Quantum chemistry-based method and system for predicting electrostatic spark sensitivity of energetic materials

Abstract

The application discloses a kind of quantum chemistry-based energetic material electrostatic spark sensitivity prediction method and system, it is related to energetic material performance prediction field, the method includes: new energetic material sample carries out quantum chemistry calculation, obtains the multidimensional influence parameter data set corresponding to the new energetic material sample, and multidimensional influence parameter data set is input into energetic material electrostatic spark sensitivity value prediction model, obtains the electrostatic spark sensitivity value corresponding to new energetic material sample.The present application can solve the technical problems that the existing energetic material electrostatic spark sensitivity linear prediction model is narrow in scope of application, low in prediction efficiency, and it is difficult to realize high-precision prior prediction of electrostatic spark sensitivity of new unsynthesized energetic material.

Description

Technical Field

[0001] This invention relates to the field of energetic material performance prediction technology, and in particular to a method and system for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemical calculations and BP neural networks. Background Technology

[0002] With the continuous evolution of modern battlefield combat modes, the need for research and development upgrades of high-performance weaponry is becoming increasingly urgent. Designing and synthesizing novel energetic materials with high energy density and low sensitivity are crucial prerequisites for ensuring that weaponry meets the requirements of precision, safety, and efficient destruction capabilities in modern battlefields. The electrostatic spark sensitivity of energetic materials refers to their sensitivity to electrostatic discharge under a certain energy level. Due to the high insulation properties of energetic materials, they are prone to generating and accumulating high-energy static electricity during production, storage, and transportation. If an electrostatic spark is released, it may lead to combustion or even explosion. Therefore, in the laboratory synthesis stage of designing and developing novel high-energy, low-sensitivity energetic materials, it is necessary to achieve high-precision prediction of their electrostatic spark sensitivity through theoretical models in advance. This is of great significance for replacing traditional experimental trial-and-error methods.

[0003] Currently, linear regression models establishing a set of influencing factors composed of simple molecular geometric parameters and electrostatic spark sensitivity can be used for predictive research. However, because these models only involve the most basic molecular composition parameters of the material itself and lack consideration for complex quantum chemical parameters such as molecular orbital energy, electronic state, thermodynamic functions, dipole moment, and atomic charge, the established set of influencing parameters has a relatively small impact on electrostatic spark sensitivity. Meanwhile, the relationship between the molecular structure and electrostatic spark sensitivity of energetic materials is complex and nonlinear. Traditional empirical formula prediction models tend to overlook some potential relationships between independent and dependent variables. Therefore, current prediction models for electrostatic spark sensitivity values ​​of energetic materials suffer from narrow applicability and low prediction accuracy. Summary of the Invention

[0004] To address the technical problems of existing linear prediction models for the electrostatic spark sensitivity of energetic materials having narrow applicability and low prediction efficiency, making it difficult to achieve high-precision prior prediction of the electrostatic spark sensitivity of novel, unsynthesized energetic materials, this invention provides a quantum chemistry-based method and system for predicting the electrostatic spark sensitivity of energetic materials.

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

[0006] This invention provides a quantum chemistry-based method for predicting the electrostatic spark sensitivity of energetic materials, comprising:

[0007] Obtain samples of novel energetic materials;

[0008] Quantum chemical calculations were performed on the novel energetic material sample to obtain a multidimensional influence parameter dataset corresponding to the novel energetic material sample; the multidimensional influence parameter dataset includes orbital band gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, oxygen balance, and detonation parameters;

[0009] The multidimensional influence parameter dataset corresponding to the novel energetic material sample is input into the electrostatic spark sensitivity value prediction model of energetic material to obtain the electrostatic spark sensitivity value corresponding to the novel energetic material sample.

[0010] The training process for the electrostatic spark sensitivity prediction model for energetic materials is as follows:

[0011] Obtain a dataset of electrostatic spark sensitivity experimental values ​​for existing energetic material samples;

[0012] Quantum chemical calculations were performed on the existing energetic material samples to obtain a multidimensional influence parameter dataset corresponding to the existing energetic material samples;

[0013] Using the multidimensional influence parameter dataset corresponding to the existing energetic material samples as input parameters and the electrostatic spark sensitivity experimental value dataset of the existing energetic material samples as output parameters, a BP neural network is trained to obtain an energetic material electrostatic spark sensitivity value prediction model.

