A mixed explosive formula rapid generation method for multi-scene requirements

Abstract

The application provides a mixed explosive formula rapid generation method for multi-scene requirements, and relates to the technical field of explosive formula generation, and the method comprises the following steps: mapping the formula requirement into a target function and a constraint function; solving the target function and the constraint function through a non-dominated sorting genetic algorithm to obtain a main component proportion of the mixed explosive; predicting the functional additive proportion of the mixed explosive through a plurality of trained machine learning models, the main component proportion of the mixed explosive and a performance index of the mixed explosive to obtain a predicted functional additive proportion of the mixed explosive; and combining the main component proportion of the mixed explosive and the predicted functional additive proportion of the mixed explosive to obtain a mixed explosive formula. The method realizes the function of generating different explosive formulas according to different scenes, improves the formula generation efficiency and reduces the time and economic cost.

Description

Technical Field

[0001] This invention relates to the field of explosive formulation generation technology, and in particular to a method for rapidly generating mixed explosive formulations to meet the needs of multiple scenarios. Background Technology

[0002] Energetic materials, as a special type of material, have wide applications in defense, aerospace, and explosives. Particularly in the military field, energetic materials are the energy source for the destructive and lethal effects of weapon system explosions, forming the basis for the efficient destruction of weapon equipment and a key factor determining the power of weapon systems. However, the energy and safety of commonly used single-element explosives are contradictory and unsuitable for various complex application scenarios. Researchers typically design orthogonal experiments, collect experimental data, and then analyze the data to adjust the explosive formulation.

[0003] However, traditional methods such as orthogonal experiments rely on researchers constantly adjusting components through trial and error based on experience, resulting in low design efficiency. Furthermore, existing software lacks effective methods for rapidly designing formulations according to application requirements. Therefore, it is essential to design a rapid generation method for mixed explosive formulations tailored to multiple scenarios. Summary of the Invention

[0004] The purpose of this invention is to provide a method for rapidly generating mixed explosive formulations for multiple scenarios, so as to realize the function of generating different explosive formulations according to different scenarios, thereby improving the formulation generation efficiency and reducing time and economic costs.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for rapidly generating mixed explosive formulations to meet the needs of multiple scenarios includes the following steps:

[0007] Map the recipe requirements to objective and constraint functions;

[0008] The principal component ratio of the mixed explosive is obtained by solving the objective function and constraint function using a non-dominated sorting genetic algorithm.

[0009] The ratio of functional additives in mixed explosives is predicted by using multiple trained machine learning models, the proportion of main components in mixed explosives, and the performance indicators of mixed explosives.

[0010] The mixed explosive formulation is obtained by combining the main component ratio of the mixed explosive with the predicted functional additive ratio of the mixed explosive.

[0011] Optionally, the expression for the recipe requirement is:

[0012]

[0013] Where, p n f represents the proportion of the nth principal component in the mixed explosive; m Let t be the performance evaluation function for the m-th formulation. m Let m be the m-th performance index value, where m is the number of optimized performance requirements for the mixed explosive formulation. Let i be the minimum percentage that the i-th component can have. f1 represents the maximum possible proportion of the i-th component, where f1:+ indicates that the performance optimization direction is to maximize this proportion. m :- indicates that the optimization direction for the m-th performance metric is minimization.

[0014] Optionally, the objective function can be expressed as:

[0015]

[0016] Where, x n-1 This refers to the proportion of the (n-1)th principal component of the mixed explosive;

[0017] The expression for the constraint function is:

[0018]

[0019] in, The minimum percentage of the i-th principal component in the mixed explosive. x represents the maximum percentage of the i-th principal component in the mixed explosive. i Let i be the percentage of the i-th principal component in the mixed explosive. It represents the minimum percentage of the nth principal component in the mixed explosive.

[0020] Optionally, the specific steps of the non-dominated sorting genetic algorithm include:

[0021] The population was initialized using the Hua Luogeng optimal point set method to obtain the initial population.

[0022] Calculate the objective function value and constraint function violation value of the initial population, and perform non-dominated sorting and stratification of the initial population;

[0023] The initial population is selected, crossovered, and mutated through a tournament selection mechanism to generate offspring populations;

[0024] The initial population and the offspring population are merged to obtain a new population. The new population is then subjected to non-dominated sorting stratification and crowding calculation. The new parent population is selected based on the crowding and non-dominated sorting stratification results.

