A hollow explosive grain density prediction method and device

Abstract

The application discloses a hollow explosive grain density prediction method and device, through improving the crossover rate and mutation rate in the GA-BP algorithm, embedding the improved crossover rate and mutation rate into a genetic algorithm (GA) to realize improvement of the GA-BP network model to obtain an IGA-BP model. The improved IGA algorithm improves the training convergence speed of the GA optimization algorithm, improves the accuracy of network weight of each layer of the GA optimized BP neural network, reduces the prediction error, and thus establishes an accurate hollow explosive grain density and uniformity prediction model. The IGA-BP model has higher prediction accuracy and higher model stability, has practical significance for explosive grain density prediction, and provides a method and idea for subsequent quality detection and process parameter adjustment.

Description

Technical Field

[0001] This invention relates to the field of shaped charge quality prediction technology, and in particular to a method and apparatus for predicting the density of hollow explosive charge based on the IGA-BP network model. Background Technology

[0002] With the widespread application of finite element simulation and artificial intelligence technologies in various fields, it has become possible to numerically simulate the explosive compression molding process using simulation methods. This not only helps reduce R&D costs and shorten the R&D cycle, but also enables rapid development of loading processes. Furthermore, combining simulation results with advanced algorithms such as neural networks can achieve accurate prediction of the quality of the molded explosive charge. Therefore, this method is of vital importance for optimizing the loading process, predicting explosive quality, and solving problems such as lagging manufacturing processes, inconsistent quality, and excessively long R&D cycles.

[0003] Traditional backpropagation (BP) neural networks face challenges in initializing parameters because these parameters are typically initialized randomly based on data characteristics, which may not be optimal. To address this issue, we can use a genetic algorithm (GA) to find the optimal neural network parameters.

[0004] However, traditional genetic algorithms may also face the problem of improperly adjusted step size in the later stages of iteration. This can lead to an excessively long search step size near the optimal solution, preventing individuals from converging quickly to the optimal solution. Instead, they may oscillate around the optimal solution, resulting in a slower convergence speed. Summary of the Invention

[0005] In view of the above problems, this invention provides a method and apparatus for predicting the density of hollow explosive charge to overcome or at least partially solve these problems. The optimal parameters are quickly solved by adaptively modifying the crossover and variation rates within the GA system, and the optimal parameter values ​​are input into the BP model, thereby improving the model. The established prediction model of explosive pressing process parameters and hollow charge density predicts the density of the pressed hollow charge, providing advance conditions for adjusting process parameters in the subsequent explosive pressing process.

[0006] This invention provides the following solution:

[0007] A method for predicting the density of a hollow explosive charge includes:

[0008] Obtain an input parameter set, which includes several hollow explosive charge compression process parameters;

[0009] The input parameter set is input into the trained target prediction model so that the target prediction model outputs the hollow explosive charge density prediction result; the target prediction model includes a neural network model and an improved genetic algorithm, the improved genetic algorithm is used to find the optimal parameters of the neural network model, and the improved genetic algorithm includes an improved crossover rate and an improved mutation rate;

[0010] The improved crossover rate and the improved mutation rate are obtained in the following ways:

[0011] By combining the average fitness value of the previous generation of chromosomes with the nonlinear action function of neurons, the improved crossover rate and the improved mutation rate, both of which can be adaptively adjusted, are obtained.

[0012] Preferably, the nonlinear action function is expressed by the following formula:

[0013]

[0014] The average fitness value of the previous generation chromosome set is expressed by the following formula:

[0015]

[0016] In the formula: best_finess(i) is the best fitness value in the i-th population of the previous generation, num is the total number of iterations, ave_errf is the average fitness difference of chromosome groups, and group_num is the number of populations.

[0017] Preferably, the improved crossover rate is expressed by the following formula:

[0018]

[0019] In the formula: when i=1, pc is 0.1, and worst_finess(i) is the maximum fitness value in the i-th chromosome set of the previous generation.

[0020] Preferably, the improved mutation rate is expressed by the following formula:

[0021]

[0022] In the formula: when i = 1, pm is 0.1.

[0023] Preferably, the training method for the target prediction model includes:

[0024] Input a training set, which includes several samples;

[0025] Determine the structure of the neural network model;

[0026] The improved genetic algorithm is initialized by selecting the number of chromosome groups and the number of iterations, and randomly generating the gene values ​​in each chromosome.