[0014] This invention also provides a quantum chemistry-based electrostatic spark sensitivity prediction system for energetic materials, comprising:

[0015] A novel energetic material sample acquisition module is used to acquire samples of novel energetic materials.

[0016] A multidimensional influence parameter dataset construction module is used to perform quantum chemical calculations on the novel energetic material sample to obtain the multidimensional influence parameter dataset corresponding to the novel energetic material sample; the multidimensional influence parameter dataset includes orbital band gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, oxygen balance and detonation parameters.

[0017] The electrostatic spark sensitivity prediction module is used to input the multidimensional influence parameter dataset corresponding to the novel energetic material sample into the energetic material electrostatic spark sensitivity prediction model to obtain the electrostatic spark sensitivity value corresponding to the novel energetic material sample.

[0018] The training process for the electrostatic spark sensitivity prediction model for energetic materials is as follows:

[0019] Obtain a dataset of electrostatic spark sensitivity experimental values ​​for existing energetic material samples;

[0020] Quantum chemical calculations were performed on the existing energetic material samples to obtain a multidimensional influence parameter dataset corresponding to the existing energetic material samples;

[0021] Using the multidimensional influence parameter dataset corresponding to the existing energetic material samples as input parameters and the electrostatic spark sensitivity experimental value dataset of the existing energetic material samples as output parameters, a BP neural network is trained to obtain an energetic material electrostatic spark sensitivity value prediction model.

[0022] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0023] To address the shortcomings of existing technologies, this invention provides a quantum chemistry-based method and system for predicting the electrostatic spark sensitivity of energetic materials. The optimal multidimensional influence parameter dataset is determined through quantum chemistry and molecular topology calculations. Then, a backpropagation (BP) neural network is used to establish a model for predicting the electrostatic spark sensitivity of energetic materials. This allows for high-precision prediction of the electrostatic spark sensitivity of novel energetic materials, facilitating efficient and accurate early assessment of whether the designed and synthesized novel energetic materials meet the target safety performance requirements. Attached Figure Description

[0024] 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.

[0025] Figure 1 This is a flowchart illustrating the electrostatic spark sensitivity prediction method for energetic materials based on quantum chemistry provided in Embodiment 1 of the present invention.

[0026] Figure 2 This is a flowchart illustrating the electrostatic spark sensitivity prediction method for energetic materials based on quantum chemistry provided in Embodiment 2 of the present invention.

[0027] Figure 3 This is a histogram of Pearson correlation coefficients of the influencing factors in Embodiment 2 of the present invention;

[0028] Figure 4 This is a graph showing the changing trend of the number of neurons in the hidden layer and the error in Embodiment 2 of the present invention;

[0029] Figure 5 This is a schematic diagram of the BP neural network structure in Embodiment 2 of the present invention;

[0030] Figure 6 This is a graph of the loss function of the BP neural network in Embodiment 2 of the present invention;

[0031] Figure 7 This is a scatter plot comparing the predicted and experimental values ​​output by the electrostatic spark sensitivity prediction model for energetic materials in Embodiments 2, 3, 4, and 5 of the present invention.

[0032] Figure 8 This is a residual diagram of the predicted values ​​output by the electrostatic spark sensitivity prediction model for energetic materials in Embodiments 2, 3, 4 and 5 of this invention.

[0033] Figure 9 This is an application domain diagram of the predicted values ​​output by the electrostatic spark sensitivity prediction model for energetic materials in Embodiments 2, 3, 4 and 5 of the present invention;

[0034] Figure 10 This is a schematic diagram of the structure of the quantum chemistry-based electrostatic spark sensitivity prediction system for energetic materials provided in an embodiment of the present invention. Detailed Implementation

[0035] 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.

[0036] 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.

[0037] Example 1

[0038] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemistry, which includes the following steps.

[0039] Step 100: Obtain samples of novel energetic materials.

[0040] Step 200: Perform quantum chemical calculations on the novel energetic material sample to obtain a multidimensional influence parameter dataset corresponding to the novel energetic material sample; the multidimensional influence parameter dataset includes orbital band gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, oxygen balance, and detonation parameters.

