Structure optimization design method and system for combined impact of anti-detonation wave and fragment group

Abstract

The present application relates to the field of impact-resistant structure optimization design, in particular to a structure optimization design method and system for resisting combined impact of blast wave and fragment group. The method comprises: obtaining a structure design scheme for resisting combined impact of blast wave and fragment group, performing first numerical analysis on the structure design scheme to obtain a sample set; constructing a classification surrogate model, training the classification surrogate model by using the sample set; introducing a genetic algorithm, improving fitness evaluation mechanism of the genetic algorithm by using the classification surrogate model; selecting parents for iteration of the genetic algorithm based on the improved fitness evaluation mechanism to obtain an improved genetic algorithm; adopting a hybrid point addition strategy to optimize selection of sample points of the classification surrogate model and iteration process of the improved genetic algorithm, and determining an optimal structure design scheme for resisting combined impact of blast wave and fragment group. The method solves the problems of insufficient optimization precision and high calculation cost faced by using a single point addition criterion.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of impact-resistant structure optimization design, in particular to a structure optimization design method and system for resisting combined impact of blast wave and fragment cloud. BACKGROUND

[0002] The combined damage effect of blast wave and high-speed fragment cloud on a structure is significantly different from the effect of either of them alone, and under certain conditions, it has a cumulative enhancement effect. Therefore, the design of new protective structures needs to consider the combined damage effect of blast wave and fragment cloud. At present, the design strategies for protective structures mainly include structure design based on traditional metal materials and development of new composite protective structures using multi-phase materials to improve the protective performance of the structure against combined loads. However, the above structure design relies mainly on engineering experience and lacks the support of mathematical optimization methods, making it difficult to fully tap the potential of the structure performance. The optimization of structures against combined impact involves strong nonlinearity of materials, geometry, and component contact under the combined load of blast wave and fragment cloud, resulting in time-consuming structure response solving and difficulty in sensitivity solving, which poses great challenges to the optimization design of structures.

[0003] Surrogate-based optimization (SBO) replaces time-consuming numerical analysis in the optimization process by constructing a direct mapping between input and output, and combines with the corresponding intelligent optimization algorithm to realize the optimization design of structures. This method has been successfully applied to the optimization design of structure parameters against blast wave and fragment / pellet single load, and has achieved certain results. The commonly used surrogate models in SBO include polynomial response surface, Kriging model, radial basis function neural network, and support vector machine. Among them, the classification surrogate model based on support vector machine does not need to predict the specific value of the mechanical response of the structure under extreme load, and has the advantage of lower sample set calculation cost.

[0004] The accuracy of the surrogate model has a great influence on the optimization result. In the "one-step" optimization design framework, that is, the surrogate model is directly established using the samples obtained by one sampling for optimization and solving, and no point adding process is involved. Therefore, the accuracy of the surrogate model is often difficult to guarantee, and the optimization result is biased. In the process of structure optimization design, the study on the accuracy of the surrogate model is also of great significance. Therefore, it is urgent to develop an effective optimization design method for structures against combined impact. SUMMARY

[0005] In view of the defects in the prior art, the present application provides a structure optimization design method and system for resisting combined impact of blast wave and fragment cloud, which solves the problems of insufficient optimization precision and high calculation cost faced by single point adding criterion.

[0006] In order to achieve the above object, one aspect of the present application provides a structure optimization design method for resisting combined impact of blast wave and fragment group, the method comprising: obtaining a structure design scheme for resisting combined impact of blast wave and fragment group, performing first numerical analysis on the structure design scheme to obtain a sample set; constructing a classification surrogate model, training the classification surrogate model by using the sample set; introducing a genetic algorithm, improving fitness evaluation mechanism of the genetic algorithm by using the classification surrogate model; based on the improved fitness evaluation mechanism, selecting parents for iteration of the genetic algorithm to obtain an improved genetic algorithm; adopting a hybrid sampling strategy to optimize selection of sample points of the classification surrogate model and iteration process of the improved genetic algorithm, and determining an optimal structure design scheme for resisting combined impact of blast wave and fragment group.

[0007] The present application can accurately simulate the response of the structure under the combined impact of blast wave and fragment group by obtaining a sample set through numerical analysis and training a classification surrogate model, and can provide reliable basis for design. The improved fitness evaluation mechanism of the genetic algorithm enables the algorithm to more efficiently screen out individuals adapted to the environment and avoid falling into local optimal solution. The hybrid sampling strategy optimizes sample point selection and algorithm iteration, further explores the design space, improves optimization quality, accelerates convergence speed, and reduces calculation cost.

[0008] Optionally, the improving the fitness evaluation mechanism of the genetic algorithm by using the classification surrogate model comprises: dividing individuals in a population of the genetic algorithm into feasible solutions and infeasible solutions by using the classification surrogate model; calculating diversity difference between the infeasible solutions and the individuals in the population to obtain infeasible solution diversity difference values; and improving the fitness evaluation mechanism of the genetic algorithm according to the infeasible solution diversity difference values.

[0009] The present application uses a classification surrogate model for prediction, providing data support for evaluating the population, and can quickly distinguish between feasible solutions and infeasible solutions. The infeasible solution diversity difference values obtained by calculating the diversity difference between the infeasible solutions and the population introduce a new dimension for fitness evaluation, avoiding premature convergence of the algorithm to a local optimal solution. Finally, the fitness evaluation mechanism is improved according to the difference values, making the genetic algorithm more in line with the target requirements of the present application, while improving the scientificity and flexibility of the genetic algorithm in the iteration process, enhancing its global search ability, and effectively improving the search efficiency.

[0010] Optionally, the improving the fitness evaluation mechanism of the genetic algorithm according to the infeasible solution diversity difference values comprises: calculating a target function value of the improved genetic algorithm according to the infeasible solution diversity difference values; sorting individuals in the population according to the target function value to obtain a sorting position; and calculating fitness values of the individuals in the population according to the sorting position to obtain population fitness values.

