] Step S5-2-1, based on the local acceleration of GPO, establish the proxy model of GPR, and find the global optimal individual PB in step S5-1 gj Nearby, according to distance PB gj The number of points closest to the target is NP×D, and a new learning sample set (X sample , Y sample ), generate a local proxy model;

[0144] Step S5-2-2: Based on the established GPR model, the GPO local acceleration technology is used to quickly find the current local optimal point BL. ij , and BL ij The position information of is input into the numerical model and its fitness FE(BL ij );

[0145] Step S5-2-3, determine the acceleration effect, based on the BL found in step S5-2-2 ij The fitness FE(BL ij ) and the current optimal point PB found in step S5-1-3 gj The fitness FE(PB gj ) are compared, and according to the evaluation results, if FE(BL ij )>FE(PB gj ), prove that the acceleration is successful, then let the global optimal point PB gj =BL ij ; If FE(BL ij ) <FE(PB gj ), then the acceleration is unsuccessful, then PB gj After the judgment is completed, the global optimization phase of the improved grey wolf algorithm MWOA is returned to step S5-1-2. In step S5-1-2, the optimal point PB is selected. gj Continue the iterative process of global optimization and local acceleration until the convergence conditions are met.

[0146] The entire process from building an implicit GPR model to finding the local optimal solution is Gaussian process optimization (GPO). The steps for building a GPR proxy model are as follows:

[0147] Step b-1: Generation process of Gaussian process model:

[0148] The statistical characteristics of the GPR regression model are completely determined by its mean m(X) and covariance function k(X,X / ) to determine the definition, the definition is:

[0149] f(X)~GP(m(X),k(X,X / )) (50)

[0150] Assume there is a training set of n observation data D={(x i ,y i )|i=1,2,,,n},x i For the d-dimensional input vector, the target value y i ∈R, if the matrix X represents the d×n-dimensional input matrix and the vector y represents the output vector, then the training set is abbreviated as D=(X,y); in the actual process, the noise ε is an independent random variable, that is:

[0151] ε~N(0,σ 2 n ) (51)

[0152] Then, with the addition of Gaussian noise to the standard regression model, the prior distribution of the input vector y is:

[0153] y~N(0,K+σ 2 n I) (52)

[0154] Where: K = K(X,X) is an m×m order symmetric positive definite covariance matrix, and any item k in the K matrix ij Measured x i and x j (i, j = 1, 2, ... m) correlation, I is the m × m unit matrix;

[0155] Step b-2: Gaussian process model fitting new sample points:

[0156] For a new test input matrix X * , GP model predicts the same * The corresponding output vector y * , then the output vector y of the training sample and the test output vector f * The resulting joint distribution Gaussian prior distribution is:

[0157]

[0158] Where: K=K(X,X * ) is the test input matrix X * The m×1 order covariance matrix with the input training sample matrix X can be abbreviated as K(X * ), K(X * ,X * ) is the test point input matrix X * The covariance matrix of itself can be abbreviated as k(X * );

[0159] The mean and variance of the function's predicted values are:

[0160]

[0161]

[0162] In the above formula, It is the fitting regression model based on the training sample points, where the individual position coordinates X are implicitly included. * With fitness f * The corresponding relationship is used to replace the mapping function curve between the dynamic parameters and the target adaptation function;

[0163] In the GPR agent model, the optimal solution of the covariance function hyperparameter θ is obtained by maximizing the marginal likelihood p(y i |X i ,θ) and adaptively obtained; specifically, by taking the negative logarithm -log(p(yi |X i ,θ)), transform the maximization problem into a minimization problem, and then use the conjugate gradient descent method to achieve the optimal hyperparameter θ lI Adaptive acquisition of

[0164] Step b-3: Local accelerated optimization process based on the agent model:

[0165] Based on the fitting model established in formula (17), new samples can be obtained by the extreme value of the convergence function, which is referred to as the LCB criterion:

[0166]

[0167] in: kσ are the mean and variance of the function prediction value, and the calculation formula of the constant k expressing the confidence range is

[0168] k=3-2.9×i / M (57)

[0169] Among them, i is the current iteration number iter, and M is the maximum iteration number Maxiter.

[0170] Step S6: Repeated iteration of global optimization and local optimization until the convergence iteration criterion meets the adaptation accuracy and the maximum allowed number of iterations, calculates the mechanical parameters corresponding to the minimum target value, exits the optimization process, and outputs the optimal structural dynamic parameters. The convergence iteration criterion includes the maximum total number of algorithm fitness function calculations ITR max , which is also the maximum number of finite element calls; the convergence value error of the algorithm, the calculated objective function min f(x) satisfies the convergence value, which is generally set to 1e -3 .