[0027] Calculate the current fitness of each chromosome;

[0028] Update the improved crossover rate and improved mutation rate;

[0029] Update chromosome set;

[0030] Calculate and update the optimal fitness value;

[0031] The optimal chromosome is found through iterations and used as the network output, and the model parameters are saved.

[0032] Preferably, the fitness is calculated using the following formula:

[0033]

[0034] In the formula, f(ε) i y represents the fitness of the i-th chromosome. k Let be the predicted value of the i-th chromosome in the k-th sample group. Let be the actual value of the i-th chromosome in the k-th sample group, and n be the sample size.

[0035] Preferably: Chromosomes are updated using the following formula:

[0036]

[0037]

[0038] In the formula: pc is the crossover rate; fg = pmg(1 - g / G) max ) 2 pm is the mutation rate, g is the current iteration number, and G max Let r be the maximum number of iterations, and r be a random number ∈ [0,1].

[0039] Preferably: the samples are normalized, and the normalization is achieved by the following formula:

[0040]

[0041] In the formula: x is the sample parameter, x min and x max These represent the minimum and maximum values ​​of the experimental data, y. max y is 1 min It is -1.

[0042] Preferably, the process parameters for pressing the hollow explosive charge include material preheating temperature, mold preheating temperature, pressing pressure, vacuum time, holding time, vacuum degree, and finished product density.

[0043] A device for predicting the density of a hollow explosive charge, comprising:

[0044] An input parameter set acquisition unit is used to acquire an input parameter set, which includes several hollow explosive charge compression process parameters.

[0045] The hollow explosive charge density prediction result output unit is used to input the input parameter set into the trained target prediction model so that the target prediction model outputs the hollow explosive charge density prediction result; the target prediction model includes a neural network model and an improved genetic algorithm, the improved genetic algorithm is used to find the optimal parameters of the neural network model, and the improved genetic algorithm includes an improved crossover rate and an improved mutation rate;

[0046] The improved crossover rate and the improved mutation rate are obtained in the following ways:

[0047] By combining the average fitness value of the previous generation of chromosomes with the nonlinear action function of neurons, the improved crossover rate and the improved mutation rate, both of which can be adaptively adjusted, are obtained.

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

[0049] This application provides a method and apparatus for predicting the density of hollow explosive charge. By improving the crossover and mutation rates in the GA-BP algorithm and embedding the improved crossover and mutation rates into a genetic algorithm (GA), an improved IGA-BP model is obtained. The improved IGA algorithm enhances the training convergence speed of the GA optimization algorithm. It also improves the accuracy of the weights in each layer of the GA-optimized BP neural network, reducing prediction errors and thus establishing an accurate prediction model for the density and uniformity of hollow explosive charge. The IGA-BP model not only has higher prediction accuracy but also significantly improved model stability, which is of practical significance for predicting explosive charge density and provides methods and ideas for subsequent quality inspection and process parameter adjustment.

[0050] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0052] Figure 1 This is a flowchart of a method for predicting the density of a hollow explosive charge provided in an embodiment of the present invention;

[0053] Figure 2 This is a flowchart of the training process for a hollow explosive charge density prediction method provided in an embodiment of the present invention;

[0054] Figure 3 This is an improved GA optimization flowchart provided in an embodiment of the present invention;

[0055] Figure 4 This is a schematic diagram of the BP network structure provided in an embodiment of the present invention;

[0056] Figure 5 This is a schematic diagram comparing the prediction results of GA-BP and IGA-BP provided in an embodiment of the present invention;

[0057] Figure 6 This is another schematic diagram showing the comparison of GA-BP and IGA-BP prediction results provided in this embodiment of the invention;

[0058] Figure 7 This is a schematic diagram of a hollow explosive charge density prediction device provided in an embodiment of the present invention;

[0059] Figure 8 This is a schematic diagram of a hollow explosive charge density prediction device provided in an embodiment of the present invention. Detailed Implementation

[0060] 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 a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0061] See Figure 1 This invention provides a method for predicting the density of a hollow explosive charge, as exemplified by [example of the invention]. Figure 1 As shown, the method may include:

[0062] S101: Obtain an input parameter set, which includes several hollow explosive charge pressing process parameters; in specific implementation, the hollow explosive charge pressing process parameters in this application embodiment may include material preheating temperature, mold preheating temperature, pressing pressure, vacuum time, holding time, vacuum degree and finished product density.