[0011] The application calculates the objective function by the diversity difference value of infeasible solutions, so that the objective function is more in line with the actual problem and the exploration ability of the solution space is enhanced. The population individuals are sorted based on the objective function value, which can intuitively distinguish the advantages and disadvantages of the individuals and provide clear guidance for subsequent operations. The fitness value is calculated according to the sorting position, which avoids the problem of excessive or insufficient selection pressure caused by directly using the objective function value, and reasonably balances the population diversity and convergence speed. This not only prevents the algorithm from converging too early, but also speeds up the process of approaching the optimal solution, so that the algorithm can more efficiently find high-quality solutions in complex optimization problems.

[0012] Optionally, the improved fitness evaluation mechanism is used to select parents for the iteration of the genetic algorithm to obtain an improved genetic algorithm, including: in the iteration process of the genetic algorithm, the roulette selection method is used to calculate the probability of individuals in the population being selected as parents according to the population fitness value; and the parents of the genetic algorithm in the iteration process are determined according to the probability to obtain the improved genetic algorithm.

[0013] The roulette selection method is used to make the probability of individuals in the population being selected proportional to their fitness value, so that the probability of becoming parents is calculated. This method has randomness, can avoid the algorithm from falling into a local optimal solution too early, and enhances the global optimization ability of the genetic algorithm to find the optimal solution faster.

[0014] Optionally, the mixed sampling strategy is used to optimize the selection of sample points of the classification proxy model and the iteration process of the improved genetic algorithm to determine the optimal structural design scheme of the combined impact of the blast wave and the fragment group, including: obtaining a decision boundary according to the classification proxy model; updating the sample set according to the decision boundary and the mixed sampling strategy; optimizing and training the classification proxy model by using the updated sample set; and determining the optimal structural design scheme of the combined impact of the blast wave and the fragment group by using the optimized and trained classification proxy model and the improved genetic algorithm for optimization and solving.

[0015] The decision boundary is obtained by the classification proxy model, which can clearly define the effective area of the solution space. Based on this boundary and the mixed sampling strategy, the optimization sample points are generated, which can accurately supplement high-quality sample points, improve the sample quality of the classification proxy model, and be used for optimizing and training the classification proxy model, thereby further improving the prediction accuracy of the classification proxy model. Combined with the improved genetic algorithm for optimization and solving, the global optimal solution can be efficiently searched under the guidance of a more accurate model. Through the iterative optimization of the cycle, the optimal structural design scheme of the combined impact of the blast wave and the fragment group can be quickly and accurately determined, and the structural protection capability is effectively improved.

[0016] Optionally, the updating the sample set according to the decision boundary and the hybrid sampling strategy comprises: obtaining a structure optimization design set and a first structure optimization design according to the improved genetic algorithm; obtaining one or more second structure optimization designs by calculating distances from the structure optimization design set to the decision boundary; performing a second numerical analysis on the first structure optimization design and the second structure optimization design, and updating the sample set according to a result of the second numerical analysis.

[0017] The application improves the data quality of the sample set by adding new high-quality data sample points.

[0018] Optionally, the optimization solving by using the classification agent model trained by optimization and the improved genetic algorithm to determine the optimal structure design scheme under the combined impact of blast waves and fragment groups comprises: obtaining a local structure optimization design by using the classification agent model trained by optimization and the improved genetic algorithm for optimization solving; obtaining a local structure optimal design according to sample points of the classification agent model; calculating a maximum value of relative deviation between the local structure optimization design and the local structure optimal design to obtain a maximum relative deviation value; setting a threshold value, and determining whether the maximum relative deviation value is less than the threshold value, and if so, determining the local structure optimal design as the optimal structure design scheme under the combined impact of blast waves and fragment groups.

[0019] The application uses the classification agent model trained by optimization and the improved genetic algorithm for solving, can efficiently explore the design space, and obtain a local structure optimization design. The local structure optimal design is obtained through sample points, which provides a reliable reference for scheme evaluation. The relative deviation between the two is calculated and compared with a threshold value, which can effectively determine the convergence degree of the design scheme. If the relative deviation is less than the threshold value, the local structure optimal design is determined as the final scheme, which not only ensures the accuracy and stability of the scheme, but also avoids excessive calculation, and improves the efficiency of determining the optimal structure design scheme under the combined impact of blast waves and fragment groups.

[0020] Optionally, the structure optimization design set comprises a plurality of structure optimization designs, and the obtaining one or more second structure optimization designs by calculating distances from the structure optimization design set to the decision boundary comprises: calculating distances from each structure optimization design in the structure optimization design set to the decision boundary; selecting one or more structure optimization designs in the structure optimization design set according to the distances to form candidate sample points; and screening the candidate sample points by calculating differences between the candidate sample points and the sample set to obtain one or more second structure optimization designs.

[0021] The application calculates the distance of each structure optimization design in the structure optimization design set to the decision boundary, and selects sample points according to the distance, can focus on the key area close to the decision boundary, and excavate design points with potential value, which often contain key information determining the advantages and disadvantages of the design scheme. Secondly, by calculating the difference of the selected sample points and the sample set and removing the points with small difference, the sample redundancy can be effectively avoided, and the diversity of the sample can be increased.

[0022] Optionally, the target function value satisfies the following formula:

[0023] ,

[0024] wherein, is the target function processed based on the constraint processing strategy, is an individual in the population of the genetic algorithm, is the structure quality of the individual in the population is a feasible solution, is the maximum structure quality of the individual in the population, is an infeasible solution, is the difference value of the infeasible solution diversity.

[0025] The fitness value calculation formula of the application includes the infeasible solution in the calculation of the whole fitness value, fully considers the potential effective information of the infeasible solution diversity, and improves the applicability of the application.