[0025] The tournament selection mechanism and population merging are repeated until the number of iterations reaches the preset maximum value.

[0026] Alternatively, the calculation formula for Hua Luogeng's optimal point set method is as follows:

[0027]

[0028] Where n is the dimension of the current space, and m is the population size. For the best point set, r ji Let a be the value of the j-th dimension of the i-th point, where mod is the modulo function, and a j b is the lower bound of the dimension. j The upper limit of the dimension, The value of the j-th dimension of the i-th individual Let j be the value of the j-th dimension of the i-th data point in the set of best points.

[0029] Optionally, the formula for calculating the constraint function violation value is:

[0030]

[0031] Among them, g j (x) represents the inequality constraint, h k (x) represents the equality constraint. <g j (x)> is if g j (x)≤0, then g j (x) = 0, otherwise g j (x)=|g j (x)|, where CV(x) is the constraint function violation value and x is the independent variable.

[0032] Optionally, the specific steps of a non-dominated sort include:

[0033] The best individual in the population is stored in a set, and the set is used as the first non-dominated layer.

[0034] Decrease the number of dominated individuals in the first non-dominated layer by 1, and use the reduced number of dominated individuals as the second non-dominated layer.

[0035] Alternatively, the formula for calculating congestion is:

[0036]

[0037] Where d represents the congestion level. Let be the maximum value of the i-th objective function across all solution sets. Let X be the minimum value of the i-th objective function in all solution sets. j_neighbor1 and X j_neighbor2 These are the two individuals that are closest to the objective function value, where n is the total performance and f is the value of the objective function. i (X j_neighbor1 ) and f i (X j_neighbor2 ) are respectively Xj_neighbor1 and X j_neighbor2 The corresponding objective function.

[0038] Optionally, the trained machine learning models include: linear regression model, random forest regression model, k-nearest neighbor regression model, and multilayer perceptron model;

[0039] The formula for calculating the linear regression model is: in, For model output, Let x be the eigenvector. n For the nth mixed explosive formulation feature, w represents the weight vector of the influence of different characteristics of mixed explosive formulations on the prediction of functional additive ratios. n is the influence weight of the nth mixed explosive formulation feature in the prediction of functional additive ratio, and b is the bias term;

[0040] The formula for calculating the random forest regression model is: in, The model output is X, where X is the formula data to be predicted, and f is the formula output. i (X) represents the output of the i-th decision tree, and m represents the number of decision trees;

[0041] The formula for calculating the k-nearest neighbor regression model is: in, For model output, y i Let be the label value of the i-th real recipe data, and k be the number of sample values ​​preset by the model;

[0042] The calculation formula for the multilayer perceptron model is as follows: in, Let w be the set of weight parameters for the j-th neuron in the i-th hidden layer. in Let be the weight parameters of the nth neuron in the i-th hidden layer, and Q be the node output. b is the output of the i-th hidden layer. in X is the bias term for the nth neuron in the i-th hidden layer. in Let σ be the output of the nth neuron in the i-th hidden layer, and let σ be the nonlinear transformation operation of the activation function.

[0043] A rapid generation system for mixed explosive formulations for multiple scenarios includes: an interaction module, a preprocessing module, and a model module;

[0044] The interactive module is used to set recipe requirements;

[0045] The preprocessing module maps the formulation requirements into objective and constraint functions and solves the objective and constraint functions using a non-dominated sorting genetic algorithm to obtain the principal component distribution ratio of the mixed explosive.

[0046] The model module is used to predict the proportion of functional additives in mixed explosives by using multiple trained machine learning models, the proportion of main components of mixed explosives, and the performance indicators of mixed explosives, so as to obtain the predicted proportion of functional additives in mixed explosives.