[0171] As an example, the method for interoperating between the algorithm and the finite element numerical calculation software during the optimal parameter solution process includes: the main iterative optimization process of the MWOA-GPO algorithm is executed in a command stream environment; during the global optimization process, the population position information required for the MWOA-GPO optimization process (corresponding to a set of structural dynamic parameters) is saved in file A; the finite element model is called through interactive commands to read the parameter file; the dynamic response of the protective structure under the blast impact under this set of dynamic parameters is calculated; and the calculated dynamic response value (structural deformation displacement, structural overpressure value, structural acceleration value) is then stored in interface file B, thereby obtaining the fitness value of the individual. After iteration, the objective function value is calculated to determine whether it meets the convergence level; otherwise, the iterative optimization process will continue until the convergence condition is met.

[0172] Then the present application also provides an identification device for the method of identifying the impact dynamic parameters of the large explosion-proof structure, such as Figure 3 As shown, it includes a geometric model M1, a dynamic response monitoring module M2, a numerical model module M3, a parameter calculation module M4, a storage and transmission module M5 and a control and display module M6 that are electrically connected to each other;

[0173] The geometric model M1 is used to measure the geometric model parameters of the explosion-proof protection structure. Common geometric information includes: structure length change, height change, thickness change, etc.; the geometric physical information of the measured structure will be used in the finite element generation module M3.

[0174] The dynamic response monitoring module M2 fits the numerical model with the real structure, that is, uses the numerical model to approximate the real structure. The reference standard of the numerical approximation is the real response value of the structure under a real explosion. It includes a structural damage and deformation measurement unit M2-1 for measuring the damage and deformation displacement values of key points of the large structure under the explosion impact load, a structural overpressure measurement unit M2-2 for measuring the overpressure values of key points of the large structure under the explosion impact load, and a structural acceleration measurement unit M2-3 for measuring the vibration acceleration values of key points of the large structure under the explosion impact load.

[0175] The numerical model module M3 includes an initial model unit M3-1 and a model iteration update unit M3-2. The initial model unit M3-1, before dynamically optimizing the dynamic parameters, establishes initial basic model information based on the finite element information in the geometry module M1. The numerical model established in the initial model unit M3-1 is based on the initial optimization parameters. Based on the measured geometric parameters of the structure in the geometry module M1, a finite element numerical model consistent with the actual structure is established. The numerical model module M3 and the parameter calculation module M4 provide feedback to each other. In the initial state, the numerical model module M3 provides the parameter calculation module M4 with a numerical model of the initial parameters and calculation results. Based on the dynamic response value of the structure under the impact load of the explosion in the dynamic response monitoring numerical model module M3, the dynamic parameter calculation results of the MWOA-GPO algorithm in the parameter calculation module M4 are used to update the dynamic parameters of the material, thereby updating the numerical model of the structure. The dynamic parameters obtained in each iteration are fed into the model iteration update unit M3-2 to calculate the fitness.

[0176] The parameter calculation module M4 includes a structural parameter initialization unit M4-1 for setting the initial solution interval of the structure, a MWOA calculation unit M4-2 for solving the dynamic parameters of the structure in the global space using the MWOA algorithm, and a GPR acceleration unit M4-3 for using the GPR model to accelerate the solution speed of the MWOA in the local space; the structural parameter initialization unit M4-1 is used to determine the required parameters of the MWOA algorithm according to the actual situation on site, including the following items: the search interval of the MWOA algorithm population, the number of MWOA algorithm population searches, etc. (in the embodiment, including the number of gray wolf populations NP, the dimension D of the algorithm, the search interval [a, b, ..., n] of the MWOA algorithm, the convergence criterion ε of the algorithm, and the number of global optimization searches T max , local optimization number startup time Iter max ); MWOA calculation unit M4-2, used to perform global optimization according to the population update strategy of the MWOA algorithm before local acceleration is started according to the parameter setting structure; GPO acceleration unit M4-3, used to use GPO local optimization technology in the local space according to the actual situation on site to accelerate the solution speed of MWOA in the local space, wherein the GPO parameter setting unit includes: the start time Iter from the MWOA global state to the GPO local acceleration max , the optimal hyperparameters for GPR machine learning, and the selection of the GPR model kernel function. The MWOA-GPO algorithm is used to solve the optimal structural dynamic parameters in the global space. The numerical model calculations based on this set of parameters are highly consistent with the actual dynamic response of the structure under blast loads. The numerical model parameters corresponding to this numerical model are the actual dynamic parameters of the structure.