[0026] Another aspect of the application also provides a structure optimization design system against combined impact of blast wave and fragment group, the system comprising a processor, an input device, an output device and a memory, the processor, the input device, the output device and the memory are connected with each other, wherein the memory is used for storing a computer program, the computer program comprises program instructions, the processor is configured to call the program instructions, and execute the structure optimization design method against combined impact of blast wave and fragment group provided in the previous aspect of the application.

[0027] The structure optimization design system against combined impact of blast wave and fragment group has the advantages of compact structure, stable performance, high integration degree and simple constitution, can stably execute the structure optimization design method against combined impact of blast wave and fragment group provided in the previous aspect of the application, and further improves the overall applicability and practical application ability of the application. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is a flow chart of the structure optimization design method against combined impact of blast wave and fragment group of the embodiment of the application;

[0029] Figure 2This is a flowchart of the structural optimization design method for resisting the combined impact of blast waves and fragmentation swarms according to an embodiment of the present invention.

[0030] Figure 3 This is a schematic diagram of the structural optimization design system for resisting the combined impact of blast waves and fragmentation groups according to an embodiment of the present invention. Detailed Implementation

[0031] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0032] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0033] Figure 1 This is a flowchart of the structural optimization design method for resisting combined blast wave and fragmentation group impact according to an embodiment of the present invention. To address the problems of insufficient optimization accuracy and high computational cost associated with a single point-addition criterion, such as... Figure 1 The method shown includes the following steps:

[0034] Step S1: Obtain a structural design scheme for resisting the combined impact of blast wave and fragmentation group, and perform a first numerical analysis on the structural design scheme to obtain a sample set.

[0035] In this embodiment, obtaining a structural design scheme resistant to the combined impact of blast wave and fragmentation swarm specifically includes the following sub-steps:

[0036] Step S101: Determine the optimal design variables for the structural design scheme that resists the combined impact of blast wave and fragmentation group, and set the value range of the optimal design variables.

[0037] Optimization design variables include the type of lattice sandwich core. (Trapezoidal or concave hexagonal), number of lattice layers Dot matrix wallboard thickness Height Folding angle Front panel thickness Rear panel thickness Polyurea plate thickness Ceramic plate thickness Lamination sequence of both polyurea plate and ceramic plate Lamination sequence of dot matrix structure and ceramic polyurea . The embodiment takes no damage to the rear plate ( ) as an optimization constraint and lightweight as an optimization objective to carry out performance optimization design under structural combined load.

[0038] In an optional embodiment, the value range of the optimization design variable includes , , , , , , , , , And . Wherein, in the formula, Indicates that the structure protection side is not damaged, Indicates that the polyurea is above the ceramic, and vice versa, Indicates that the dot matrix structure is above the ceramic / polyurea layer, and vice versa; Indicates that the dot matrix type is trapezoidal, and vice versa for concave hexagonal.

[0039] Step S102, according to the optimization design variable and the value range of the optimization design variable, a Latin hypercube sampling method is used to determine a plurality of structure design schemes for resisting combined impact of blast wave and fragment group as initial sample points in the design space.

[0040] In the present embodiment, according to the optimization design variable, an optimization design combination composed of the value range of the optimization design variable forms a design space, and a Latin hypercube sampling method is used to determine a plurality of sample points as initial sample points in the design space, wherein each initial sample point is a structure design scheme for resisting combined impact of blast wave and fragment group.

[0041] ​The Latin hypercube sampling method is an efficient sampling technique for generating samples of a multivariate distribution. The principle of this method is to divide the value range of each optimization design variable into several equal probability intervals, and then randomly select a sample point in each interval. These sample points are combined to form a sample set. This sampling method can ensure that each interval of each optimization design variable has a sample point, and the sample points are more uniformly distributed in the entire value range. Compared with random sampling, it can more effectively cover the optimization design variable space.

[0042] After obtaining the structure design scheme subjected to combined impact of blast wave and fragment cloud by the above method, numerical analysis is performed on the structure design scheme to obtain a sample label of the structure design scheme , and a sample set is formed . In the construction process of the numerical model, first, geometric modeling is performed, the constitutive model is determined, the material domain is discretized, and geometric boundary conditions and contact conditions are set. Through numerical simulation, the protection performance value of each structure design scheme is obtained. If the performance value meets the protection index, the label of the structure design scheme is set to -1, otherwise, it is set to 1, and finally the sample set is obtained.

[0043] Specifically, numerical analysis is a mathematical modeling and computing method used to predict and evaluate the performance of a structure design scheme under specific conditions. It begins with geometric modeling, which involves creating a three-dimensional model of the structure using computer-aided design (CAD) software. Geometric modeling needs to accurately reflect the dimensions, shape, and material distribution of the design scheme to ensure the accuracy of subsequent analysis.

[0044] After completing the geometric modeling, the next step is to determine the constitutive model, which is a mathematical model that describes the behavior of the structure. This may include elastic, plastic, viscoelastic, or other more complex material behavior models. The selection of the constitutive model is crucial to the accuracy of the simulation results, as it determines how the material responds under stress.

[0045] Subsequently, the material domain is discretized, which means dividing the continuous material domain into a finite number of small elements that can be processed by the structure analysis software. The quality and method of discretization directly affect the accuracy and computational efficiency of numerical simulation.

[0046] After discretization, geometric boundary conditions and contact conditions need to be set. Geometric boundary conditions define the fixed and free boundaries of the structure, while contact conditions describe the interaction between the structure and other objects or materials. These conditions are crucial for simulating the performance of the structure in the actual working environment.

[0047] After completing the above steps, the model is run through numerical simulation to obtain the protection performance value of each structure design scheme​ The protective performance values include, but are not limited to, stress, strain, deformation, energy absorption, and other key indicators. According to the comparison of these performance values with the preset protective indicators, a label is assigned to each structural design scheme If the performance values meet the protective indicators, the label is set to -1, indicating that the design scheme is feasible; if not, the label is 1, indicating that it is not feasible. Finally, all structural design schemes and their corresponding labels are combined to form a sample set This sample set contains a comprehensive evaluation of the initial design scheme, providing necessary data support for subsequent optimization algorithms.