[0047] According to specific embodiments provided by the present invention, the following technical effects are disclosed: The present invention provides a rapid generation method for mixed explosive formulations for multiple scenarios. This method includes: mapping formulation requirements to an objective function and a constraint function; solving the objective function and constraint function using a non-dominated sorting genetic algorithm to obtain the principal component ratios of the mixed explosive; predicting the proportion of functional additives in the mixed explosive using multiple trained machine learning models, the principal component ratios of the mixed explosive, and performance indicators of the mixed explosive to obtain the predicted proportions of functional additives; and combining the principal component ratios of the mixed explosive and the predicted proportions of functional additives to obtain the mixed explosive formulation. This method enables the generation of different explosive formulations according to different scenarios, improving formulation generation efficiency and reducing time and economic costs. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart of a method for rapidly generating mixed explosive formulations for multiple scenarios according to an embodiment of the present invention;

[0050] Figure 2 This is a flowchart of the preprocessing process according to an embodiment of the present invention;

[0051] Figure 3 This is a flowchart of the non-dominated sorting genetic algorithm according to an embodiment of the present invention;

[0052] Figure 4 This is a structural diagram of a multilayer perceptron model according to an embodiment of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] like Figure 1 As shown, this embodiment of the invention provides a method for rapidly generating mixed explosive formulations for multiple scenarios, including the following steps:

[0056] Step 100: Map the recipe requirements to the objective function and constraint functions; specific steps are as follows: Figure 2 As shown, it includes:

[0057] Step 101: Unify the independent variables of the formula representing the main explosive performance index; the specific implementation process is as follows: First, define the main components of the mixed explosive formulation as (p1, p2, ..., p n ), where p i This represents the proportion of the i-th principal component, while p i The following conditions must be met: in, The minimum proportion of the i-th principal component The maximum proportion of the i-th principal component, and n is the number of principal components. Then, the principal components (p1, p2, ..., p...) are... n Unify them into independent variables (x1, x2, ..., x). n-1 ).

[0058] Specifically, the formulation requirements include the main explosive performance indicators, performance optimization directions, performance calculation formulas, and component variables. The expression for the formulation requirements is:

[0059]

[0060]

[0061] Where, p n f represents the proportion of the nth principal component in the mixed explosive; m Let t be the performance evaluation function for the m-th formulation. m Let m be the m-th performance index value, where m is the number of optimized performance requirements for the mixed explosive formulation. Let i be the minimum percentage that the i-th component can have. f1 represents the maximum possible proportion of the i-th component, where f1:+ indicates that the performance optimization direction is to maximize this proportion. m :- indicates that the optimization direction for the m-th performance metric is minimization.

[0062] Step 102: Combine the performance calculation formula and the performance optimization direction, and unify the optimization direction as minimization to construct the objective function;

[0063] Specifically, the expression for the objective function is:

[0064]

[0065] Where, x n-1 This refers to the proportion of the (n-1)th main component of the mixed explosive.

[0066] Step 103: Preprocess the objective function to obtain the constraint function; the specific steps are: construct the constraint function based on the performance indicators and the objective function, and unify the direction of the inequalities.

[0067] Specifically, the direction of the unified inequality is to split the minimization objective function into forms that are less than or equal to 0.

[0068] More specifically, the expression for the constraint function is:

[0069]

[0070] in, x represents the minimum percentage of the i-th principal component in the mixed explosive. i Let i be the percentage of the i-th principal component in the mixed explosive. It represents the minimum percentage of the nth principal component in the mixed explosive.

[0071] Step 200: Solve the objective function and constraint function using a non-dominated sorting genetic algorithm to obtain the principal component distribution ratio of the mixed explosive;

[0072] Specifically, the steps of the non-dominated sorting genetic algorithm are as follows: Figure 3 As shown, it includes:

[0073] Step 201: Initialize the population using Hua Luogeng's optimal point set method to obtain the initial population;

[0074] Furthermore, random initialization is generally used for population initialization. However, randomly generated populations are unevenly distributed throughout the solution space, sometimes overly clustered in certain regions and overly dispersed in others, resulting in low utilization of the overall space. In this embodiment, to better find the globally optimal solution set, the Hua Luogeng optimal point set method is used to sample a uniformly distributed initial population from the n-dimensional space of the independent variable distribution. The calculation formula for the Hua Luogeng optimal point set method is:

[0075]

[0076]

[0077] Where n is the dimension of the current space, and m is the population size. For the best point set, r ji Let a be the value of the j-th dimension of the i-th point, where mod is the modulo function, and a j b is the lower bound of the dimension. j The upper limit of the dimension, The value of the j-th dimension of the i-th individual Let j be the value of the j-th dimension of the i-th data point in the set of best points.