[0063] S102: Input the set of input parameters into the trained target prediction model so that the target prediction model outputs the hollow explosive charge density prediction result; the target prediction model includes a neural network model and an improved genetic algorithm, the improved genetic algorithm is used to find the optimal parameters of the neural network model, and the improved genetic algorithm includes an improved crossover rate and an improved mutation rate;

[0064] The improved crossover rate and the improved mutation rate are obtained in the following ways:

[0065] By combining the average fitness value of the previous generation of chromosomes with the nonlinear action function of neurons, the improved crossover rate and the improved mutation rate, both of which can be adaptively adjusted, are obtained.

[0066] The method for predicting the density of hollow explosive charge provided in this application improves the crossover rate (pc) and mutation rate (pm) in the GA algorithm (genetic algorithm) to obtain improved crossover and mutation rates. These improved pc and pm are then embedded into the GA algorithm to replace the original crossover and mutation rates, resulting in an improved IGA-BP network. This IGA-BP network model is then used to predict the density of the hollow explosive charge. The improved IGA-BP network increases the training convergence speed of the GA optimization algorithm. It also improves the accuracy of the weights in each layer of the GA-optimized BP neural network, reducing prediction errors and thus establishing an accurate prediction model for the density and uniformity of the hollow explosive charge.

[0067] The nonlinear action function (sigmoid function) is expressed by the following formula:

[0068]

[0069] The average fitness value of the previous generation chromosome set is expressed by the following formula:

[0070]

[0071] In the formula: best_finess(i) is the best fitness value in the i-th population of the previous generation, num is the total number of iterations, ave_errf is the average fitness difference of chromosome groups, and group_num is the number of populations.

[0072] Furthermore, the improved crossover rate is expressed by the following formula:

[0073]

[0074] In the formula: when i = 1, pc is 0.1, and worst_finess(i) is the maximum fitness value in the i-th chromosome set of the previous generation.

[0075] The improved mutation rate is expressed by the following formula:

[0076]

[0077] In the formula: when i = 1, pm is 0.1.

[0078] The training method for the target prediction model includes:

[0079] Input a training set, which includes several samples; in practical applications, normalize the samples using the following formula:

[0080]

[0081] In the formula: x is the sample parameter, x min and x max These represent the minimum and maximum values ​​of the experimental data, y. max y is 1 min It is -1.

[0082] Determine the structure of the neural network model;

[0083] The improved genetic algorithm is initialized by selecting the number of chromosome sets and the number of iterations, and randomly generating the values ​​of each gene in each chromosome.

[0084] Calculate the current fitness of each chromosome;

[0085] Update the improved crossover rate and improved mutation rate;

[0086] Update chromosome set;

[0087] Calculate and update the optimal fitness value;

[0088] The optimal chromosome is found through iterations and used as the network output, and the model parameters are saved.

[0089] The fitness is calculated using the following formula:

[0090]

[0091] In the formula, f(ε) i y represents the fitness of the i-th chromosome. k Let be the predicted value of the i-th chromosome in the k-th sample group. Let be the actual value of the i-th chromosome in the k-th sample group, and n be the sample size.

[0092] Update chromosomes using the following formula:

[0093]

[0094]

[0095] In the formula: pc is the crossover rate; fg = pmg(1 - g / G) max ) 2 pm is the mutation rate, g is the current iteration number, and G max Let r be the maximum number of iterations, and r be a random number ∈ [0,1].

[0096] The methods provided in the embodiments of this application will be described in detail below, such as... Figure 2 As shown.

[0097] The GA algorithm is an algorithm that finds the global optimal solution to a problem by simulating natural selection and mating variation in biological evolution.

[0098] Embedding pc and pm into the GA algorithm involves using MATLAB software to write the improved crossover rate and mutation rate as program modules, which are then added to GA to replace the default crossover rate and mutation rate.

[0099] A BP neural network is a multilayer feedforward neural network that uses an error backpropagation algorithm for error correction.

[0100] The parameters are the weight coefficients and biases between neurons in each layer of the BP network.