[0048] In this way, numerical analysis not only verifies the feasibility of the design scheme, but also provides a foundation for the fitness evaluation mechanism of the genetic algorithm, guiding the optimization process to meet the design requirements. This process is an indispensable part of structural optimization design, ensuring the effectiveness and reliability of the design scheme in practical applications.

[0049] Step S2, a classification proxy model is constructed, and the classification proxy model is trained using the sample set.

[0050] In this embodiment, a support vector machine is used to build a classification proxy model, and the sample set is used to train the classification proxy model. The support vector machine divides the design space into two regions by finding a decision boundary, where the design points that meet the structural protective performance indicators are on one side, and the design points that do not meet the protective performance indicators are on the other side.

[0051] The support vector machine has many advantages in building a classification proxy model. First, it divides different categories by finding a maximum margin hyperplane. This optimization goal based on geometric margin makes the model have strong generalization ability in classification, and can learn a relatively robust decision boundary from limited training samples. Second, the support vector machine uses the kernel trick, which can map linearly inseparable data in low-dimensional space to high-dimensional space, thus realizing complex nonlinear classification and greatly expanding its application range. In addition, the complexity of the classification proxy model mainly depends on the number of support vectors, not the dimension of the data, which makes it perform well in handling high-dimensional data and has high computational efficiency. Finally, the support vector machine has a solid theoretical foundation based on the VC dimension and structural risk minimization principle of statistical learning theory, which can effectively avoid overfitting in small sample cases, ensuring the stability and reliability of the model.

[0052] Step S3, a genetic algorithm is introduced, and the fitness evaluation mechanism of the genetic algorithm is improved using the classification proxy model.

[0053] In the present embodiment, the genetic algorithm is an optimization algorithm simulating the biological evolution process, which is based on natural selection and genetic mechanisms, and gradually evolves the optimal solution from the initial population through operations such as selection, crossover and mutation. The core idea of genetic algorithm is "survival of the fittest", that is, individuals with high fitness have a higher probability of being selected for reproduction, so as to pass on good genes to the next generation.

[0054] The improvement of the fitness evaluation mechanism of the genetic algorithm by using the classification agent model specifically includes the following sub-steps:

[0055] Step S301, using the classification agent model to divide the individuals in the population of the genetic algorithm into feasible solutions and infeasible solutions.

[0056] In the present embodiment, the classification agent model can divide the population of the genetic algorithm into feasible solutions and infeasible solutions, and the population of the genetic algorithm comes from the sample set. In an optional embodiment, the population of the genetic algorithm can also come from new sample points.

[0057] Step S302, calculating the diversity difference between the infeasible solution and the individuals in the population to obtain the infeasible solution diversity difference value.

[0058] For constrained optimization problems, it is usually necessary to convert them into unconstrained optimization problems to be suitable for genetic algorithm solving. The criterion requires that the feasible solution is always better than any infeasible solution, and the individual solution with smaller objective function value in the feasible solution is dominant, and the individual solution with low constraint violation degree in the infeasible solution is dominant. Therefore, the unconstrained optimization problem obtained according to the traditional processing method is:

[0059] ,

[0060] Wherein, is the objective function of the converted unconstrained optimization problem, is the structural quality of the individual in the population, is the feasible solution; is the maximum structural quality of the individual in the population, is the predicted value of the structural performance of the classification agent model.

[0061] However, as can be seen from the above formula, the individual in the infeasible solution is always +1, which leads to the inability to determine the dominant individual solution. Therefore, the present application proposes to use individual diversity instead of violation degree evaluation strategy to solve the problem of infeasible solution unable to determine the individual dominant.

[0062] The diversity difference refers to the difference between the infeasible solution and other individuals in the population, therefore, the infeasible solution diversity difference value satisfies the following formula:

[0063] ,

[0064] wherein, is an infeasible solution diversity difference value, is a population size in a genetic algorithm, is a sample point dimension of a feature, i.e. dimension of an optimization design variable, is a value of an infeasible solution in the i-th dimension, is a value of the j-th individual in the population in the i-th dimension.

[0065] Step S303, improving a fitness evaluation mechanism of the genetic algorithm according to the infeasible solution diversity difference value.

[0066] wherein, improving the fitness evaluation mechanism of the genetic algorithm according to the infeasible solution diversity difference value specifically includes the following sub-steps:

[0067] Step S30301, calculating a target function of the improved genetic algorithm according to the infeasible solution diversity difference value, to obtain a target function value.

[0068] The target function value satisfies the following formula:

[0069] ,

[0070] wherein, is a target function processed based on a constraint processing strategy, is an individual in a population of a genetic algorithm, is a structural quality of the individual in the population, is a feasible solution, is a maximum structural quality of the individual in the population, is an infeasible solution diversity difference value.

[0071] Step S30302, sorting the individual in the population according to the target function value, to obtain a sorting position.

[0072] In the embodiment, sorting the individual in the population according to the target function value can clearly show the relative advantages and disadvantages of each individual in the current population. By the quantitative standard of the target function value, all individuals are placed in the same scale for measurement, so that the algorithm can intuitively distinguish which individual is closer to the ideal solution of the problem and which individual has a larger improvement space. This clear understanding of the performance of the individual provides a clear direction for subsequent genetic operations such as selection, crossover and mutation.