[0041] Step 300: Input the multidimensional influence parameter dataset corresponding to the novel energetic material sample into the electrostatic spark sensitivity prediction model of the energetic material to obtain the electrostatic spark sensitivity value corresponding to the novel energetic material sample.

[0042] In this embodiment of the invention, step 300 specifically includes:

[0043] Step 301: Perform Pearson correlation analysis and clean up similar redundant variables in the multidimensional influence parameter dataset corresponding to the novel energetic material sample.

[0044] Step 302: The cleaned data is normalized using the min-max method to obtain the normalized multidimensional influence parameter dataset corresponding to the new energetic material samples.

[0045] Step 303: Input the normalized multidimensional influence parameter dataset corresponding to the novel energetic material sample into the electrostatic spark sensitivity prediction model of the energetic material to obtain the electrostatic spark sensitivity value corresponding to the novel energetic material sample.

[0046] The training process for the electrostatic spark sensitivity prediction model for energetic materials is as follows:

[0047] (1) Obtain the dataset of electrostatic spark sensitivity experimental values ​​of existing energetic material samples. Specifically, collect and organize the dataset of electrostatic spark sensitivity experimental values ​​of existing energetic material samples. The selected electrostatic spark sensitivity experimental values ​​are mainly from experimental electrostatic spark sensitivity values ​​given in publicly published journal articles.

[0048] (2) Perform quantum chemical calculations on the existing energetic material samples to obtain the multidimensional influence parameter dataset corresponding to the existing energetic material samples.

[0049] The direct influencing factor on the electrostatic spark sensitivity of energetic materials is their molecular structure. While considering the molecular structure, complex quantum chemical parameters such as orbital energy, electronic state, thermodynamic functions, dipole moment, and atomic charge must also be taken into account. Therefore, the number of atoms, molecular weight, oxygen balance, orbital gap, atomic charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, and detonation performance parameters are selected as parameters affecting the electrostatic spark sensitivity of energetic materials.

[0050] (3) Using the multidimensional influence parameter dataset corresponding to the existing energetic material samples as input parameters and the electrostatic spark sensitivity experimental value dataset of the existing energetic material samples as output parameters, the BP neural network is trained to obtain the electrostatic spark sensitivity value prediction model of energetic materials.

[0051] Furthermore, step (3) above specifically includes:

[0052] 1) Perform Pearson correlation analysis and clean up similar redundant variables on the data in the multidimensional influence parameter dataset corresponding to the existing energetic material samples.

[0053] 2) The data in the cleaned multidimensional influence parameter dataset corresponding to the existing energetic material sample and the data in the electrostatic spark sensitivity experimental value dataset corresponding to the existing energetic material sample are normalized using the min-max method to obtain the normalized multidimensional influence parameter dataset and the normalized electrostatic spark sensitivity experimental value dataset corresponding to the existing energetic material sample.

[0054] The normalization formula is as follows: In the formula, x* is the value after normalization, x is the value before normalization, and x... min =min(x), x max = max(x).

[0055] 3) Using the normalized multidimensional influence parameter dataset corresponding to the existing energetic material samples as input parameters and the normalized electrostatic spark sensitivity experimental value dataset of the existing energetic material samples as output parameters, the BP neural network is trained to obtain the electrostatic spark sensitivity value prediction model of energetic materials.

[0056] Furthermore, before performing step 3), the following is also included:

[0057] Based on the normalized multidimensional influence parameter dataset and the normalized electrostatic spark sensitivity experimental value dataset of the existing energetic material samples, a BP neural network is constructed. This step specifically includes:

[0058] Step ①: Using the normalized multidimensional influence parameter dataset corresponding to the existing energetic material sample as the input parameter, and the normalized electrostatic spark sensitivity experimental value dataset of the existing energetic material sample as the output parameter, the number of neurons in the input layer and the number of neurons in the output layer of the BP neural network are determined according to the output parameter and the input parameter.

[0059] The number of neurons in the input layer is determined by the number of input parameters. The number of neurons in the output layer is determined by the number of output parameters.