[0078] Step 202: Calculate the objective function value and constraint function violation value of the initial population, and perform non-dominated sorting and stratification of the initial population;

[0079] Specifically, the constraint function violation value is used to quantitatively describe the degree to which a solution violates constraints, and the calculation formula is as follows:

[0080]

[0081] Among them, g j (x) represents the assumed inequality constraint, h k (x) represents the assumed equality constraint. <g j (x)> is if g j (x)≤0, then g j (x) = 0, otherwise g j (x)=|g j (x)|, where CV(x) is the constraint function violation value and x is the independent variable.

[0082] More specifically, the smaller the CV value, the lower the degree to which the solution violates the constraints, and the better the solution. At the same time, when a solution is feasible, its CV value is 0, and when it is infeasible, its CV value is greater than 0.

[0083] Furthermore, to incorporate the degree of constraint violation into the adaptive consideration, a penalty term is added to the objective function of all solutions; the formula for calculating the penalty term is:

[0084] p = ρ × CV(x);

[0085] Here, p is the penalty term, and ρ is a relatively large value relative to the range of the objective function, i.e., the penalty coefficient. The greater the degree to which the solution violates the constraints, the more the penalty term increases the objective function value of the solution, making it more likely to be at a disadvantage in subsequent non-dominated comparisons, resulting in lower fitness than other solutions and making it more likely to be eliminated in the iteration.

[0086] Furthermore, since equality constraints are more restrictive than inequality constraints, using too small a penalty coefficient will weaken the constraint, while using too large a coefficient will cause even minor constraint violations to result in a penalty term that is much larger than the objective function value, to the point that the objective function degenerates into a penalty function. This makes the algorithm more inclined to seek solutions that satisfy equality constraints rather than solutions to the optimization problem during the iteration process.

[0087] This embodiment uses an equality removal process to handle equality constraints, thus eliminating equality constraints. For example, the equality constraints are incorporated into the objective function, and the x-related constraints in the objective function are included. n Partial replacement with Simultaneously, equality constraints will be applied. Convert to This ensures that the processed solution automatically satisfies the equality constraints.

[0088] Specifically, the implementation process of non-dominated sort is as follows:

[0089] First, the non-dominated sorting algorithm assigns two parameters N to each individual (solution) in the population. i and S i N i S represents the number of individuals (solutions) that dominate the i-th individual (solution). i This represents the number of individuals (solutions) dominated by the i-th individual (solution);

[0090] Secondly, all those satisfying N i Individuals (solutions) with a value of 0 are stored in a set as the first non-dominated layer, such that the individuals (solutions) in this set constitute the current optimal set of individuals (solution set); where N i =0 indicates that there are no other individuals (solutions) in the current population (solution set) that are better than this individual (solution);

[0091] Then, the S of the first non-dominated layer i Decrease by 1, and satisfy S i Individuals with -1 = 0 are stored in a set as the second non-dominated layer, where S i -1 represents S after subtracting 1. i ;

[0092] Finally, the subsequent non-dominated layers are constructed sequentially according to the above non-dominated sorting algorithm, so that the solutions in the lower non-dominated layers are better than the solutions in the higher layers.

[0093] Step 203: Select, crossover, and mutate the initial population through a tournament selection mechanism to generate a progeny population; the specific steps are as follows: randomly select individuals with good fitness as parents through a tournament selection operator to crossover and mutate, so as to breed new individuals to form a progeny population.

[0094] Specifically, fitness is represented by a fitness function, where the fitness of the initial reproduction is represented by the non-dominant layer level of the individual, and the smaller the value, the better; the fitness of subsequent reproduction is represented by a partial order relation; that is, the smaller the dominant layer level, the higher the fitness, and the greater the crowding at the same layer level, the higher the fitness.

[0095] Step 204: Merge the initial population and the offspring population to obtain a new population, perform non-dominated sorting stratification and crowding calculation on the new population, and select a new parent population based on the crowding and stratification results; the size of both the initial population and the offspring population is M.

[0096] Specifically, if the number of individuals in the current non-dominant layer is less than M, then all individuals in that layer are retained in the new parent population; otherwise, they are included in the new parent population in descending order of crowding, until the size of the new parent population reaches M.