[0073] ​Based on the ranking position, the algorithm can be more targeted when selecting parent individuals. For example, during the selection process, individuals with higher ranking positions, i.e., better performance in the objective function, are more likely to be selected as parents, which is consistent with the concept of survival of the fittest in natural selection, helping to pass on good genes to the next generation and thus promoting the evolution of the population towards better directions. At the same time, ranking also provides a basis for self-adaptive adjustment of genetic algorithm parameters. According to the ranking results, the algorithm can dynamically adjust key parameters such as crossover probability and mutation probability. For individuals with higher ranking positions, the mutation probability is appropriately reduced to protect good gene combinations. For individuals with lower ranking positions, the mutation probability is appropriately increased to increase the possibility of exploring new solution spaces, thereby improving the global search ability of the algorithm.

[0074] In addition, the sorting operation also facilitates performance evaluation of the algorithm. By observing the ranking changes of individuals at different iteration times, the convergence trend of the algorithm can be intuitively understood. If the top-ranked individuals gradually stabilize and the objective function value continuously optimizes in consecutive iterations, it indicates that the algorithm is effectively converging. Conversely, if the individual ranking fluctuates greatly and the objective function value does not significantly improve, it suggests that the algorithm may be trapped in a local optimum or the search process has problems, and the algorithm needs to be adjusted. This ranking-based performance evaluation method provides strong support for the optimization and improvement of the algorithm, enabling it to better handle various complex optimization problems

[0075] Step S30303, according to the ranking position, the fitness value of the individual in the population is calculated, and the population fitness value is obtained.

[0076] In this embodiment, the ranking-based method is used to calculate the fitness value of individuals in the population, and the application can effectively avoid the imbalance of selection pressure caused by the large difference in fitness value. The traditional method of calculating fitness based on the objective function may cause some individuals with extremely high fitness to dominate the selection process, leading to premature convergence to a local optimal solution. By calculating the fitness based on the ranking position, regardless of the difference in the actual objective function value of individuals, the fitness is assigned according to their relative order in the population, so that different individuals have a relatively reasonable opportunity to participate in reproduction during the selection process, which helps to maintain the diversity of the population.

[0077] At the same time, this method enhances the adaptability of the algorithm to complex problems. In many problems, the objective function has a complex form and there are multiple local optimal regions. By calculating the fitness based on the ranking position, the algorithm does not excessively focus on the current individual that appears to be the best, but can explore a wider solution space. It encourages the algorithm to fully exploit potential solutions in different regions during the early search stage, and gradually filters out better individuals as the iteration progresses, making it more likely to find a global optimal solution.

[0078] In addition, the fitness value is calculated based on the ranking position, and the calculation complexity is relatively low. Without complex mathematical operations on the fitness of each individual, the fitness value can be assigned according to the preset rule only by determining the ranking position of the individual in the population. When dealing with large-scale populations, this can significantly reduce the consumption of computing resources and computing time, improve the running efficiency of the algorithm, and enable the algorithm to more efficiently handle large-scale complex problems, thereby providing a more feasible solution for practical applications.

[0079] The population fitness value satisfies the following formula:

[0080] ,

[0081] In the formula, is the fitness of the ith individual, is the ranking position of the ith individual in the population, is the selection pressure difference.

[0082] The selection pressure difference reflects the potential ability of the selection operation to improve the population fitness, and measures the ability of the selection operator to select relatively good individuals from the current population.

[0083] A larger selection pressure difference means that the average fitness of the selected parent individuals is significantly higher than the average fitness of the population. This enables the genes of good individuals in the population to spread more quickly in the next generation, accelerating the convergence of the population to the optimal solution. However, if the selection pressure difference is too large, it may cause the algorithm to converge prematurely, missing the global optimal solution. Because the algorithm relies too much on the current good individuals, it ignores other potential solution spaces.

[0084] A smaller selection pressure difference indicates that the selection operation does not distinguish much between individuals, and the average fitness of the parent individuals is close to the average fitness of the population. This helps to maintain the diversity of the population, as more individuals with different fitness levels have the opportunity to participate in reproduction. However, if the selection pressure difference is too small, the evolution speed of the algorithm will be slow, as the advantages of good individuals are not fully utilized and the overall fitness of the population improves slowly.

[0085] Therefore, the value of the selection pressure difference can be flexibly selected according to different actual situations, or gradually adjusted during the calculation process.

[0086] Step S4, based on the improved fitness evaluation mechanism, selecting parents for the iteration of the genetic algorithm to obtain an improved genetic algorithm.

[0087] Wherein, based on the improved fitness evaluation mechanism, selecting parents for the iteration of the genetic algorithm to obtain an improved genetic algorithm specifically includes the following sub-steps:

[0088] Step S401, in the iteration process of the genetic algorithm, the probability of individuals in the population as parents is calculated according to the fitness value of the population by roulette selection method.

[0089] The probability of individuals in the population as parents satisfies the following formula:

[0090]

[0091] Wherein, is the probability of individuals in the population as parents, is the number of population in the genetic algorithm, is the fitness of the ith individual.

[0092] In this embodiment, the genetic algorithm calculates the probability of individuals as parents in the iteration process according to the fitness value of the population by roulette selection method. Roulette selection method closely follows the theory of natural selection, and takes fitness as the key indicator to measure the quality of individuals. The individuals with high fitness have a greater chance of being selected as parents in the "roulette" rotation, so as to pass on the excellent genes and guide the population to develop in the direction of higher fitness, gradually optimizing the population to approach the optimal solution of the problem.

[0093] This selection method ingeniously integrates randomness, and even individuals with poor fitness also have a certain probability to participate in the reproduction process. This feature is of great significance to maintain the diversity of the population, and can effectively prevent the algorithm from falling into the dilemma of local optimal solution in the early stage. In a complex search space, the seemingly ordinary individuals in the early stage may bloom unique value in the subsequent gene combination and evolution process, and open up new paths to find the global optimal solution.

[0094] Another outstanding advantage of roulette selection method is its simplicity. Only the proportion of individual fitness value in the total fitness of the population is calculated, and the probability of individual as parent can be determined. This simple calculation method greatly reduces the computational complexity of the algorithm, and can significantly improve the running efficiency of the algorithm when facing large-scale population or limited computing resources, so that it can run efficiently. Therefore, roulette selection method can easily integrate with other operators such as crossover and mutation in genetic algorithm, and promote the stable operation of genetic algorithm, providing strong support for solving various complex problems.