[0060] Step 2: Determine the number of hidden layers, and then preliminarily calculate the number of neurons in the hidden layers based on the number of neurons in the input layer and the number of neurons in the output layer.

[0061] Increasing the number of hidden layers can improve the prediction accuracy of a backpropagation (BP) neural network. However, setting too many hidden layers increases the complexity of training the BP neural network, leading to instability and overfitting. Therefore, when determining the number of hidden layers in a BP neural network, it is advisable to learn from previously established high-performing models. If no BP neural network is available for reference, it is generally advisable to start with 1 to 2 layers.

[0062] Increasing the number of neurons in the hidden layer can improve the prediction accuracy of a backpropagation (BP) neural network. However, setting too many neurons can lead to overfitting, significantly increasing training time and making convergence difficult. Historically, the number of neurons is generally between the number of neurons in the input and output layers. Therefore, a middle value is chosen first, and then other ranges are explored. The optimal number of neurons for training results is determined based on the trend of the BP neural network's training performance. The general calculation formula is: b is the number of neurons in the hidden layer, a is the number of neurons in the input layer (i.e., the number of input parameters), c is the number of neurons in the output layer (i.e., the number of output parameters), and γ is the adjustment constant, which typically ranges from 1 to 10.

[0063] The number of hidden layer neurons in the BP neural network is set within the above range, and the BP neural network is trained in an increasing trend to obtain the trend of mean squared error with the number of hidden layer neurons. Then, based on the trend, and taking into account the convergence and accuracy of the BP neural network, the optimal number of hidden layer neurons is determined.

[0064] Furthermore, step 3) specifically includes:

[0065] Using the normalized multidimensional influence parameter dataset corresponding to existing energetic material samples as input parameters and the normalized electrostatic spark sensitivity experimental value dataset of existing energetic material samples as output parameters, a prediction model for the electrostatic spark sensitivity value of energetic materials is obtained by continuously adjusting the network parameters of a BP neural network. The network parameter optimization includes: the number of hidden layer nodes, learning rate, number of iterations, maximum allowable error, and training function type.

[0066] During the learning and training process of the BP neural network, the learning rate that is most suitable for predicting electrostatic spark sensitivity is determined by continuous adjustment, and the appropriate number of training iterations and the maximum allowable error are set according to the convergence of the training.

[0067] Different training functions also have a certain impact on the performance of the network. For the BP neural network, the training function with the most stable error reduction and moderate training time is selected by comparing the training errors of different functions. After multiple model training attempts, the loss function curve of the BP neural network gradually stabilizes, and the BP neural network is fully trained, resulting in a prediction model for the electrostatic spark sensitivity value of energetic materials. The number of iterations at which no further training is required is taken as the final number of iterations for the prediction model.

[0068] The learning rate determines the amount of weight update during each training iteration of the network, and its selection is crucial. If the learning rate is too small, the error reduction will be very slow. To ensure the stability of network training, learning rates are typically set to 0.1, 0.01, 0.001, and 0.0001, and the final learning rate is determined by observing the loss behavior of the network in the initial stages.

[0069] The maximum allowable error is typically set between 0.001 and 0.00001. The system terminates the iterative calculation when the error between two iterations is less than this value. By training networks with different error values, the optimal allowable error value is obtained.

[0070] The training functions include trainlm using the Levenberg-Marquardt algorithm, traingda using the steepest gradient descent algorithm with a variable learning rate, and traingdm using the momentum backpropagation algorithm to correct the weights and thresholds.

[0071] In this embodiment of the invention, the quantum chemical calculation steps involved are as follows:

[0072] Quantum chemistry calculations typically require the use of Gaussian software. To balance computational resources and efficiency, this study employs the DFT method using the 6-31 G* basis set and B3LYP density functional in Gaussian 2009 to perform quantum chemical calculations on energetic material samples, outputting .out and .chk files. Specifically:

