Detonation product state equation parameter calculation model based on machine learning

Abstract

The invention provides a detonation product state equation parameter calculation model based on machine learning. The model is constructed by adopting the following method: step 1, cleaning sample data; step 2, calculation model construction: based on machine learning, constructing a high-energy explosive detonation product state equation parameter estimation calculation model by adopting different types of regression model algorithms; the high-energy explosive detonation product state equation parameter estimation calculation model is a calculation model. In the step 2, the regression model algorithm is an artificial neural network model, a support vector machine model or a nearest neighbor model. And step 3, calculating model training. And step 4, calculating model performance evaluation. The calculation model construction method is based on a machine learning technology method. When the calculation model is constructed, use of various regression models is considered, state equation parameters of detonation products of different types of high-energy explosives can be accurately estimated and calculated, and the method has wide adaptability.

Description

Technical Field

[0001] The invention belongs to the field of mixed explosives, relates to state equation parameters of detonation products of high-energy explosives, and specifically relates to a detonation product state equation parameter calculation model based on machine learning. Background Art

[0002] Explosive work capacity is an important aspect of characterizing the detonation performance of explosives. The parameters of the state equation of the detonation products are an important data representation of the explosive work capacity. At present, the explosive cylinder test is mostly used to obtain the cylinder wall expansion displacement curve or velocity curve, and then the state equation parameters of the high-energy explosive detonation products are obtained by fitting.

[0003] However, conducting explosive cylinder tests often requires a lot of financial and time costs. Moreover, due to the limitations of test conditions and test technology, there may be shortcomings in obtaining incomplete test data. In addition, each test can only be conducted on one test sample, which has an unfavorable limiting factor on the large-scale promotion of the test method.

[0004] From a theoretical perspective, the work done by the detonation products of high-energy explosives can be approximated as an adiabatic isentropic expansion process. The detonation theory of explosives, shock wave theory and thermodynamics theory can be combined to make assumptions, infer and calculate the parameters of the state equation of the detonation products from a mathematical physics perspective, which is of great significance for explaining the work done by the detonation of explosives.

[0005] The Chinese invention patent application number CN 202310270149.0 discloses a method for calculating the detonation performance of molten-cast explosives based on machine learning, which is suitable for calculating the detonation performance parameters of molten-cast explosives under the consideration of the melt-casting process parameters. However, the invention does not involve the calculation model and method of the state equation parameters of the detonation products.

[0006] At the same time, based on the machine learning method, a detonation product state equation parameter calculation model is constructed, and the detonation product state equation parameters are estimated and calculated, which is helpful to promote the overall improvement of the design level of energy characteristics of explosive formulations. Therefore, it is necessary to invent a method for constructing a detonation product state equation parameter calculation model based on machine learning, which is used for the estimation and calculation of the detonation product JWL state equation parameters. Summary of the invention

[0007] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a detonation product state equation parameter calculation model based on machine learning to solve the technical problem that the accuracy of the calculation model in the prior art needs to be further improved.

[0008] In order to solve the above technical problems, the present invention adopts the following technical solutions to achieve the above problems:

[0009] A detonation product state equation parameter calculation model based on machine learning is constructed using the following method:

[0010] Step 1: Sample data cleaning:

[0011] The state equation parameters of high-energy explosive detonation products are collected and sorted to form an original sample data set; the original sample data set is cleaned and preprocessed.

[0012] Step 2, computational model construction:

[0013] Based on machine learning, different types of regression model algorithms are used to construct a high-energy explosive detonation product state equation parameter estimation calculation model; the high-energy explosive detonation product state equation parameter estimation calculation model is a calculation model.

[0014] In step 2, the regression model algorithm is an artificial neural network model, a support vector machine model or a nearest neighbor model.

[0015] Step 3, computational model training:

[0016] Based on the optimization algorithm, different types of model training methods are used respectively, and the training data set and the test data set formed by the segmentation of the model data set in step 1 are combined to train and optimize the parameters of the state equation estimation calculation model for the detonation products of high-energy explosives constructed in step 2.

[0017] The model training method includes a gradient descent algorithm or a quasi-Newton algorithm.