[0095] Step S402, the parents in the iteration process of the genetic algorithm are determined according to the probability, and the improved genetic algorithm is obtained.

[0096] ​In this embodiment, determining the parent generation in the genetic algorithm iteration based on probability simulates natural selection, giving individuals with high fitness a greater chance to pass on their genes and driving the population towards a better direction. Simultaneously, probabilistic selection provides opportunities for individuals with low fitness, preventing premature convergence, maintaining population diversity, and helping the algorithm explore more effectively in complex search spaces, ultimately finding the global optimum.

[0097] Based on the improvements to the above genetic algorithm, an improved genetic algorithm is obtained.

[0098] Step S5: Using a hybrid point addition strategy, the selection of sample points for the classification proxy model and the iterative process of the improved genetic algorithm are optimized to determine the optimal structural design scheme for resisting the combined impact of blast wave and fragmentation swarm.

[0099] The process of using a hybrid point-addition strategy to optimize the selection of sample points in the classification proxy model and the iterative process of the improved genetic algorithm to determine the optimal structural design scheme for resisting the combined impact of blast waves and fragmentation swarms includes the following sub-steps:

[0100] Step S501: Obtain the decision boundary based on the classification agent model.

[0101] In this embodiment, to find a decision boundary with high classification accuracy, the "interval" between the data points and the decision boundary needs to be as large as possible, while simultaneously satisfying the constraint of correct classification:

[0102] ,

[0103] Where, in the formula, The number of training samples. Let i be the Lagrange multiplier of sample i. Let j be the Lagrange multiplier of sample j. For the sample The tag, For the sample The tag, A kernel function used to evaluate input samples. With the sample set Sample similarity, This is the penalty parameter.

[0104] Solving the above equation using optimization methods yields the Lagrange multipliers for each sample point. Using any support vector Then the bias can be determined. The formula is as follows:

[0105] ,

[0106] determining and After that, the decision boundary of the support vector machine can be obtained for classification of the structural protection performance.

[0107] The decision boundary satisfies the following formula:

[0108] ,

[0109] wherein, is the decision boundary, is the number of training samples, is the Lagrange multiplier of the sample point is the label, is the kernel function for calculating the similarity between the sample and is the bias term. According to the value of , the influence of the sample point on the decision boundary can be determined. When

[0110] , the sample point is usually far away from the decision boundary and has been correctly classified, and has no direct influence on the decision boundary; when , the sample point is near the classification boundary and is a support vector, directly affecting the classification interval and the position of the decision boundary. Step S502, updating the sample set according to the decision boundary and the mixed adding point strategy.

[0111] In this embodiment, the core idea of the mixed adding point strategy is to ensure that the optimization algorithm has good local development and global search ability. The mixed adding point criterion adds one sample through local adding point and adds m samples through global adding point in one sequence optimization cycle.

[0112] Wherein, updating the sample set according to the decision boundary and the mixed adding point strategy specifically includes the following sub-steps:

[0113] Step S50201, obtaining a structure optimization design set and a first structure optimization design according to the improved genetic algorithm.

[0114] In this embodiment, the structure optimization design set refers to the improved genetic algorithm optimization population obtained after the population of the improved genetic algorithm passes through the steps of constraint processing, fitness calculation, individual selection and crossover mutation of the improved genetic algorithm. The first structure optimization design refers to the best individual in the optimization population, and the best individual is the individual with the highest fitness.

[0115]

[0116] ​​The first structure optimization design forms a local point sample, and the minimum model prediction criterion is a point adding method first applied in sequential optimization design. It is widely used because of its simple operation and easy implementation. This method can quickly guide the convergence of the optimization algorithm and has strong local development capability. Therefore, the minimum model prediction criterion is used for local point adding in the present technology. In each cycle, the current surrogate model is optimized and solved, the obtained optimal design point is added to the sample set, and the surrogate model is updated accordingly. The mathematical expression of this criterion is:

[0117] ,

[0118] In the formula, "arg" represents the mapping relationship between the independent variable and the dependent variable, and the return value is the current minimum target corresponding to the structure parameter.

[0119] However, in the optimization design problem of resisting combined load, due to the large number of optimization variables and the complex proxy relationship between the optimization variables and the protection performance, the global classification accuracy of the proxy model is low, which leads to the global optimal point being easily ignored. In addition, the structural response analysis of the structure resisting combined load has high computational overhead, and structural response analysis needs to be performed for each optimal point. If only one optimal point is added each time, the number of iterations will increase significantly, which leads to high computational overhead.

[0120] Therefore, the present technology proposes a global point adding method for support vector machine classification proxy model, which aims to improve the accuracy of the proxy model by sampling in the sensitive area with large prediction deviation, so as to guide the optimization to the direction of the global optimum.

[0121] Step S50202, one or more second structure optimization designs are obtained by calculating the distance of the structure optimization design set to the decision boundary.

[0122] Wherein, the one or more second structure optimization designs are obtained by calculating the distance of the structure optimization design set to the decision boundary, and the specific steps are as follows:

[0123] Step S5020201, the distance of each structure optimization design in the structure optimization design set to the decision boundary is calculated.

[0124] In this embodiment, the distance of each structural optimization design in the structural optimization design set to the decision boundary is calculated. The decision boundary not only determines the position of the classification boundary, but also quantifies the distance of the sample point to the boundary. The greater the distance, the higher the confidence of the classification. The smaller the distance, the more significant the contribution of the sample point to the classification, i.e., the support vector. Since the samples near the decision boundary are more likely to be misclassified, this area is considered to be a sensitive area. By sampling in these areas, the prediction accuracy of the surrogate model can be effectively improved.