[0073] (1) Use databases such as PubChem and Chemical CAS Database to check and verify the molecular structure information of the selected energetic materials to ensure that all molecular structure information is accurate. (2) Use ChemDraw to draw the molecular structure to be calculated, and input the .cdx format file output by ChemDraw into Chem3D. Chem3D is used to check and modify the three-dimensional spatial structure model of the molecule and outputs it as a .gjf format file that can be read by Gaussian calculation software. (3) Input the constructed three-dimensional spatial structure model of the molecule into the DFT-B3LYP method set in Gaussian09 software, and perform global optimization of the input three-dimensional spatial structure of the molecule at the basis set level of 6-31G*. The optimized structure is subjected to frequency analysis and no imaginary frequency is found, ensuring that the obtained geometric configuration is in the minimum energy state. (4) When using the supercomputing center, first set the calculation level and the running memory and number of cores used when calling the supercomputing in Gaussian09 software. Since energetic material molecules are not considered macromolecules in the broader field of materials science, 10GB of memory and 64 cores were used for structure optimization and frequency calculations. These calculation parameters were input into the supercomputing platform as a .gjf example file. Within the supercomputing platform, after selecting the output folder and the supercomputing partition, the job was submitted, and calculations were performed to obtain the output .out and .chk files.

[0074] In this embodiment of the invention, the formula for calculating oxygen balance is: OB = 1600(z - 2x - 0.5y) / MW, where OB is the oxygen balance, x, y, and z are the number of C, H, and O atoms in the molecular formula, and MW is the molecular weight.

[0075] In this embodiment of the invention, the orbital energy gap (ΔE) gap ΔE refers to the absolute value of the difference between the energy of the highest occupied molecular orbital (HOMO) and the energy of the lowest unoccupied molecular orbital (LUMO). gap It can be used to measure the ease with which an electron transitions from a HOMO to a LUMO, ΔE gap A smaller value indicates a easier transition, and easier electronic transitions mean the molecule is more easily excited, meaning it is less stable. Electrostatic spark sensitivity essentially reflects the degree to which a molecule is excited; the ΔE value of the molecule... gap The smaller the value, the more sensitive the molecule. Import the output .chk file into GaussianView to view the orbital energies of HOMO and LUMO in the molecular structure information.

[0076] ΔE gap The formula for calculation is: ΔE gap =|E HOMO -E LUMO |

[0077] In this embodiment of the invention, nitro explosives constitute a large proportion of the energetic material. The N-NO2, C-NO2, or O-NO2 bonds in the molecular structure of the energetic material are generally considered to be the "trigger bonds" for decomposition. NO2 is a strong electron-withdrawing group; the more electrons near the NO2 group, the more stable the molecule, and the lower the electrostatic spark sensitivity. The electron configuration near the nitro group can be represented by the nitro charge. The sum of the Mulliken charges of all atoms of different types of nitro groups in each energetic molecule can be calculated using quantum chemistry. Then, the smallest nitro charge value is selected through comparison. Therefore, the process for determining the nitro charge in this embodiment of the invention is as follows: find the Mulliken charges of all atoms corresponding to any nitro group in the energetic material in the .out file; according to formula Q... Nitro =Q N +Q O1 +Q O2 Calculate the sum of the Mulliken charges of all atoms corresponding to any nitro group in the energetic material; compare and screen the sum of the Mulliken charges of all atoms corresponding to the nitro group, and determine the minimum sum of Mulliken charges as the nitro charge of the energetic material. Where Q N Q O1 Q O2 Q represents the Mulliken charge values ​​of the nitrogen atom and the two oxygen atoms in the nitro group, respectively. Nitro It refers to the sum of the Mulliken charges of the three atoms in the nitro group of an energetic material molecule.

[0078] In this embodiment of the invention, electron energy (E) ele Generally, it includes four items: (1) the kinetic energy of the electron; (2) the Coulomb repulsion energy between electrons; (3) the Coulomb repulsion energy between nuclei; and (4) the Coulomb attraction energy between the electron and the nucleus. The simplest task in quantum chemistry is the single-point calculation task, which refers to calculating the electron energy for a given geometric structure. This is also called single-point energy. Tasks other than single-point calculation can also provide electron energy, but their purpose is not to specifically calculate electron energy; electron energy is just generated incidentally. For example, in geometric structure optimization, the electron energy output at the last time corresponds to the electron energy under the optimized geometric structure. The electron energy data comes from the Summary in the Gaussian View window.