[0018] The present invention also has the following technical features:

[0019] Specifically, the step 1 comprises the following steps:

[0020] Step 11: Processing of missing values ​​in the original sample data set.

[0021] Step 12: preprocessing of the original data of the original sample data set.

[0022] Step 13: Segmentation of the model data set.

[0023] Preferably, the specific process of step 11 is: converting and supplementing invalid data and blank data contained in the original sample data set; converting inappropriate non-numeric data contained in the original sample data set into reasonable numerical data, supplementing blank values ​​and zero values ​​contained in the original sample data set with the average value of related data, supplementing missing values ​​contained in the original sample data set with the average value of related data, eliminating duplicate data contained in the original sample data set, and deleting erroneous data out of range contained in the original sample data set; after processing the missing values, a complete original sample data set is formed.

[0024] Preferably, the specific process of step 12 is: normalizing the original data contained in the complete original sample data set obtained in step 11; dividing the original data in the original sample data set into input data and output data, wherein the input data includes the element composition of the explosive formula and the explosive charge density, and the output data includes the state equation parameters of the detonation product; normalizing the input data and the output data respectively to form a preprocessed model data set.

[0025] Further preferably, in step 12, the normalization processing includes extreme value normalization, mean normalization or nonlinear normalization.

[0026] Preferably, the specific process of step 13 is: segmenting the model data set preprocessed by step 12; segmenting the model data set into a training data set and a test data set by random sampling.

[0027] Preferably, in step 2, the artificial neural network model includes an input layer, a hidden layer and an output layer; the structure of the hidden layer includes 3 or 4 network layers.

[0028] Preferably, in step 2, the support vector machine model includes support vectors, hyperplanes and kernel functions; the kernel function is a polynomial kernel function, a radial basis kernel function or a Laplace kernel function.

[0029] Preferably, in step 2, the nearest neighbor model includes a distance metric and a nearest neighbor number; the value range of the nearest neighbor number is any positive integer in [3,30].

[0030] Furthermore, step 4 is also included.

[0031] Specifically, step 4, calculation model performance evaluation:

[0032] The performance of the constructed state equation parameter estimation calculation model for high-energy explosive detonation products was evaluated. The evaluation indicators included accuracy, mean square error and determination coefficient.

[0033] Compared with the prior art, the present invention has the following technical effects:

[0034] The computational model construction method involved in the present invention is based on a machine learning technology method. When constructing the computational model, the use of various regression models is considered, and the parameters of the state equation of detonation products of different types of high-energy explosives can be estimated and calculated more accurately, and it has wide adaptability.

[0035] The specific contents of the present invention are further explained in detail below in conjunction with embodiments. DETAILED DESCRIPTION

[0036] It should be noted that all raw materials, algorithms, methods, techniques and theories in the present invention, unless otherwise specified, are raw materials, algorithms, methods, techniques and theories known in the prior art.

[0037] It should be noted that, in the present invention, the International System of Units (SI) can be used as the main unit system, and other commonly used derived unit systems are recommended for some parameters. Among them, the element composition of the explosive formula is expressed as the molar number of various constituent elements; the unit of the explosive charge density is g / cm 3 The state equation of high explosive detonation products refers to the JWL state equation of detonation products. The specific parameters and units are: parameter A, parameter B and parameter C, all in GPa; parameter R1 and parameter R2, dimensionless quantities; parameter w is a dimensionless quantity; parameter E0, in GPa; the conversion relationship is 1GPa = 10 11 Pa. All model parameters in the state equation calculation model of high-energy explosive detonation products are dimensionless quantities.

[0038] In view of the shortcomings of the prior art, the present invention discloses a method for constructing a detonation product state equation parameter calculation model based on machine learning. Combining the basic theory and technical methods of machine learning, a high-energy explosive detonation product JWL state equation parameter calculation model is constructed, which is used for the estimated calculation of the JWL state equation parameters of detonation products of different types of high-energy explosives.