[0177] The storage and transmission module M5 is used to store model calculation data and communicate data between modules. It includes a local storage unit M5-1 for storing the calculated data in a local data center centralized device, a backup storage unit M5-2 for backing up the data and storing it in the cloud, and a transmission unit M5-3 for sending the calculated data to a receiving platform or a corresponding client via a universal wireless, wired, or selective transmission method.

[0178] The control and display module M6 is used to display the computational progress between modules and to provide manual control over the computational progress. It includes an information display unit M6-1 for displaying various types of information and a time control unit M6-2. The entire model operates under automated instructions, but allows for interactive control by administrators. Real-time computational control is possible based on administrator needs. Administrators can pause the process and enter any module to direct the progress. The control and display module M6 displays the parameter settings and model generation for each module, facilitating real-time management of the computational process.

[0179] Example 1:

[0180] The geometric shape of the complex explosion-proof structure measured in Example 1 is as follows Figure 4 As shown. The structure of this example is a double-layer composite structure, and the main material used is reinforced concrete. In reinforced concrete, the main necessary materials that determine the explosion impact resistance of the structure include steel bars and concrete. According to actual needs, the impact dynamic parameters of the large explosion-proof structure are determined in step S1, mainly including the dynamic tensile modulus DT of the longitudinal steel bars (abbreviated as: longitudinal bars), the dynamic enhancement coefficient DS of the longitudinal bars, the dynamic friction angle DΦ1 of the concrete, the dynamic compressive strength DE of the concrete, and the dynamic enhancement coefficient DF of the concrete.

[0181] The optimization range and initial values of the dynamic parameters to be solved are shown in Table 1 below.

[0182] Table 1 Optimization interval and initial value of dynamic parameters to be solved

[0183] parameter DT(MPa) DS <![CDATA[DΦ1(rad)]]> DE(MPa) DF Optimization interval [500,800] [0,3] [0,3] [50,200] [0,3] Initial value 600 1 1 80 1

[0184] Step S2, the geometric shape of the complex explosion-proof structure measured in this Example 1 is as follows Figure 4 As shown, the main measured geometric parameters include: height H i , length B i , width C i The main geometric and physical parameters of the measured structures are shown in Table 2.

[0185] Table 2 Geometric parameters of the structure

[0186] H1(m) B1(m) C1(m) B2(m) C2(m) 23.26 24.98 6.12 20.28 5.2

[0187] Step S3, measuring the dynamic response of the structure to be identified under the impact of the explosion. In order to measure the dynamic response of the explosion-resistant structure under the explosive load, this example uses the explosion impact test of the structure under the impact of 5kg TNT explosive, with the distance between the explosive and the structure being 0.2m. The main monitoring dynamic response contents include: the dynamic displacement DW of the structure at the key point under the explosive load i (x), overpressure value KP of the structure at the key point i (x), the acceleration value AC of the structure at the key point i (x),i is the arrangement of key points on the structure. Figure 5 As shown, it includes DW1, DW2, DW3, DW4, DW5, KP1, KP2, KP3, KP4, AC1, AC2, and x is the structural dynamic parameter of this group of structures.

[0188] Step S4: Establish a numerical model of the structure to be identified under the impact of the explosion. Step S1 determines the dynamic mechanical parameters that need to be calculated, and in step S2, the structural geometry physics is used to establish the numerical model of the structure. In this Example 1, a finite element numerical model was established using the commercial ABAQUS numerical model software as the calculation medium to simulate the dynamic response of the structure under the explosion of 5kg TNT. In the subsequent step S5, the numerical model is embedded in the Python command stream of MWOA-GPO to perform calculations and find a suitable set of structural dynamic parameters.

[0189] In the embodiment, based on the dynamic response value calculated by the numerical model and the measured geometric model of the structure in step S2, the initial finite element numerical model of the structure is established according to the initial values of the dynamic mechanics in Table 1 in step S1. The numerical model in this embodiment 1 is as follows: Figure 6 shown.