[0097] Specifically, the crowding degree is calculated as follows: First, calculate the function values ​​of all objective functions. Then, find two individuals (solutions) with similar function values ​​and calculate their differences. The crowding degree of this individual (solution) is the sum of the differences between all adjacent individuals (solutions) of the objective function. A lower crowding degree indicates that there are many similar individuals nearby. Generally, individuals with higher crowding degrees are preferred because their surroundings are more open and have greater search value. The specific calculation formula is:

[0098]

[0099] Where d represents the congestion level. Let be the maximum value of the i-th objective function across all solution sets. Let X be the minimum value of the i-th objective function in all solution sets, m be the population size, and X be the minimum value of the i-th objective function. j_neighbor1 and X j_neighbor2Let `neighbor1` and `neighbor2` be the objective functions corresponding to `neighbor1` and `neighbor2`, respectively. `neighbor1` and `neighbor2` are the two individuals (solutions) whose objective function values ​​are closest to the objective function value of this individual (solution). The former is the closest individual (solution) among all those whose objective function value is greater than the objective function value of this individual (solution), and the latter is the closest individual (solution) among all those whose objective function value is less than the objective function value of this individual (solution). For boundary points, their crowding degree is set to infinity. The denominator of the calculation formula is used for normalization because different objective function values ​​have different ranges and may vary significantly. If normalization is not performed, the objective function with a larger value will dominate the crowding degree value.

[0100] More specifically, this embodiment also provides two new crowding calculation formulas for recipe design in certain scenarios:

[0101]

[0102] Furthermore, when there are specific expected values ​​for various performance values ​​of the formulation, and it is desired to improve the fitness of solutions similar to the expected vector, a new crowding formula (1) is adopted. Where, d max is the maximum crowding degree among all individuals (solutions) in the level to which this individual (solution) is located, d is the value calculated according to the original crowding degree, and sim is the expected approximation of the solution set (with a value of 0 to 1).

[0103] Furthermore, the new congestion formula (1) is divided into distance approximation and cosine approximation; assuming Let be the expected vector. The objective function value vector for an individual (solution).

[0104] The formula for calculating the approximate distance is:

[0105]

[0106]

[0107] Where distance_max represents the vector of objective function values ​​for all individuals in the non-dominated layer of the individual. With expectation vector The maximum distance, Represents the vector of individual objective function values With expectation vector The distance is given by e1, where e1 represents the expected energy value of a certain formula, v1 represents the objective function value of that performance obtained through individual solutions, and pre represents the approximation between the objective function value vector and the expected vector, i.e., the approximation between the solved performance value and the expected performance value. The reference standard is the solution vector that is furthest from the expected vector among all solutions. When The closer the value of the objective function vector is to 0, that is, the closer the individual (solution) objective function value vector is to the expected vector, the greater the fitness improvement.

[0108] The formula for calculating the cosine approximation is:

[0109]

[0110] The larger the cosine approximation, the greater the improvement in fitness.

[0111] Furthermore, when it is desired to increase the priority of a certain performance of the formula during the search process, a new crowding formula (2) is adopted. Here, f(X) is the objective function value of the individual with priority.

[0112] Step 205: Repeat steps 203 and 204 until the number of iterations reaches the preset maximum value.

[0113] Step 300: Predict the proportion of functional additives in the mixed explosive by using multiple trained machine learning models, the proportion of main components of the mixed explosive, and the performance indicators of the mixed explosive, and obtain the predicted proportion of functional additives in the mixed explosive.

[0114] Specifically, the trained machine learning models include: linear regression models, random forest regression models, k-nearest neighbor regression models, and multilayer perceptron models;

[0115] The linear regression model considers the influence of multiple input variables on the output variable, and its calculation formula is as follows:

[0116]

[0117] in, For model output, Let x be the eigenvector. n For the nth mixed explosive formulation feature, w represents the weight vector of the influence of different characteristics of mixed explosive formulations on the prediction of functional additive ratios. n is the influence weight of the nth mixed explosive formulation feature in the prediction of functional additive ratio, and b is the bias term; and b are determined by minimizing the loss function, which is calculated using the following formula:

[0118]

[0119] Among them, y i Label values ​​representing actual formula data (such as sensitivity, energy). The model output is for the i-th input variable.