[0039] The basic design concept of the present invention includes the following aspects:

[0040] First, the cleaning of sample data based on data statistical analysis. The ultimate goal of sample data cleaning is to form a model data set for computational model construction and computational model training. First, it is necessary to collect and organize the parameters of the JWL (Jones-Wilkins-Lee) state equation of various types of high-energy explosive detonation products, and collect and classify various types of literature data based on artificial intelligence methods. The sources of literature data include monographs, scientific papers, professional books and periodicals, industry conferences, scientific reports, patents, network data and technical documents. According to the data structure and vector characteristics of the JWL state equation of high-energy explosive detonation products, various literature data are clustered and subject modeled, and the data structure style of the original sample data set is constructed, and the literature data is extracted and gradually filled into the original sample data set. The design concept mainly includes step one in the present invention, namely, sample data cleaning.

[0041] Second, the construction of a computational model based on a machine learning method. The ultimate goal of constructing a computational model is to form a machine learning model and establish a mapping relationship model from input data to output data, so as to be applied to the estimated calculation of the parameters of the JWL state equation of high-energy explosive detonation products. The input data includes the composition of explosive formula elements and the density of explosive charge, and the output data includes the parameters of the JWL state equation of high-energy explosive detonation products. Both types of data show the characteristics of continuous change and uniform distribution in their respective effective ranges. From the perspective of implementation effect, the problem of establishing a mapping relationship from one set of continuous data to another set of continuous data usually belongs to a regression problem, and the input data and output data of the present invention both contain multiple variables, so it also belongs to a multi-input and multi-output regression problem. The computational model construction method involved in the present invention is based on three commonly used regression model algorithms in machine learning methods, including artificial neural network models, support vector machine models and nearest neighbor models, and any one of the models is selected. The design concept mainly includes step two in the present invention, that is, computational model construction.

[0042] Third, the evaluation of model performance based on data statistical indicators. It mainly includes two aspects: computational model training and computational model performance evaluation. The ultimate goal of computational model training is to use the training data set and the optimization design algorithm to optimize the training and dynamic adjustment of the model parameters contained in the constructed computational model, so as to form an executable computational model and realize the estimated calculation of the JWL state equation parameters of the detonation products of high-energy explosives. The ultimate goal of computational model performance evaluation is to use the test data set and the statistical evaluation indicators to analyze and evaluate the performance of various computational models, form a complete calculation result, and provide a reference for users. This design concept mainly includes steps three and four in the present invention, namely, computational model training and computational model performance evaluation.

[0043] Specific embodiments of the present invention are given below. It should be noted that the present invention is not limited to the following specific embodiments, and all equivalent changes made on the basis of the technical solution of this application fall within the protection scope of the present invention.

[0044] Example:

[0045] This embodiment provides a detonation product state equation parameter calculation model based on machine learning, and the model is constructed using the following method:

[0046] Step 1: Sample data cleaning:

[0047] Collect and organize the state equation parameters of high-energy explosive detonation products to form an original sample data set; perform data cleaning and preprocessing on the original sample data set for subsequent computational model training and computational model performance evaluation.

[0048] Specifically, step 1 includes the following steps:

[0049] Step 11: Processing of missing values ​​in the original sample data set:

[0050] Convert and supplement the invalid data and blank data contained in the original sample data set; convert the inappropriate non-numeric data contained in the original sample data set into reasonable numeric data, supplement the blank values ​​and zero values ​​contained in the original sample data set with the average value of the relevant data, supplement the missing values ​​contained in the original sample data set with the average value of the relevant data, eliminate the duplicate data contained in the original sample data set, and delete the erroneous data out of range contained in the original sample data set; after processing the missing values, form a complete original sample data set.

[0051] In this embodiment, the main purpose of step 11 is to perform preliminary processing on the non-standard data format and incomplete data types in the original sample data set to form a complete original sample data set.

[0052] The abnormal situations considered when processing missing values ​​in the original sample data set mainly include the following types:

[0053] First, the type conversion of invalid data, such as converting non-numeric data such as text or character data to zero value or corresponding floating-point data.

[0054] Second, add and supplement blank data, such as filling blank data with Null data.

[0055] Third, average filling of blank values ​​and zero-value data, such as filling Null data and zero-value data with the average value of row and column data.

[0056] Fourth, duplicate data is found and removed, such as trimming the excess parts of the duplicate data to retain only a set of non-duplicate independent data.

[0057] Fifth, the judgment and deletion of erroneous data, such as deleting abnormal data that exceeds the range and obviously erroneous data. Through a series of processing of missing values, a standardized and complete original sample data set is formed.