[0190] Then, according to the dynamic response values of the structure measured in step S3: displacement, acceleration and overpressure value, the optimization objective function established is DW(x) is the “measured displacement” of the key points of the explosion-resistant structure under the actual explosion load in step S3. is the “numerical displacement” of the key point of the explosion rush calculated on the numerical platform by the finite element numerical model updated by each set of optimal position parameters; KP(x) is the “measured overpressure” of the key point of the explosion-resistant structure under the actual explosion load in step S3, is the “numerical overpressure” of the key points of the explosion rush calculated on the numerical platform by the finite element numerical model updated by each set of optimal position parameters; AC(x) is the “measured acceleration” of the key points of the explosion-resistant structure under the actual explosion load in step S3, It is the "numerical acceleration" of the key point of the explosion rush calculated on the numerical platform by the finite element numerical model updated by each set of optimal position parameters; when the value of the objective function meets the convergence condition of the objective function min f(x), it can be considered that the numerical model and the real structure have good consistency. In this case, the dynamic parameter values in the numerical model are the dynamic parameters in the real structure.

[0191] Step S5, first in step S5-1: set the MWOA algorithm search parameters, including: search population number NP = 80, algorithm dimension D = 5, MWOA algorithm optimization times Iter max =50 Enter the GPO local optimization stage;

[0192] Then, in step S5-1, according to the optimization parameters set in step S5-1, the MWOA algorithm is started to perform global optimization, and the position parameters are updated according to the optimization step S5-1-2 of the MWOA algorithm. After the wolf pack optimization strategy of the MWOA algorithm is optimized, the new position of the individual and the position information of the 400 sets of structural dynamic parameters are stored in the data interface file Position. This interface file contains the optimal structural dynamic parameters found by the current iterative optimization; the numerical model of the structure in ABAQUS is called through the PYTHON command flow program to calculate the "numerical response values" such as the structural deformation displacement, structural overpressure peak value and structural acceleration value under the impact of the explosion load under this set of dynamic parameters and calculate the "numerical response values" The objective function value min f(x) between the “field monitoring response value” DW(x), KP(x), and AC(x) is calculated. The optimal individual corresponding to the minimum fitness value FE among all particles in the iterative process is recorded as the current global optimal individual PB. gj , the corresponding objective function value BFE(PB gj ) is the current global optimal solution. When the number of global optimization iterations reaches t=Iter max Then, the optimization phase of GPR is entered. In the next step S5-2, around the current optimal point PB gj Accelerate the local space.

[0193] Then, according to step S5-2, enter the local optimization stage, at the current optimal point BFE (PB gj ) to build a local proxy model near the current optimal point PB found in step S5-1. gj Nearby, a local proxy model is constructed according to the distance. The current population size of the GPR model is 80×3=240, and the sample set D=(X sample , Y sample );

[0194] The current optimal point BL found based on the GPR model ij , follow the steps of GPO optimization algorithm, and also set BL ij The dynamic parameters in the numerical model in ABAQUS are dynamically updated, and the "numerical response values" FE (BL) such as the structural deformation displacement, structural overpressure peak and structural acceleration value under the impact of the explosion load are calculated under this set of dynamic parameters. ij ).

[0195] In step S5-2, it is necessary to judge whether the acceleration effect is successful or not. ij The fitness FE(BL ij) and the current optimal point PB found in step S5-1-3 gj The fitness FE(PB gj ) are compared, and according to the evaluation results, if FE(BL ij )>FE(PB gj ), prove that the acceleration is successful, then let the global optimal point PB gj =BL ij ; If FE(BL ij ) <FE(PB gj ), then the acceleration is unsuccessful, then PB gj After the judgment is completed, the global optimization phase of the improved gray wolf algorithm MWOA is returned to step S5-1-2. In step S5-1-2, the current optimal point PB is gj Continue the iterative process of global optimization and local acceleration until the convergence condition is met;

[0196] The above step S5 is an iterative optimization process, which continues until step S6, until the algorithm's objective function convergence condition ε = min f(x) = 1 × 10 -3 and the maximum number of calculations T of the global numerical model max =300, the calculation stops and the position coordinates of the individual (i.e., the optimal values of the dynamic mechanical parameters of the structure) are output.

[0197] The dynamic mechanical parameters of the explosion-proof protective structure in this embodiment include: the dynamic tensile strength DT of the longitudinal steel bars (abbreviated as: longitudinal bars), the dynamic enhancement coefficient DS of the longitudinal bars, the dynamic friction angle DΦ1 of the concrete, the dynamic compressive strength DE of the concrete, and the dynamic enhancement coefficient DF of the concrete.