[0101] GA-BP neural network refers to the improved GA-BP neural network model that uses GA to find the optimal parameters of BP.

[0102] Methods to improve the crossover rate (pc) include:

[0103] By combining dynamic discreteness with the sigmoid function, a nonlinear dynamic adaptive method is obtained, which improves the pc in the GA algorithm, resulting in the improved pc.

[0104] The pc controls the frequency of crossover operations. If the pc is too large, it increases the search intensity of the algorithm but also affects the overall search efficiency. If the pc is too small, the search will stagnate.

[0105] The sigmoid function refers to The nonlinear action function of a neuron.

[0106] The PC improvement method includes the following steps:

[0107] Average fitness value of the previous generation chromosome set:

[0108]

[0109] In the formula: best_finess(i) is the best fitness value in the i-th population of the previous generation, and similarly, worst_finess(i) is the largest fitness value in the i-th chromosome set of the previous generation, num is the total number of iterations, ave_errf is the average fitness difference of chromosome sets, and group_num is the number of populations;

[0110] To calculate the dynamically adaptive pc, combine ave_errf with the sigmoid function to obtain the formula for the adaptively adapted pc:

[0111]

[0112] In the formula: when i = 1, pc is 0.1.

[0113] Methods to improve the mutation rate (pm) include:

[0114] To calculate the dynamically adaptively adjusted PM, combine ave_errf with the sigmoid function to obtain the formula for the adaptively adjusted PM:

[0115]

[0116] In the formula: when i = 1, pm is 0.1.

[0117] like Figure 3 As shown, the improved pc and improved pm are embedded into the GA algorithm to obtain the improved GA. The optimal parameters of the BP network model are found using the improved GA algorithm, and the GA-BP network model is established for the prediction of the density of hollow explosive charge.

[0118] After improving the model, it can be trained. The training method is as follows:

[0119] The training set is normalized and input into the training set to train the GA-BP network model to obtain the optimized model. Then, the model is tested on the test set to output the predicted density of the hollow explosive charge.

[0120] The normalization refers to the use of a function to process the evolution of samples, as expressed by the following formula:

[0121]

[0122] In the formula, x is the sample parameter, x min and x min These represent the minimum and maximum values ​​of the experimental data, y. max y is 1 min =-1;

[0123] The method for predicting the density of hollow explosive charge includes the following steps:

[0124] S41. The input sample parameters of the network model include material preheating temperature, mold preheating temperature, pressing pressure, vacuum time, holding time, vacuum degree, and finished product density.

[0125] S42. Normalize the input samples and divide the processed samples into training set and test set;

[0126] S43. Train the GA-BP network using the training set to obtain the network parameters;

[0127] S44. Load the network model parameters and use the test set to predict the backpropagation (BP).

[0128] The GA-BP network is trained using the training set to obtain the network model, including the following steps:

[0129] S4301. Input the training set, which contains 456 sets of data.

[0130] S4302. Determine the BP network structure; set the input layer to 6 nodes, and the output result to be the predicted density or uniformity, which is a single output. Therefore, determine the number of hidden layer neurons to be 10. The structure diagram of the BP network is shown in Appendix 4. Use the GA algorithm to optimize the structural parameters of the BP network model. In essence, the dimension of the chromosome is the sum of the number of weights and biases to be optimized in the BP network.

[0131] The number of hidden layer nodes is determined using the following formula.

[0132]

[0133] In the formula: n is the number of hidden layer nodes, m is the number of input layer nodes, p is the number of output layer nodes, and d is a preset constant; the activation function of the hidden layer is set to the hyperbolic tangent function Tansig, and the activation function of the output layer is the linear function Purelin.

[0134] The chromosomes are initialized to determine the number of chromosome sets and the number of iterations. The training set is input, the current fitness of the chromosome is calculated according to the fitness function, and the current fitness is compared with the optimal fitness of the previous iteration to obtain the optimal fitness value. The chromosome position corresponding to the optimal fitness is updated according to the comparison result.

[0135] S4303. Initialize the GA algorithm, select the number of chromosome groups as m, the initial generation iteration number as t, randomly generate the value of each "gene" in each chromosome, and set the gene value range as (1,10); calculate the fitness value corresponding to the initial chromosome through the fitness calculation formula, use the loop statement to calculate the fitness value corresponding to all initialized chromosomes in the chromosome group, and select the smallest one as the initial optimal fitness value.