[0058] Step 12: Preprocessing of the original sample data set:

[0059] The original data contained in the complete original sample data set obtained in step 11 is normalized; the original data in the original sample data set is divided into input data and output data, wherein the input data includes the composition of explosive formula elements and the density of explosive charge, and the output data includes the parameters of the detonation product state equation; the input data and the output data are normalized respectively to form a preprocessed model data set.

[0060] In this embodiment, the main goal of step 12 is to normalize the raw data in the original sample data set, transform the data features, reduce the impact of data sensitivity, form a model data set, and use it for subsequent computational model construction and computational model training, and improve the accuracy of the estimated computational results. The raw data contained in the original sample data set can be divided into input data and output data according to the difference between the data category and the vector feature. Among them, the input data includes the explosive formula element composition and the explosive charge density, and the output data includes the JWL state equation parameters of the high-energy explosive detonation product. The types of chemical elements involved in the explosive formula element composition in this embodiment include C, H, N, O, Al, Cl and F. When converting the explosive formula element composition, the conversion is mainly based on the chemical formula, relative molar mass and mass percentage of various raw materials in the explosive formula. When calculating, it is assumed that the total mass of various raw materials is 100. The formula used in the conversion is as follows:

[0061]

[0062]

[0063] Where:

[0064] element represents the elemental composition of the explosive formula. The chemical elements include C, H, N, O, Al, Cl and F;

[0065] n represents the number of moles of the substance of each explosive raw material;

[0066] i represents the number of types of explosive raw materials;

[0067] j represents the number of chemical elements;

[0068] Fraction represents the mass percentage of each explosive raw material;

[0069] molwt represents the relative atomic mass of each chemical element.

[0070] The conversion of explosive charge density is mainly based on the theoretical density and mass percentage of various raw materials in the explosive formula and the density ratio. The total mass of various raw materials is assumed to be 100 parts. The formula used for conversion is as follows:

[0071]

[0072] Where:

[0073] density represents the density of explosive charge;

[0074] i represents the number of types of explosive raw materials;

[0075] Fraction indicates the mass percentage of each raw material;

[0076] TMD represents the theoretical density of each raw material;

[0077] Ratio indicates the density ratio.

[0078] The JWL state equation of high explosive detonation products can be expressed as follows:

[0079]

[0080] Or it can be expressed as:

[0081]

[0082]

[0083] The connotations represented by the two forms are consistent.

[0084] Where:

[0085] Parameter A, parameter B and parameter C are all in GPa;

[0086] Parameters R1 and R2 are dimensionless quantities;

[0087] The parameter w is a dimensionless quantity;

[0088] Parameter E0, unit is GPa;

[0089] v represents relative specific volume.

[0090] As a preferred solution of this embodiment, in step 12, the normalization processing includes extreme value normalization, mean normalization or nonlinear normalization.

[0091] In this embodiment, specifically, when normalizing the original data in the original sample data set, the normalization method used mainly includes the following methods:

[0092] First, the extreme value normalization processing method can be expressed as the following formula:

[0093]

[0094] Second, the mean normalization method can be expressed as the following formula:

[0095]

[0096] In formula (6) to formula (7):

[0097] x represents the original sample data;

[0098] x norm Represents the normalized data;

[0099] x min Represents the minimum value in the original sample data;

[0100] x max Represents the maximum value in the original sample data;

[0101] x μ Represents the mean in the original sample data.

[0102] x σ Represents the standard deviation in the original sample data.

[0103] Third, nonlinear normalization methods, including logarithmic transformation, square root transformation and exponential transformation.

[0104] After normalizing the input data and output data respectively, a preprocessed model data set is formed.

[0105] Step 13, segmentation of model data set:

[0106] The model data set preprocessed in step 12 is segmented; the model data set is segmented into a training data set and a test data set by random sampling, which are used for subsequent computational model training and computational model performance evaluation, respectively.

[0107] In this embodiment, the main goal of step 13 is to segment the model data set to form a training data set and a test data set, respectively, for subsequent computational model training and computational model performance evaluation. When the model data set is segmented, the data contained in the model data set is respectively included in the training data set and the test data set based on the methods of stratified sampling and random sampling. The data contained in the training data set and the test data set are mutually exclusive, and the data contained in the training data set and the test data set just constitute a complete model data set when added together. In general, the ratio of the number of data contained in the training data set to the number of data contained in the test data set is between 8:2 and 7:3.