[0198] After the above algorithm optimization steps, the final values of the above parameter inversion are shown in Table 3 below.

[0199] Table 3 Comparison of dynamic parameter values obtained by the method of the present invention and experimental values

[0200] parameter DT(MPa) DS <![CDATA[DΦ1(rad)]]> DE(MPa) DF MWOA-GPO calculated value 700 1.54 0.87 100 1.25 Mechanical experimental test value 706 1.49 0.83 101 1.26 Relative error 0.84％ 3.2％ 4.5％ 1％ 0.8％

[0201] Table 3 lists the experimental values of the structure and the original mechanical experiments (uniaxial loading, stamping loading and other mechanical experiments), and compares the relative errors between the two. It can be seen from Table 3 that the relative errors between the identification results and the mechanical experimental values are small.

[0202] In addition, based on the structural dynamic parameters in Table 3, the measured key point displacements calculated by substituting this set of parameters into the established numerical model are shown in Table 4 below.

[0203] Table 4 Comparison of measured dynamic response values and numerical calculation values at key points

[0204]

[0205] It can be seen from Table 4 that the differences between the measured values and the numerical calculation values of each group are small. Under the calculation of this group of parameters, the number of calculations of the numerical model reached 180 times, and the final convergence value of the objective function min f(x) was 9×10 -4 , which satisfies the convergence criterion.

[0206] At the same time, according to the above steps, the traditional optimization particle swarm optimization algorithm (PSO), whale algorithm (HOA), and locust algorithm (GOA) are used to invert the same set of mechanical parameters at the same time, and the time consumption of the algorithm is calculated.

[0207] Table 5 Comparison of algorithm calculation time

[0208] algorithm PSO HOAs GOA MWOA-GPO Number of numerical model calls 92 71 66 21 Calculation time (h) 27.68 21.36 20.14 9.62

[0209] It can be seen from Table 5 that the time consumption for calculating the dynamic parameters of the structure in this application is the lowest, which has significant advantages.

[0210] Example 2:

[0211] In order to achieve the purpose of the above-mentioned embodiment 1, the present application provides an identification device for a method for identifying impact dynamic parameters of a large explosion-proof structure. Figure 7 The figure shows a schematic diagram of the device layout for acquiring structural dynamic parameters provided in the present application, including an electrically connected geometric model M1, a dynamic response monitoring module M2 (structural damage and deformation measurement unit M2-1, structural overpressure measurement unit M2-2, and structural acceleration measurement unit M2-3), a numerical model module M3, a parameter calculation module M4, a storage and transmission module M5, and a control and display module M6.

[0212] The numerical model module M3 is used in this application to match the measured deformation of the structure in the dynamic response monitoring module M2, to receive the structure size information in the geometry module M1, to construct a numerical model of the structure under the explosion impact, and then interactively connect to the parameter calculation module M4.

[0213] The parameter calculation module M4 includes both user-defined and default parameter values, allowing users to select parameters as needed. This improves the module's operability, eliminating the need for extensive algorithm experience. This improves the method's adaptability to a broad user base and significantly reduces computation time.

[0214] The storage and transmission module M5 transmits the calculated key mechanical parameters, structural explosion load deformation values, and other necessary information wirelessly, wired, or in a hybrid manner, depending on the response needs of each module. It can be used to transmit information between all module devices. The local storage unit M5-1 is used to store the calculated data in a centralized device in the local data center. It will store the last three months of data according to instructions. The local storage unit will regularly clear the stored data to improve efficiency and storage space. The backup storage unit M5-2 is used to back up the data and store it in the cloud. The transmission unit M5-3 is used to send the calculated data to the receiving platform or the corresponding client via universal wireless, wired, or selective transmission methods. The storage function between the local storage unit M5-1 and the backup storage unit M5-2, and the transmission function of the transmission unit M5-3, respectively, operate under two independent instructions without interfering with each other, aiming to improve the independence of information storage and transmission.

[0215] The storage medium includes, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination thereof, including but not limited to portable computer disks, hard disks, random access memories, cloud storage platforms, etc. Wired transmission devices include: optical fibers, cables, circuits, etc.; wireless transmission devices include Wi-Fi transmission, bridges, or long-distance 5G communications, etc., or a combination of one or more of these.