[0136] The parameters to be optimized in the BP neural network (weights and biases of each layer) are encoded into chromosomes represented by real-valued strings. Simultaneously, a certain number of chromosomes are randomly generated to form an initial chromosome set. Since the output node has 1, the number of output weights is the same as the number of hidden layer neurons. The number of input layer nodes is intnum, and the number of hidden layer nodes is hidnum. Each input node must interact with all hidden layer neurons once, so the number of data center parameters is hidnum * intnum. The dimension of the chromosome is Lnum, specifically: Lnum = hidnum * intnum + hidnum + hidnum * 1 + 1.

[0137] S4304. Perform a selection operation on the chromosome set. Select chromosomes from the initial fitness values ​​of m chromosomes using a roulette wheel selection algorithm to form a new chromosome set. Calculate the current fitness f(ε) of each chromosome. i The formula for calculating the fitness function is as follows:

[0138]

[0139] In the formula, f(ε) i y represents the fitness of the i-th chromosome. k Let be the predicted value of the i-th chromosome in the k-th sample group. Let be the actual value of the i-th chromosome in the k-th sample group, and n be the sample size.

[0140] S4305. Update pc and pm according to the above adaptive method;

[0141] S4306. Update the chromosome set by performing crossover mutation operations on the new chromosome set according to the crossover mutation method described above; the chromosome update formula is as follows:

[0142]

[0143]

[0144] In the formula, pc is the crossover probability; the mutation operation selects the j-th gene on the i-th chromosome for mutation; fg = pmg(1-g / G) max ) 2 pm is the mutation probability, g is the current iteration number, and G max Let r be the maximum number of iterations, and r be a random number ∈ [0,1].

[0145] S4307. Next, calculate the fitness value corresponding to the new chromosome set, compare the fitness value of this set with the initial optimal fitness value, and update the optimal fitness value according to the formula shown:

[0146]

[0147] In the formula: f b f is the current optimal fitness value. n (i) represents the fitness value of the i-th chromosome in the current chromosome set; update the optimal fitness value and save the chromosome corresponding to the optimal fitness value.

[0148] S4308. Loop through and find the optimal chromosome as the network output, and save the model parameters. Iterate through and update the chromosome set. If the optimization reaches the maximum number of iterations, the process ends and the chromosome corresponding to the current optimal fitness value is returned. Otherwise, t = t + 1, jump to S4205, continue to update the chromosome set and calculate the corresponding fitness value. Use the optimal chromosome as the output of the GA network and input it into the BP neural network to obtain the optimized BP prediction model.

[0149] The improved GA algorithm updates the mutation rate of the current chromosome based on the current iteration count, thereby updating the current chromosome. In the early stages of iteration, when the number of iterations is low, the improved method increases the mutation rate, allowing the chromosome to traverse all possibilities within the specified interval as much as possible, enhancing the algorithm's global search capability. As the number of iterations increases, when the number of iterations is close to the limit, the mutation rate is reduced using the above formula to prevent the algorithm from failing to converge accurately to the optimal parameters due to an excessively high mutation rate. During the iterative training of the network, the fitness of the current chromosome is compared with the initial optimal fitness value. If it is less than the initial optimal fitness value, the initial optimal fitness value is updated; otherwise, iterative updates continue until the optimal parameters are found, completing the model training.

[0150] Loading model parameters and using backpropagation (BP) to predict the test set includes the following steps:

[0151] S4401. Load the trained model parameters into the improved GA network; input the optimization output into the BP neural network to obtain the optimized BP neural network.

[0152] S4402. Load the test set into the improved BP network to predict the density of the hollow explosive charge; input the 244 samples in the test set into the optimized model to obtain the prediction results of the hollow explosive charge density.

[0153] S4403. Evaluate the accuracy of the network model according to the following formula.

[0154]

[0155] Where test represents the predicted result data, true represents the collected real data, and k represents the number of test sample groups, with a maximum of n.

[0156] The model's prediction performance was evaluated. Finally, the improved IGA-BP was compared with the unoptimized GA-BP. The mean squared error of prediction for IGA-BP was 9.3 × 10⁻³, while that for GA-BP was 1.02 × 10⁻². This shows that the optimized model performed better than GA-BP. The comparison results are as follows: Figure 5 , Figure 6 As shown.