[0125] In step S5020202, one or more structural optimization designs in the structural optimization design set are selected as candidate sample points according to the distance.

[0126] In this embodiment, all sample points are sorted according to the distance, and one or more structural optimization designs closest to the decision boundary are selected as candidate sample points.

[0127] These sample points are located near the decision boundary and contain rich structural feature information. The decision boundary is a key limit that distinguishes different design performance regions. Sample points near it can reflect subtle changes from one performance to another, which helps to understand the complex relationship between design variables and performance.

[0128] Secondly, subsequent analysis based on these sample points can effectively improve the optimization efficiency. Compared with blind exploration in the entire design set, focusing on the area near the decision boundary can more accurately locate the optimization direction and avoid wasting computational resources in a large number of invalid regions far from the boundary.

[0129] Furthermore, candidate sample points can help discover potential innovative designs. Since they are in the sensitive region of performance change, in-depth study of them may reveal new ideas that break through traditional designs, thereby bringing more innovative and competitive solutions to structural optimization and promoting the development of structural design.

[0130] In step S5020203, the candidate sample points are screened by calculating the difference between the candidate sample points and the sample set, and one or more second structural optimization designs are obtained.

[0131] In this embodiment, in order to avoid repeated selection of samples, it is necessary to ensure that the candidate sample points have sufficient difference from the existing sample set. Therefore, among the selected one or more samples, samples with less difference need to be removed, and the remaining samples are retained as second structural optimization designs.

[0132] The difference between the candidate sample points and the sample set satisfies the following formula:

[0133] ,

[0134] wherein, represents the difference between sample i and sample j. For a threshold is set, and when the newly selected sample is discarded.

[0135] Step S50203, performing a second numerical analysis on the first structure optimization design and the second structure optimization design, and updating the sample set according to the result of the second numerical analysis.

[0136] In this embodiment, numerical analysis is performed on the first structure optimization design and the second structure optimization design, a label is obtained, and the result after numerical analysis is added to the sample set to realize data updating of the sample set. The method is to improve the data quality of the sample set. The updated sample set.

[0137] Step S503, using the updated sample set to optimize and train the classification agent model.

[0138] In this embodiment, the above optimization calculation process can obtain more accurate structure scheme designs at the boundary between feasible schemes and infeasible schemes and optimized structure scheme designs. These structure scheme designs are used to update the sample set, so the sample set reflects the current optimal and most diverse individuals in the population. Using this updated sample set to retrain the classification agent model can make the classification agent model learn the latest design space distribution and constraint conditions. This helps the classification agent model to more accurately identify feasible solutions that meet the design requirements and infeasible solutions that need to be further improved. In addition, the classification agent model after optimization training can more effectively guide the selection operation of the genetic algorithm, and preferentially select individuals that are more likely to produce high-quality offspring. Ultimately, this process not only improves the performance of the classification agent model, but also speeds up the convergence speed of the genetic algorithm and reduces the number of iterations required to reach the optimal solution. Through continuous iteration and optimization, a more accurate and reliable optimal structure design scheme can be obtained to meet the complex design requirements of anti-blast wave and fragment group combined impact.

[0139] Step S504, using the classification agent model after optimization training and the improved genetic algorithm to perform optimization solving to determine the optimal structure design scheme of anti-blast wave and fragment group combined impact.

[0140] wherein, using the classification agent model after optimization training and the improved genetic algorithm to perform optimization solving to determine the optimal structure design scheme of anti-blast wave and fragment group combined impact specifically includes the following sub-steps:

[0141] Step S50401, the classification agent model and the improved genetic algorithm are used for optimization solving to obtain a local structure optimization design.

[0142] In the embodiment, the updated classification agent model improves the accuracy of classification prediction, and the improved genetic algorithm is used for optimization solving. After the iteration of the improved genetic algorithm ends, the individual with the highest fitness is taken as the local structure optimization design.

[0143] Step S50402, the local structure optimal design is obtained according to the sample points of the classification agent model.

[0144] In the embodiment, the structure optimal design in the sample points of the classification agent model is the first structure optimization design determined after the last iteration of the genetic algorithm ends, and the first structure optimization design is taken as the local structure optimal design.

[0145] Step S50403, the maximum value of the relative deviation between the local structure optimization design and the local structure optimal design is calculated to obtain a maximum relative deviation value.

[0146] The maximum relative deviation value satisfies the following formula:

[0147] ,

[0148] is the maximum relative deviation value, is the first component of the local structure optimization design, is the first component of the local structure optimal design, is the first component of the local structure optimization design, is the first component of the local structure optimal design, is the first component of the local structure optimal design, is the first component of the local structure optimal design, is the feature dimension of the sample point.

[0149] Step S50404, a threshold value is set, and it is judged whether the maximum relative deviation value is less than the threshold value. If yes, the local structure optimal design is determined as the optimal structure design scheme of the combined impact of the blast wave and the fragment group.

[0150] In the embodiment, the threshold value is set to set the convergence condition. The threshold value can be flexibly selected according to different required accuracies. The smaller the threshold value, the higher the final optimization design accuracy, and at the same time, the calculation cost will also increase.

[0151] When the relative deviation value is less than the threshold value, it is indicated that the optimization structure design scheme output by the improved genetic algorithm has met the requirements, and it is also indicated that the classification agent model has reached the ideal effect at this time, and further optimization training is not needed.​

[0152] When the relative deviation value is greater than the threshold value, it indicates that the optimization structure design scheme output by the improved genetic algorithm does not meet the requirements, and further optimization training needs to be performed on the classification agent model, at this time, the algorithm enters a loop.