[0216] The control display module M6 can be displayed on a variety of current clients, including smartphones, laptops, and desktop computers. A dedicated process control program is included on the client, allowing administrators to access the device's computational process through client commands and recalculate or set parameters based on user experience needs. The computational progress of the aforementioned geometry, monitoring, numerical, and calculation modules is displayed on the display module. The time control unit M6-2 allows real-time observation of the calculated data and models, facilitating personnel management and improving monitoring standardization and model fault tolerance.

[0217] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as such herein, will not be interpreted in an idealized or overly formal sense.

[0218] The meaning of "and / or" in this application means that both situations where each exists alone or both exist at the same time are included.

[0219] The term “connection” as used in this application may mean a direct connection between components or an indirect connection between components via other components.

[0220] With the above-described preferred embodiments of the present invention as a guide, and with reference to the above description, relevant personnel are fully capable of making various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the contents of the specification and must be determined according to the scope of the claims.

Claims

1. A method for identifying impact dynamic parameters of large explosion-resistant structures, characterized by: The following steps are involved: Step S1: determining the impact dynamic parameters of the structure to be identified; Step S2: Determine the geometric parameters of the structure to be identified; Step S3: measuring the dynamic response of the structure to be identified under the impact of the explosion; Step S4: establishing a numerical model of the structure to be identified under the impact of the explosion and constructing the objective function; Step S5: Using the improved MWOA-GPO algorithm to quickly optimize in the parameter interval, that is, using the improved gray wolf algorithm to optimize in the global space. When the number of global optimization reaches the preset number of iterations, the optimization in the global space is transformed into the optimization in the local space. At this stage, the GPO local optimization technology is started; Step S6: Repeated iteration of global optimization and local optimization until the convergence iteration criterion meets the adaptation accuracy and the maximum allowed number of iterations, calculates the mechanical parameters corresponding to the minimum target value, exits the optimization process, and outputs the optimal structural dynamic parameters; The impact dynamic parameters of the structure to be identified determined in step S1 include the dynamic mechanical parameters of a single material if the structure is a single-medium structure, and the dynamic mechanical parameters of the materials of each component if the structure is a multi-medium composite structure; The aforementioned material dynamic mechanical parameters include determining the dynamic parameters required in actual engineering design, the initial values of unknown parameters, and the upper and lower limits of the preset parameter search range; In step S2, the geometric parameters of the structure to be identified are measured using a contact or non-contact measurement method; In step S3, the dynamic response of the structure to be identified under the impact of the explosion is the measured dynamic response of the key points of the structure. Including the impact damage deformation value W of the structure kp1 、W kp2 ,...,W kpn , the impact overpressure value P of the structure kp1 、P kp2 ,...,P kpn And the acceleration value A of the structure kp1 、A kp2 ,...,A kpn ; In step S4, the geometric parameters of the structure to be identified obtained in step S2 are combined with the initial values of the position parameters in step S1 to establish an initial numerical model of the structure; The constructed objective function is Among them, x is a set of material dynamic mechanical parameters, S i is the dynamic response value of the i-th key monitoring point on the structure after the explosion, is the calculated dynamic response value of the finite element at the i-th displacement monitoring point under this set of mechanical parameters; When the value of the objective function satisfies the convergence condition of the objective function minf(x), the dynamic parameter values in the numerical model are the dynamic parameters in the real structure; The specific steps of step S5 are: Step S5-1, performing iterative optimization of the improved grey wolf algorithm in the global space; Step S5-2, dynamic acceleration of GPO in local space; Step S5-1 is specifically as follows: Step S5-1-1, set the parameters of the MWOA algorithm, including the number of gray wolf populations to be searched NP, the dimension of the algorithm D, and the number of local optimization times and the start time Iter max , when the MWOA algorithm reaches the fitness evaluation times t in the single cycle stage, max When , it enters the GPO local acceleration stage; Step S5-1-2, before each iterative optimization of the MWOA algorithm, calculate the fitness value of each wolf in the search population, and save the position information of the wolf with the best fitness value in the population as α, the position information of the wolf with the second best fitness value in the population as β, the position information of the gray wolf with the third best fitness value in the population as δ, and the position information of the remaining wolves as ω; when the four classes in the wolf pack are determined, start the optimization in the global space, and update the position of the next generation of wolf pack particles according to the iterative formula of the improved gray wolf algorithm. When the wolf population calculation is updated in each iterative update, the number of fitness calculations calculated numerically at this time is t=t+1; Step S5-1-3, in the optimization process, the potential optimal particle swarm needs to be brought into the real fitness function to calculate the real fitness value, that is, according to the position I of the optimal individual of the current gray wolf population ij In the numerical model established in the dynamic update step S4, the current parameter I is calculated according to the updated numerical model. ij The fitness value FE(I ij ), where the population composed of particles with the best fitness in each group is recorded as QBE(i), and the best individual corresponding to the minimum fitness value FE among all particles is recorded as the current global best individual PB gj , the corresponding objective function value BFE(PB gj ) is the current global optimal solution; the number of calculations of the numerical model reaches t = Iter max After that, it enters the GPO acceleration phase; Step S5-2 is specifically as follows: Step S5-2-1, establish a Gaussian process agent model, and find the global optimal individual PB in step S5-1 gj Nearby, according to distance PB gj The number of points closest to the target is NP×D, and a new learning sample set (X sample , Y sample ), generate a local proxy model; Step S5-2-2: Based on the established GPR model, the GPO local acceleration technology is used to quickly find the current local optimal point BL. ij , and BL ij The position information of is input into the numerical model and its fitness FE(BL ij ); Step S5-2-3, based on the BL found in step S5-2-2 ij The fitness FE(BL ij ) and the current optimal point PB found in step S5-1-3 gj The fitness FE(PB gj ) are compared, and according to the evaluation results, if FE(BL ij )>FE(PB gj ), prove that the acceleration is successful, then let the global optimal point PB gj =BL ij ; If FE(BL ij )<FE(PB gj ), then the acceleration is unsuccessful, then PB gj After the judgment is completed, the global optimization phase of the improved grey wolf algorithm MWOA is returned to step S5-1-2. In step S5-1-2, the optimal point PB is selected. gj Continue the iterative process of global optimization and local acceleration until the convergence conditions are met.