[0157] The dataset consists of 700 sets of sample data, including six parameters: friction, aspect ratio, relative density, pressing speed, temperature, and holding time, and one output feature: finished product density.

[0158] Training and prediction results:

[0159] Analysis of the prediction results shows that the optimal training result is obtained when the chromosome set number is set to 20 and the number of iterations is set to 50. When the number of iterations is 30, the fitness value tends to stabilize, and the optimal fitness value is 0.001283. The IGA-BP model, which uses dynamic adaptive adjustment of crossover and mutation rates, converges more rapidly. The mean square error of the final predicted data (9.3 × 10⁻³) is also smaller than the root mean square error (RMSE) of the GA-BP model (1.02 × 10⁻²), a decrease of 8.8%, thus improving the accuracy of drug column compression density prediction.

[0160] In summary, the hollow explosive charge density prediction method provided in this application improves the crossover and mutation rates in the GA-BP algorithm by embedding the improved crossover and mutation rates into a genetic algorithm (GA), thereby obtaining an improved IGA-BP model. The improved IGA algorithm enhances the training convergence speed of the GA optimization algorithm. It also improves the accuracy of the weights in each layer of the GA-optimized BP neural network, reducing prediction errors and establishing an accurate prediction model for the density and uniformity of hollow explosive charges. The IGA-BP model not only has higher prediction accuracy but also significantly improved model stability, making it practically significant for explosive charge density prediction and providing methods and ideas for subsequent quality inspection and process parameter adjustment.

[0161] See Figure 7 This application embodiment can also provide a hollow explosive charge density prediction device, such as... Figure 7 As shown, the device may include:

[0162] The input parameter set acquisition unit 701 is used to acquire an input parameter set, which includes several hollow explosive charge compression process parameters.

[0163] The hollow explosive charge density prediction result output unit 402 is used to input the input parameter set into the trained target prediction model so that the target prediction model outputs the hollow explosive charge density prediction result; the target prediction model includes a neural network model and an improved genetic algorithm, the improved genetic algorithm is used to find the optimal parameters of the neural network model, and the improved genetic algorithm includes an improved crossover rate and an improved mutation rate;

[0164] The improved crossover rate and the improved mutation rate are obtained in the following ways:

[0165] By combining the average fitness value of the previous generation of chromosomes with the nonlinear action function of neurons, the improved crossover rate and the improved mutation rate, both of which can be adaptively adjusted, are obtained.

[0166] This application embodiment can also provide a hollow explosive charge density prediction device, the device including a processor and a memory:

[0167] The memory is used to store program code and transmit the program code to the processor;

[0168] The processor is used to execute the steps of the hollow explosive charge density prediction method described above according to the instructions in the program code.

[0169] like Figure 8 As shown in the figure, the hollow explosive charge density prediction device provided in this application embodiment may include: a processor 10, a memory 11, a communication interface 12, and a communication bus 13. The processor 10, memory 11, and communication interface 12 all communicate with each other through the communication bus 13.

[0170] In this embodiment, the processor 10 may be a central processing unit (CPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic devices.

[0171] The processor 10 can call the program stored in the memory 11. Specifically, the processor 10 can execute the operations in the embodiment of the hollow explosive charge density prediction method.

[0172] The memory 11 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment, the memory 11 stores at least a program for implementing the following functions:

[0173] Obtain an input parameter set, which includes several hollow explosive charge compression process parameters;

[0174] The input parameter set is input into the trained target prediction model so that the target prediction model outputs the hollow explosive charge density prediction result; the target prediction model includes a neural network model and an improved genetic algorithm, the improved genetic algorithm is used to find the optimal parameters of the neural network model, and the improved genetic algorithm includes an improved crossover rate and an improved mutation rate;

[0175] The improved crossover rate and the improved mutation rate are obtained in the following ways:

[0176] By combining the average fitness value of the previous generation of chromosomes with the nonlinear action function of neurons, the improved crossover rate and the improved mutation rate, both of which can be adaptively adjusted, are obtained.

[0177] In one possible implementation, the memory 11 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function (such as file creation or data read / write). The data storage area may store data created during use, such as initialization data.

[0178] In addition, memory 11 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0179] Communication interface 12 can be an interface for the communication module, used to connect with other devices or systems.