[0153] Specifically, as shown in Figure 2 , the present application starts optimization calculation, generates initial sample points by a sampling method and inputs , and then uses a numerical model to obtain sample point responses , at this time, a sample set is obtained, the sample set is used to train a support vector machine agent model to obtain a classification agent model, the optimization problem of resisting combined loads is solved by using the agent model to assist the genetic algorithm, and then it is judged whether the convergence condition is converged, if the convergence condition is converged, the final optimization design is output, and the whole optimization calculation process is ended, if the convergence condition is not converged, new sample points are obtained through a point adding strategy, and corresponding protective performance is obtained through parallel calculation to form a new sample set , the new sample set is used to continue training the classification agent model, and the next iteration is entered until the convergence condition is finally reached.

[0154] As shown in Figure 3 , in another aspect, the present application also provides a structure optimization design system for resisting blast waves and fragment groups combined impact, comprising: a processor, an input device, an output device and a memory, the processor, the input device, the output device and the memory are connected with each other, wherein the memory is used for storing a computer program, the computer program comprises program instructions, the processor is configured to call the program instructions to execute the related steps of the related embodiments in the structure optimization design method for resisting blast waves and fragment groups combined impact of the present application.

[0155] The structure optimization design system for resisting blast waves and fragment groups combined impact provided by the present application can integrate each functional component in one processing component, or each component can exist physically alone, or two or more components can be integrated in one component. The above integrated components can be realized in the form of hardware or in the form of software function, further improving the overall applicability and practical application ability of the present application.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the description of the present application.

Claims

1. A method for optimal design of structures against combined blast wave and fragment cloud impact, characterized in that, The method comprises: obtaining a structure design scheme subjected to combined impact of blast waves and fragment clouds, performing first numerical analysis on the structure design scheme to obtain a sample set; constructing a classification agent model and training the classification agent model by using the sample set; introducing a genetic algorithm and improving a fitness evaluation mechanism of the genetic algorithm by using the classification agent model; based on the improved fitness evaluation mechanism, selecting parents for iteration of the genetic algorithm to obtain an improved genetic algorithm; adopting a hybrid sampling strategy to optimize selection of sample points of the classification agent model and an iteration process of the improved genetic algorithm, and determining an optimal structure design scheme subjected to combined impact of blast waves and fragment clouds, comprising: obtaining a decision boundary according to the classification agent model; updating the sample set according to the decision boundary and the hybrid sampling strategy, comprising: obtaining a structure optimization design set and a first structure optimization design according to the improved genetic algorithm; obtaining one or more second structure optimization designs by calculating distances from the structure optimization design set to the decision boundary, comprising: the structure optimization design set comprises a plurality of structure optimization designs; calculating distances from each of the structure optimization designs in the structure optimization design set to the decision boundary; selecting one or more of the structure optimization designs in the structure optimization design set according to the distances to form candidate sample points; screening the candidate sample points by calculating differences between the candidate sample points and the sample set to obtain one or more second structure optimization designs; performing second numerical analysis on the first structure optimization design and the second structure optimization design, and updating the sample set according to a result of the second numerical analysis; optimizing training the classification agent model by using the updated sample set; determining the optimal structure design scheme subjected to combined impact of blast waves and fragment clouds by using the optimized and trained classification agent model and the improved genetic algorithm.

2. The method according to claim 1, wherein The improvement of the fitness evaluation mechanism of the genetic algorithm by using the classification agent model comprises: dividing individuals in a population of the genetic algorithm into feasible solutions and infeasible solutions by using the classification agent model; calculating diversity differences between the infeasible solutions and the individuals in the population to obtain infeasible solution diversity difference values; improving the fitness evaluation mechanism of the genetic algorithm according to the infeasible solution diversity difference values.

3. The method according to claim 2, wherein The improvement of the fitness evaluation mechanism of the genetic algorithm according to the infeasible solution diversity difference values comprises: calculating a target function of the improved genetic algorithm according to the infeasible solution diversity difference values to obtain a target function value; sorting the individuals in the population according to the target function value to obtain sorting positions; calculating fitness values of the individuals in the population according to the sorting positions to obtain population fitness values.

4. The method according to claim 3, wherein The selection of parents for iteration of the genetic algorithm based on the improved fitness evaluation mechanism to obtain the improved genetic algorithm comprises: In the iteration process of the genetic algorithm, according to the population fitness value, a roulette selection method is used to calculate the probability of individuals in the population as parents; According to the probability, the parents of the genetic algorithm in the iteration process are determined, and the improved genetic algorithm is obtained.

5. The method of optimal design of structures against blast wave and fragment cloud combined impact according to claim 1, characterized in that, The optimization training of the classification agent model and the improved genetic algorithm are used for optimization solving to determine the optimal structure design scheme of the anti-blast wave and fragment group combined impact includes: The optimization training of the classification agent model and the improved genetic algorithm are used for optimization solving to obtain the local structure optimization design; According to the sample points of the classification agent model, the local structure optimal design is obtained; The maximum value of the relative deviation between the local structure optimization design and the local structure optimal design is calculated to obtain the maximum relative deviation value; A threshold value is set to determine whether the maximum relative deviation value is less than the threshold value, and if so, the local structure optimal design is determined as the optimal structure design scheme of the anti-blast wave and fragment group combined impact.

6. The method of optimal design of structures against blast waves and clusters of fragments according to claim 3, characterized in that, The target function value satisfies the following formula: ， wherein, is the objective function after being processed based on the constraint handling strategy, is an individual in the population of the genetic algorithm, is the structural mass of the individual in the population is the structural mass of the individual in the population, is the feasible solution, is the maximum structural mass of the individual in the population, is the infeasible solution diversity difference value.

7. A system for the optimal design of structures against blast wave and fragment cloud combined impact, characterized in that, It includes: A processor, an input device, an output device and a memory, the processor, the input device, the output device and the memory are connected with each other, wherein the memory is used to store a computer program, the computer program includes program instructions, the processor is configured to call the program instructions, and executes the structure optimization design method of the anti-blast wave and fragment group combined impact as claimed in any one of claims 1 to 6.

Images