2. The method for identifying impact dynamic parameters of a large explosion-resistant structure according to claim 1, characterized in that: The optimization process of the MWOA algorithm in the global space in step S5-1-2 includes: Step S5-1-2-1, wolf pack initial position information P ij =I ij , the first generation population P ij Substitute into the numerical model built in step S4 to perform fitness evaluation and obtain the first generation fitness value FE of all individuals 1st (P ij ), the particle P with the smallest fitness value in the current set xj is the current global optimal grasshopper, and its corresponding objective function value FE 1st (P xj ) is the current first-generation global optimal solution; Step S5-1-2-2, update the search particle position according to the improved wolf pack algorithm MWOA algorithm optimization formula to obtain the next search population position P nextij ; Step S5-1-2-3, update the current wolf pack position P nextij Bring it into step S5-1-3 to perform fitness evaluation and obtain the objective function value FE (P nextij ); Current P nextij The wolf individual N with the smallest fitness function value mingj is the current global optimal wolf pack individual, and its corresponding objective function value FE(N mingj ) is the current global optimal solution; Step S5-1-2-4, based on the current wolf pack position P nextij , repeat the above steps, at this time the number of iterations t=t+1, after several iterations, the number of global optimization iterations t=Iter max , entering the GPO acceleration phase.

3. The method for identifying impact dynamic parameters of a large explosion-resistant structure according to claim 2, characterized in that: The specific optimization steps of the improved grey wolf algorithm using the positions of α, β and δ are as follows: Step a-1: Surround the prey Formula (1) represents the distance between gray wolf individuals. The update method of offspring gray wolf individuals is expressed as formula (2): D=|C·Y p (t c )-Y(t c )| (1) Y(t c +1)=Y p (t c )-A·Dist (2) Where Dist is the distance between the hunting wolf pack and the target, X p (t c ) is the prey position, t c is the current iteration number, X(t c ) is the current position of the wolf pack, A and C are coefficient control vectors, and the control parameter C is mainly responsible for controlling the gray wolf algorithm's ability to explore the unknown space. The control parameter C is not a fixed random value, but changes dynamically with the number of iterations. The control parameter C modified by MWOA is calculated according to the following formula: A=2×au×r1-au (3) C=2×r2-au / 2 (4) During the iteration process, the value of au decreases linearly from 2 to 0 as the number of iterations increases. r1 and r2 are random variables and are in the range of [0, 1]. Step a-2: Capturing the prey The position of ω is updated based on the positions of α, β, and δ as follows: Dist α =|C1·Y a (t c )-Y(t c )| (5) Dist β =|C2·Y β (t c )-Y(t c )| (6) Dist δ =|C3·Y δ (t c )-Y(t c )| (7) Then, the ω wolf pack updates towards the positions of α, β and δ as follows: Y1(t c +1)=Y a (t c )-A1·Dist a (8) Y2(t c +1)=Y β (t c )-A2·Dist β (9) Y3(t c +1)=Y δ (t c )-A3·Dist δ (10) The final position of the offspring wolf pack is: Y3=(Y1+Y2+Y3) / 3 (11); The technical steps for GPO acceleration are as follows: Step b-1: Generation process of Gaussian process model: The statistical characteristics of the Gaussian process regression model are completely determined by its mean m(X) and covariance function k(X,X′), which are defined as follows: f(X)~GP(m(X),k(X,X′)) (12) Assume there is a training set of n observation data D={(x i ,y i )|i=1,2,,,n},x i For the d-dimensional input vector, the target value y i ∈R, if the matrix X represents the d×n-dimensional input matrix and the vector y represents the output vector, then the training set is abbreviated as D=(X, y); in the actual process, the noise ε is an independent random variable, that is: ε~N(0,σ 2 n ) (13) Then, with the addition of Gaussian noise to the standard regression model, the prior distribution of the input vector y is: y~N(0,K+σ 2 n I) (14) Where: K = K(X, X) is an m×m order symmetric