[0108] Step 2, computational model construction:

[0109] Based on machine learning, different types of regression model algorithms are used to construct a parameter estimation calculation model for the state equation of detonation products of high-energy explosives; the parameter estimation calculation model for the state equation of detonation products of high-energy explosives is the calculation model.

[0110] As a preferred solution of this embodiment, the regression model algorithm is an artificial neural network model, a support vector machine model or a nearest neighbor model.

[0111] Step 2 includes the following steps:

[0112] Step 21, constructing a computational model based on the artificial neural network model:

[0113] The artificial neural network model includes an input layer, a hidden layer and an output layer; the structure of the hidden layer contains 3 or 4 network layers.

[0114] In this embodiment, the structure of the hidden layer is designed when the computational model is constructed; the neurons included in the hidden layer are connected in a fully connected manner, and the neurons use the same activation function.

[0115] In this embodiment, the main goal of step 21 is to complete the construction of a computing model based on an artificial neural network. An artificial neural network model generally includes an input layer, a hidden layer, and an output layer, and the basic constituent unit is a neuron. When constructing a computing model based on an artificial neural network, the hidden layer is mainly designed. The number of neurons contained in the input layer is equal to the number of variables in the input data, and the number of neurons contained in the output layer is equal to the number of variables in the output data. In this embodiment. The hidden layer includes 3 or 4 network layers, and each network layer includes a certain number of neurons. Generally, the number of neurons contained in each network layer is not less than 20. The neurons contained in the hidden layer are connected in a fully connected manner, and the same activation function is used to transmit signals. Common activation functions include tanh function, sigmoid function and relu function.

[0116] First, the common expression of the tanh function is as follows:

[0117]

[0118] Second, the common expression of the sigmoid function is as follows:

[0119]

[0120] Third, the common expression of the relu function is as follows:

[0121] relu(x)=max(0,x) (10)

[0122] In formula (8) to formula (10):

[0123] x represents the input data.

[0124] In this embodiment, the nonlinear activation function is used in the calculation model based on the artificial neural network, which can retain the nonlinear characteristics in the regression calculation process to the greatest extent while maintaining the strength and depth of signal propagation.

[0125] Step 22: construct a calculation model based on the support vector machine model:

[0126] The support vector machine model includes support vectors, hyperplanes and kernel functions; the kernel function is a polynomial kernel function, a radial basis kernel function or a Laplace kernel function.

[0127] In this embodiment, the kernel function is designed when the calculation model is constructed, and the error term penalty coefficient is set to a default value.

[0128] In this embodiment, the main goal of step 22 is to complete the construction of a computing model based on a support vector machine. A support vector machine model generally includes support vectors, hyperplanes, and kernel function designs. The basic idea of ​​a support vector machine model is to maximize the interval value in the feature space, and the interval value is defined as the distance between the segmentation hyperplane or the decision boundary and the closest sample point, and the closest sample point is called a support vector. When constructing a computing model based on a support vector machine, the kernel function is mainly designed. Commonly used kernel functions include polynomial kernel functions, radial basis kernel functions, and Laplace kernel functions.

[0129] First, the common expression of the polynomial kernel function is as follows:

[0130] K(x,x′)=(γx·x′+c) d (11)

[0131] Second, the common expression of the radial basis kernel function is as follows:

[0132] K(x,x′)=exp(-γx-x′ 2 ) (12)

[0133] Third, the common expression of the Laplace kernel function is as follows:

[0134]

[0135] In formula (11) to formula (13):

[0136] K(x,x') represents the kernel function;

[0137] x represents input data;

[0138] x' represents the observed data;

[0139] γ represents the model hyperparameter;

[0140] c represents the model hyperparameters;

[0141] d represents the model hyperparameters;

[0142] represents the gradient operator;

[0143] |||| represents the norm.

[0144] The kernel functions used in the computing model based on the support vector machine in the invention all exhibit nonlinear characteristics, which can fully reflect the nonlinear characteristics in the regression calculation process. At the same time, the error term penalty coefficient is set to a default value, so that the computing model can retain a more suitable fitting ability and avoid the occurrence of overfitting.