[0120] The random forest regression model consists of multiple decision trees. Multiple subsets of real recipe data are generated through bootstrapping to train these decision trees. The final model output is the average of the outputs of all decision trees, calculated using the following formula:

[0121]

[0122] in, The model output is X, where X is the formula data to be predicted, and f is the formula output. i (X) represents the output of the i-th decision tree, and m is the number of decision trees; the loss function of the random forest regression model is calculated as follows:

[0123]

[0124] R1(k,v)=X|X k ≤v, R2(k,v)=X|X k >v;

[0125]

[0126] Where R1 and R2 represent dividing the value v of feature k into two intervals, X i Let c be the data in this interval, c be the average of the true formula data label values ​​of all data in this interval, N be the number of data points in the interval, c1 be the c value of interval R1, and c2 be the c value of interval R2.

[0127] The formula for calculating the k-nearest neighbor regression model is:

[0128]

[0129] in, For model output, y i Let be the label value of the i-th real formula data, and k be the number of sample values ​​preset by the model. The value of k is determined by grid optimization.

[0130] like Figure 4 As shown, the multilayer perceptron model is a neural network structure containing multiple layers, including an input layer, at least one hidden layer, and an output layer. The input layer takes the main explosive ratio, binder ratio, and target performance data as input, passes through the hidden layer, and outputs the functional additive ratio. Its calculation formula is:

[0131]

[0132]

[0133] in, Let w be the set of weight parameters for the j-th neuron in the i-th hidden layer. inLet be the weight parameters of the nth neuron in the i-th hidden layer, and Q be the node output. b is the output of the i-th hidden layer. in X is the bias term for the nth neuron in the i-th hidden layer. in Let σ be the output of the nth neuron in the i-th hidden layer, and let σ be the nonlinear transformation operation of the activation function.

[0134] Step 400: Combine the proportions of the main components of the mixed explosive with the predicted proportions of the functional additives of the mixed explosive to obtain the mixed explosive formulation.

[0135] This invention also provides a rapid generation system for mixed explosive formulations for multiple scenarios, including: an interaction module, a preprocessing module, and a model module;

[0136] The interactive module is used by users to configure the requirements of the formula to be designed, including the determination of components, the determination of the complete mixed explosive formula index and the main explosive index, the determination of the component proportion range, the selection of the performance optimization direction, and the selection of the performance calculation formula;

[0137] The preprocessing module maps the formulation requirements into objective and constraint functions and solves the objective and constraint functions using a non-dominated sorting genetic algorithm to obtain the principal component distribution ratio of the mixed explosive.

[0138] The model module is used to predict the proportion of functional additives in mixed explosives by using multiple trained machine learning models, the proportion of main components of mixed explosives, and the performance indicators of mixed explosives, so as to obtain the predicted proportion of functional additives in mixed explosives.

[0139] The beneficial effects of this invention are as follows:

[0140] 1) The ability to generate different explosive formulas based on different scenarios;

[0141] 2) Improved formulation generation efficiency and reduced time and economic costs.

[0142] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0143] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for rapidly generating mixed explosive formulations to meet the needs of multiple scenarios, characterized in that, Includes the following steps: The recipe requirements are mapped to an objective function and a constraint function; the expression for the recipe requirements is: ； ； ； :+，.......， :-； in, The ratio of the nth principal component of the mixed explosive; Let m be the performance evaluation function for the formulation. Let m be the m-th performance index value, where m is the number of optimized performance requirements for the mixed explosive formulation. Let i be the minimum percentage that the i-th component can have. f1 represents the maximum possible proportion of the i-th component, where f1:+ indicates that the performance optimization direction is to maximize this proportion. m :- indicates that the optimization direction for the m-th performance metric is minimization; The expression for the objective function is: ； in, This refers to the proportion of the (n-1)th principal component of the mixed explosive; The expression for the constraint function is: ； in, The minimum percentage of the i-th principal component in the mixed explosive. This represents the maximum percentage of the i-th principal component in the mixed explosive. Let i be the percentage of the i-th principal component in the mixed explosive. This represents the minimum percentage of the nth principal component in the mixed explosive; The objective function and the constraint function are solved using a non-dominated sorting genetic algorithm to obtain the principal component ratio of the mixed explosive; the specific steps of the non-dominated sorting genetic algorithm include: The population was initialized using the Hua Luogeng optimal point set method to obtain the initial population. Calculate the objective function value and constraint function violation value of the initial population, and perform non-dominated sorting and stratification on the initial population; The initial population is selected, crossovered, and mutated through a tournament selection mechanism to generate a progeny population; The initial population and the offspring population are merged to obtain a new population. The non-dominated sorting stratification and crowding degree are calculated on the new population, and a new parent population is selected based on the crowding degree and non-dominated sorting stratification results. Repeat the tournament selection mechanism and the population merging until the number of iterations reaches the preset maximum value; The ratio of functional additives in the mixed explosive is predicted by using multiple trained machine learning models, the main component ratio of the mixed explosive, and the performance index of the mixed explosive. The mixed explosive formulation is obtained by combining the main component ratio of the mixed explosive with the predicted functional additive ratio of the mixed explosive.