positive definite covariance matrix, and any item k in the K matrix ij Measured x i and x j (i, j = 1, 2, ... m) correlation, I is the m × m order identity matrix; Step b-2: Gaussian process model fitting new sample points: For a new test input matrix X ＊ , GP model predicts the same * The corresponding output vector y ＊ , then the output vector y of the training sample and the test output vector f * The resulting joint distribution Gaussian prior distribution is: Where: K = K(X, X ＊ ) is the test input matrix X ＊ The m×1 order covariance matrix with the input training sample matrix X can be abbreviated as K(X ＊ ), K(X ＊ , X ＊ ) is the test point input matrix X ＊ The covariance matrix of itself can be abbreviated as k(X ＊ ); The mean and variance of the function's predicted values are: In the above formula, It is the fitting regression model based on the training sample points, where the individual position coordinates X are implicitly included. ＊ With fitness f * The corresponding relationship is used to replace the mapping function curve between the dynamic parameters and the target adaptation function; Step b-3: Local accelerated optimization process based on the agent model: Based on the fitting model established in formula (17), new samples can be obtained by the extreme value of the convergence function, which is referred to as the LCB criterion: in: kσ are the mean and variance of the function prediction value, and the calculation formula of the constant k expressing the confidence range is k=3-2.9×i / M (19) Among them, i is the current iteration number iter, and M is the maximum iteration number Maxiter.

4. The method for identifying impact dynamic parameters of a large explosion-resistant structure according to claim 3, wherein: The convergence iteration criteria include the maximum total number of algorithm fitness function calculations ITR max , which is also the maximum number of finite element calls; the convergence value error of the algorithm, the calculated objective function min f(x) satisfies the convergence value, is set to 1e -3 .

5. An identification system for the method for identifying impact dynamic parameters of large explosion-resistant structures according to claim 3, characterized in that: It includes a geometric model M1, a dynamic response monitoring module M2, a numerical model module M3, a parameter calculation module M4, a storage and transmission module M5 and a control and display module M6 that are electrically connected to each other; The dynamic response monitoring module M2 includes a structural damage and deformation measurement unit M2-1 for measuring the damage and deformation displacement value of key points of the large structure under the impact load of the explosion, a structural overpressure measurement unit M2-2 for measuring the overpressure value of key points of the large structure under the impact load of the explosion, and a structural acceleration measurement unit M2-3 for measuring the vibration acceleration value of key points of the large structure under the impact load of the explosion; The numerical model module M3 includes an initial model unit M3-1 and a model iterative update unit M3-2; The parameter calculation module M4 includes a structural parameter initialization unit M4-1 for setting the initial solution interval of the structure, a MWOA calculation unit M4-2 for solving the dynamic parameters of the structure in the global space using the MWOA algorithm, and a GPR acceleration unit M4-3 for accelerating the solution speed of the MWOA in the local space using the GPR model; The storage and transmission module M5 includes a local storage unit M5-1 for storing the calculated data in a local data center centralized device, a backup storage unit M5-2 for backing up the data and storing it in the cloud, and a transmission unit M5-3 for sending the calculated data to a receiving platform or a corresponding client via a universal wireless, wired, or selective transmission method; The control and display module M6 includes an information display unit M6-1 for displaying various types of information and a time control unit M6-2 that can enter any module to direct the progress under the administrator's instruction.

Images