[0145] Step 23, construct a calculation model based on the nearest neighbor model:

[0146] The nearest neighbor model includes a distance metric and the number of nearest neighbors; the value range of the nearest neighbor number is any positive integer in [3,30].

[0147] In this embodiment, the calculation model is constructed with respect to the nearest neighbor number, and the selection of the nearest neighbor number is optimized in an iterative manner; the distance measurement is measured in one of the following ways: Euclidean distance, Chebyshev distance or Minkowski distance.

[0148] In this embodiment, the main goal of step 23 is to complete the construction of the nearest neighbor-based computing model. The nearest neighbor model is also called the K-nearest neighbor model, and its basic idea is that the category and data of the current sample can be represented by the category and data of its K nearest neighbor samples. The nearest neighbor model includes the design of the distance metric and the nearest neighbor number K. The distance metric includes the Euclidean distance, the Chebyshev distance, or the Minkowski distance.

[0149] First, the common expression for Euclidean distance is as follows:

[0150]

[0151] Second, a common expression for Chebyshev distance is as follows:

[0152]

[0153] Third, the common expression for Minkowski distance is as follows:

[0154]

[0155] In formula (14) to formula (16):

[0156] D(x,y) represents the distance metric;

[0157] x represents sample data;

[0158] y represents the predicted data;

[0159] i represents the number of sample data, i = 1, 2, ..., n;

[0160] p represents the power of the exponent.

[0161] In this embodiment, the calculation model based on the nearest neighbor is constructed mainly for the number of nearest neighbors, that is, the K value is iteratively optimized. Under the premise of meeting the estimated calculation requirements and the calculation model performance evaluation index not being reduced, under normal circumstances, the value range of the K value is [3,30], and any positive integer is selected. The distance metrics used in the calculation model based on the nearest neighbor in this embodiment all show nonlinear characteristics, which can effectively reveal the nonlinear characteristics in the regression calculation process.

[0162] Step 3, computational model training:

[0163] Based on the optimization algorithm, different types of model training methods are used respectively, and the training data set and the test data set formed by the segmentation of the model data set in step 1 are combined to train and optimize the parameters of the state equation estimation calculation model for the detonation products of high-energy explosives constructed in step 2.

[0164] As a preferred solution of this embodiment, in step 3, the model training method includes a gradient descent algorithm or a quasi-Newton algorithm.

[0165] In this embodiment, the main goal of step 3 is to optimize the model parameters contained in the constructed computing model based on the optimization design algorithm. Different types of model training methods are used to train the computing model parameters using the training data set.

[0166] Step 4: Calculate model performance evaluation:

[0167] The performance of the constructed state equation parameter estimation calculation model for high-energy explosive detonation products was evaluated. The evaluation indicators included accuracy, mean square error and determination coefficient.

[0168] In this embodiment, the main goal of step 4 is to evaluate the performance of the calculation model using statistical evaluation indicators.

[0169] First, the common expression of accuracy is as follows:

[0170]

[0171] Second, a common expression for mean square error is as follows:

[0172]

[0173] Third, a common expression for the coefficient of determination is as follows:

[0174]

[0175] In formulas (17) to (19):

[0176] accuracy indicates accuracy;

[0177] MSE stands for mean square error.

[0178] R 2 represents the coefficient of determination;

[0179] y pred represents the predicted value;

[0180] y represents the sample value;

[0181] i represents the number of sample data, i = 1, 2, ..., n;

[0182] x represents input data;

[0183] f(x) represents the activation function.

[0184] Application Example 1:

[0185] This application example provides a detonation product state equation parameter calculation model based on machine learning based on the above embodiment. In this application example, the high-energy explosive is numbered X1, the formula weight composition is Fox-7 / TNT / Al / paraffin=40 / 35 / 17 / 8, and the explosive charge density is 1.68. The model is constructed using the following method:

[0186] Step 1: Sample data cleaning:

[0187] Step 11: Processing missing values ​​in the original sample data set.

[0188] Step 12: Preprocessing of the original sample data set: Normalization processing: Selecting the mean normalization processing method.