2. The method for rapid generation of mixed explosive formulations for multiple scenarios according to claim 1, characterized in that, The calculation formula for Hua Luogeng's optimal point set method is as follows: ； ； ； Where n is the dimension of the current space, and m is the population size. For the best collection of points, Let be the value of the j-th dimension at the i-th point, and mod be the modulo function. As the lower bound of the dimension, The upper limit of the dimension, The value of the j-th dimension of the i-th individual Let j be the value of the j-th dimension of the i-th data point in the set of best points.

3. The method for rapid generation of mixed explosive formulations for multiple scenarios according to claim 1, characterized in that, The formula for calculating the constraint function violation value is as follows: ； in, For inequality constraints, For equality constraints, If ,but =0, otherwise = , Let x be the value of the constraint function violation, and let x be the independent variable.

4. The method for rapid generation of mixed explosive formulations for multiple scenarios according to claim 1, characterized in that, The specific steps of the non-dominated sorting include: The best individual in the population is stored in a set, and the set is used as the first non-dominated layer; The number of dominated individuals in the first non-dominated layer is reduced by 1, and the reduced number of dominated individuals is used as the second non-dominated layer.

5. The method for rapid generation of mixed explosive formulations for multiple scenarios according to claim 1, characterized in that, The formula for calculating the congestion level is: ； Where d represents the congestion level. Let be the maximum value of the i-th objective function across all solution sets. Let be the minimum value of the i-th objective function across all solution sets. and These are the two individuals that are closest to the objective function value, and n is the total performance. and They are respectively and The corresponding objective function value.

6. The method for rapid generation of mixed explosive formulations for multiple scenarios according to claim 1, characterized in that, The trained machine learning models include: linear regression model, random forest regression model, k-nearest neighbor regression model, and multilayer perceptron model; The calculation formula for the linear regression model is as follows: ; ; ;in, For model output, For feature vectors, For the nth mixed explosive formulation feature, This represents the weight vector of the influence of different characteristics of mixed explosive formulations on the prediction of functional additive ratios. is the influence weight of the nth mixed explosive formulation feature in the prediction of functional additive ratio, and b is the bias term; The calculation formula for the random forest regression model is as follows: ;in, For model output, X For the formula data to be predicted, Let m be the output of the i-th decision tree, and m be the number of decision trees. The calculation formula for the k-nearest neighbor regression model is as follows: ;in, For model output, For the first i The label value of a real formula data, k The preset number of sample values ​​for the model; The calculation formula for the multilayer perceptron model is as follows: ; ; ;in, Let be the set of weight parameters for the j-th neuron in the i-th hidden layer. Let be the weight parameters of the nth neuron in the i-th hidden layer, and Q be the node output. This is the output of the i-th hidden layer. Let n be the bias term for the nth neuron in the i-th hidden layer. This represents the output of the nth neuron in the i-th hidden layer. This is a nonlinear transformation operation for the activation function.

7. A rapid generation system for mixed explosive formulations for multiple scenarios, applied to the rapid generation method for mixed explosive formulations for multiple scenarios as described in any one of claims 1-6, characterized in that, include: Interaction module, preprocessing module, and model module; The interactive module is used to set the recipe requirements; The preprocessing module is used to map the formulation requirements into objective functions and constraint functions, and solve the objective functions and constraint functions using a non-dominated sorting genetic algorithm to obtain the principal component distribution ratio of the mixed explosive; The model module is used to predict the proportion of functional additives in the mixed explosive by using multiple trained machine learning models, the proportion of the main components of the mixed explosive, and the performance indicators of the mixed explosive, so as to obtain the predicted proportion of functional additives in the mixed explosive.

Images