[0180] Of course, it should be noted that, Figure 8 The structure shown does not constitute a limitation on the hollow explosive charge density prediction device in the embodiments of this application. In practical applications, the hollow explosive charge density prediction device may include a ratio Figure 8 More or fewer components as shown, or combinations of certain components.

[0181] This application embodiment may also provide a computer-readable storage medium for storing program code for executing the steps of the above-described hollow explosive charge density prediction method.

[0182] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0183] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0184] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0185] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method of predicting the density of a hollow explosive charge column, characterized by, The method comprises the following steps: obtaining an input parameter set, the input parameter set comprising a plurality of hollow explosive column pressing process parameters; inputting the input parameter set into a trained target prediction model, so that the target prediction model outputs a hollow explosive column density prediction result; the target prediction model comprises a neural network model and an improved genetic algorithm, the improved genetic algorithm being used to find optimal parameters of the neural network model, the improved genetic algorithm comprising an improved crossover rate and an improved mutation rate; wherein the improved crossover rate and the improved mutation rate are obtained by the following method: combining the average fitness value of the last generation chromosome set with a nonlinear function of neurons to obtain the improved crossover rate and the improved mutation rate which can be self-adaptively adjusted; the nonlinear function of neurons is represented by the following formula: the average fitness value of the last generation chromosome set is represented by the following formula: wherein: is the best fitness value in the previous generation of the th population, is the total number of iterations, is the difference between the average fitness of the chromosome set, is the population size; the improved crossover rate is represented by the following formula: wherein: when is 0.1, is the largest fitness value in the previous generation of the th chromosome; and th chromosome. the improved mutation rate is represented by the following formula: In the formula: when hour, It is 0.

1.

2. The method of hollow explosive charge column density prediction of claim 1, wherein, the training method of the target prediction model comprises the following steps: inputting a training set, the training set comprising a plurality of samples; determining the structure of the neural network model; initializing the improved genetic algorithm, selecting the number of chromosome sets and the number of iterations, and randomly generating each gene value in each chromosome; calculating the current fitness of each chromosome; updating the improved crossover rate and the improved mutation rate; updating the chromosome set; calculating and updating the optimal fitness value; cycling to find the optimal chromosome as the network output result and saving the model parameters.

3. The method of hollow explosive charge column density prediction of claim 2, wherein, the fitness is calculated by the following formula: wherein fitness of the jth chromosome, fitness of the jth chromosome, fitness of the jth chromosome, predicted value of the jth chromosome in the ith sample, predicted value of the jth chromosome in the ith sample, actual value of the jth chromosome in the ith sample, actual value of the jth chromosome in the ith sample, actual value of the jth chromosome in the ith sample, sample size.

4. The method of hollow explosive charge column density prediction of claim 3, wherein, the chromosome is updated by the following formula: wherein: is a crossover rate; , is a mutation rate, is a number of contemporary iterations, is a maximum number of iterations, is a random number and .

5. The method of hollow explosive charge column density prediction of claim 3, wherein, normalizing a plurality of samples by the following formula: wherein: is a sample parameter, and are the minimum and maximum values of the experimental data, respectively, is 1, is -1.

6. The method of hollow explosive charge column density prediction of claim 1, wherein, the hollow explosive column pressing process parameters comprise material preheating temperature, mold preheating temperature, pressing pressure, vacuum time, holding time, vacuum degree and finished product density.

7. A hollow explosive charge grain density prediction device, characterized by, The device is used for performing the hollow explosive column density prediction method of any one of claims 1-6, and the device comprises: an input parameter set acquisition unit configured to acquire an input parameter set, the input parameter set comprising a plurality of hollow explosive column pressing process parameters; a hollow explosive column density prediction result output unit configured to input the input parameter set into a trained target prediction model, so that the target prediction model outputs a hollow explosive column density prediction result; the target prediction model comprises a neural network model and an improved genetic algorithm, the improved genetic algorithm being used to find optimal parameters of the neural network model, the improved genetic algorithm comprising an improved crossover rate and an improved mutation rate; wherein the improved crossover rate and the improved mutation rate are obtained by the following method: combining the average fitness value of the last generation chromosome set with a nonlinear function of neurons to obtain the improved crossover rate and the improved mutation rate which can be self-adaptively adjusted.

Images