[0189] Step 13: Segmentation of the model data set: The ratio of the number of data contained in the training data set formed by segmentation to the number of data contained in the test data set is 8:2.

[0190] Step 2, computational model construction:

[0191] This application example builds a computational model based on an artificial neural network.

[0192] In this application example, the hidden layer contains 4 network layers, and the number of neurons in each network layer is 20, 30, 40 and 30 respectively. The activation function used is the relu function.

[0193] Step 3, computational model training:

[0194] Based on the optimization algorithm, this application example uses the gradient descent algorithm to train and optimize the parameters of the constructed high-energy explosive detonation product state equation parameter estimation calculation model.

[0195] Step 4: Calculate model performance evaluation:

[0196] The estimated calculation results of the JWL state equation parameters of the detonation product of the high-energy explosive numbered X1 are: A = 750.5GPa, B = 10.05GPa and C = 0.72GPa, R1 = 5.15 and R2 = 1.42, w = 0.21, E = 7.3GPa. The accuracy is 0.90, the mean square error is 90.5 and the determination coefficient is 0.92.

[0197] The above model and parameters are used to estimate and calculate the parameters of the JWL state equation of high-energy explosive detonation products. The model has a wide range of application and strong adaptability.

[0198] Application Example 2:

[0199] This application example provides a detonation product state equation parameter calculation model based on machine learning based on the above embodiment. In this application example, the high-energy explosive is numbered X2, the formula weight composition is HMX / VitonA=85 / 15, and the explosive charge density is 1.85. The model is constructed using the following method:

[0200] Step 1: Sample data cleaning:

[0201] Step 11: Processing missing values ​​in the original sample data set.

[0202] Step 12: Preprocessing of the original sample data set: Normalization processing: Selecting the mean normalization processing method.

[0203] Step 13: Segmentation of the model data set: The ratio of the number of data contained in the training data set formed by segmentation to the number of data contained in the test data set is 7:3.

[0204] Step 2, computational model construction:

[0205] This application example builds a computing model based on support vector machine.

[0206] In this application example, the kernel function is the radial basis kernel function, and the error term penalty coefficient is set to 1.0.

[0207] Step 3, computational model training:

[0208] Based on the optimization algorithm, this application example uses the quasi-Newton method to train and optimize the parameters of the constructed high-energy explosive detonation product state equation parameter estimation calculation model.

[0209] Step 4: Calculate model performance evaluation:

[0210] The estimated calculation results of the JWL state equation parameters of the detonation product of the high-energy explosive numbered X2 are: A = 860.1GPa, B = 14.7GPa and C = 1.2GPa, R1 = 4.7 and R2 = 1.25, w = 0.33, E = 9.7GPa. The accuracy is 0.89, the mean square error is 150.6 and the determination coefficient is 0.87.

[0211] The above model and parameters are used to estimate and calculate the parameters of the JWL state equation of high-energy explosive detonation products. The model has a wide range of application and strong adaptability.

[0212] Application Example 3:

[0213] This application example provides a detonation product state equation parameter calculation model based on machine learning based on the above embodiment. In this application example, the high-energy explosive is numbered X3, and the formula composition is CL-20 / Al / paraffin

[0214] =85 / 10 / 5, and the explosive charge density is 1.89. The model is constructed using the following method:

[0215] Step 1: Sample data cleaning:

[0216] Step 11: Processing missing values ​​in the original sample data set.

[0217] Step 12: Preprocessing of the original sample data set: Normalization processing selects the extreme value normalization processing method.

[0218] Step 13, segmentation of the model data set. The ratio of the number of data contained in the training data set formed by segmentation to the number of data contained in the test data set is 7.5:2.5.

[0219] Step 2, computational model construction:

[0220] This application example builds a calculation model based on the nearest neighbor model.

[0221] In this application example, the number of nearest neighbors is K=12, and the distance metric is the Euclidean distance.

[0222] Step 3, computational model training:

[0223] Based on the optimization algorithm, this application example uses the quasi-Newton method to train and optimize the parameters of the constructed high-energy explosive detonation product state equation parameter estimation calculation model.

[0224] Step 4: Calculate model performance evaluation:

[0225] The estimated calculation results of the JWL state equation parameters of the detonation product of the high-energy explosive numbered X3 are: A = 1190GPa, B = 26.5GPa and C = 1.3GPa, R1 = 5.2 and R2 = 1.5, w = 0.39, E = 10.6GPa. The accuracy is 0.86, the mean square error is 330.9 and the determination coefficient is 0.87.

[0226] The above model and parameters are used to estimate and calculate the parameters of the JWL state equation of high-energy explosive detonation products. The model has a wide range of application and strong adaptability.

Claims

1. A detonation product state equation parameter calculation model based on machine learning, characterized in that: The model was constructed using the following method: Step 1: Sample data cleaning: Collect and organize the state equation parameters of high-energy explosive detonation products to form an original sample data set; perform data cleaning and preprocessing on the original sample data set; Step 2, computational model construction: Based on machine learning, different types of regression model algorithms are used to construct a high-energy explosive detonation product state equation parameter estimation calculation model; the high-energy explosive detonation product state equation parameter estimation calculation model is a calculation model; The regression model algorithm is an artificial neural network model, a support vector machine model or a nearest neighbor model; Step 3, computational model training: Based on the optimization algorithm, different types of model training methods are used respectively, and the training data set and the test data set formed by the segmentation of the model data set in step 1 are combined to train and optimize the parameters of the high-energy explosive detonation product state equation parameter estimation calculation model constructed in step 2; The model training method is a gradient descent algorithm or a quasi-Newton algorithm.

2. The detonation product state equation parameter calculation model based on machine learning according to claim 1, characterized in that: The step 1 comprises the following steps: Step 11, processing of missing values ​​in the original sample data set; Step 12, preprocessing of the original data of the original sample data set; Step 13: segmentation of the model data set.

3. The detonation product state equation parameter calculation model based on machine learning according to claim 2, characterized in that: The specific process of step 11 is: converting and supplementing the invalid data and blank data contained in the original sample data set; converting the inappropriate non-numeric data contained in the original sample data set into reasonable numeric data, supplementing the blank values ​​and zero values ​​contained in the original sample data set with the average value of the relevant data, supplementing the missing values ​​contained in the original sample data set with the average value of the relevant data, eliminating the duplicate data contained in the original sample data set, and deleting the erroneous data out of range contained in the original sample data set; After processing the missing values, a complete set of original sample data is formed.

4. The detonation product state equation parameter calculation model based on machine learning according to claim 3, characterized in that: The specific process of step 12 is: normalize the original data contained in the complete original sample data set obtained in step 11; divide the original data in the original sample data set into input data and output data, wherein the input data includes the composition of explosive formula elements and the density of explosive charge, and the output data includes the parameters of the detonation product state equation; normalize the input data and the output data respectively to form a preprocessed model data set.

5. The detonation product state equation parameter calculation model based on machine learning according to claim 4, characterized in that: In step 12, the normalization processing includes extreme value normalization, mean normalization or nonlinear normalization.

6. The detonation product state equation parameter calculation model based on machine learning according to claim 4, characterized in that: The specific process of step 13 is: segmenting the model data set preprocessed by step 12; segmenting the model data set into a training data set and a test data set by random sampling.

7. The detonation product state equation parameter calculation model based on machine learning according to claim 1, characterized in that: In step 2, the artificial neural network model includes an input layer, a hidden layer and an output layer; the structure of the hidden layer includes 3 or 4 network layers.

8. The detonation product state equation parameter calculation model based on machine learning according to claim 1, characterized in that: In step 2, the support vector machine model includes support vectors, hyperplanes and kernel functions; the kernel function is a polynomial kernel function, a radial basis kernel function or a Laplace kernel function.

9. The detonation product state equation parameter calculation model based on machine learning according to claim 1, characterized in that: In step 2, the nearest neighbor model includes a distance metric and a nearest neighbor number; the value range of the nearest neighbor number is any positive integer in [3,30].

10. The detonation product state equation parameter calculation model based on machine learning according to claim 1, characterized in that: Also includes step 4; Step 4: Calculate model performance evaluation: The performance of the constructed high-energy explosive detonation product state equation parameter estimation calculation model was evaluated. The evaluation indicators included accuracy, mean square error and determination coefficient.

Images