[0079] In this embodiment of the invention, the dipole moment is a physical quantity reflecting the charge distribution in a molecule. It can represent the charge symmetry of the molecule itself to a certain extent, thereby determining the symmetry of the molecular structure. The more symmetrical the molecular structure, the higher its stability, and the lower its sensitivity. The dipole moment data is obtained from the .out file.

[0080] In this embodiment of the invention, the enthalpy of formation refers to the thermal effect at a certain temperature on the formation of a unit amount of a pure substance from the most stable elemental forms of various elements in their standard states, also known as the heat of formation. The data for the enthalpy of formation is obtained from the .out file.

[0081] In this embodiment of the invention, molecular volume refers to the volume of the space inside the molecular surface. The average molar volume of the molecule is calculated by combining the optimized molecular stability configuration with the Monte Carlo method. Theoretical density refers to the ratio of molecular molar mass to molecular volume. Detonation parameters refer to the detonation velocity, detonation pressure, detonation heat, and detonation temperature of the energetic material.

[0082] Compared with the prior art, the technical solution provided by the embodiments of the present invention has the following advantages:

[0083] 1. Based on quantum chemical calculations and Pearson correlation screening, an optimal multi-dimensional set of electrostatic spark sensitivity influencing parameters was constructed, including orbital band gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, oxygen balance, and detonation performance. This multi-dimensional set of influencing parameters can significantly improve the degree of influence on electrostatic spark sensitivity.

[0084] 2. After fully training the BP neural network to learn the complex nonlinear relationship between electrostatic spark sensitivity and its multidimensional influencing parameter set, it has high accuracy in predicting the electrostatic spark sensitivity value of novel energetic materials in the laboratory synthesis stage. This makes it easy to efficiently and accurately determine in advance whether the designed and synthesized novel energetic materials meet the target electrostatic spark sensitivity performance requirements.

[0085] Example 2

[0086] This invention provides a method for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemistry, and its flowchart is shown below. Figure 2 As shown, it specifically includes:

[0087] (1) Collect and organize the dataset of electrostatic spark sensitivity experimental values, and use it as the output parameter. The selected electrostatic spark sensitivity experimental values ​​are mainly from the electrostatic spark sensitivity values ​​(E) given in the literature. ES All of these are experimental data values ​​tested using a unified RDAD method.

[0088] (2) A set of multidimensional influence parameters based on quantum chemical calculations was constructed and used as input parameters. After analysis, the characteristic multidimensional influence parameters with the highest correlation to the electrostatic spark sensitivity value of energetic materials were selected as: oxygen balance, orbital band gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, detonation velocity, and detonation pressure.

[0089] (3) A Pearson correlation analysis was performed on the above-mentioned multidimensional influencing parameters and the electrostatic spark sensitivity value. The resulting histogram is shown below. Figure 3 As shown.

[0090] (4) Use the min-max method to normalize the input and output parameters.

[0091] (5) Based on the output and input parameters, the number of neurons in the input and output layers of the BP neural network is determined to be 10 and 1, respectively. Referring to previous related prediction studies, and considering both neural network convergence and prediction accuracy, the number of hidden layers is determined to be 1. Based on the number of neurons in the input and output layers, the number of neurons in the hidden layers is initially calculated to be between 3 and 16. The number of hidden layer neurons in the neural network is set within this range, and the network is trained in an increasing trend. The mean squared error is then analyzed as a function of the number of hidden layer neurons. Figure 4 As shown, it can be seen that the mean square error of the neural network is minimized when the number of neurons in the hidden layer is 12. Therefore, considering both network convergence and accuracy, the number of neurons in the hidden layer is determined to be 12.

[0092] (6) Based on the parameters set above, a three-layer BP neural network initial model is constructed. The structure of this model is as follows: Figure 5 As shown. During the initial learning and training process of the BP neural network model, through continuous adjustments, the optimal learning rate for predicting electrostatic spark sensitivity was determined to be 0.001, and the maximum permissible error was set to 0.65 * 10^6. -3 Different training functions also have a certain impact on the performance of the network. For the BP neural network, by comparing the training errors of different functions, the `trainlm` function, which has the most stable error reduction and a moderate training time, was selected as the training function for the BP model. After multiple model training attempts, it was found that the loss function curve of the BP neural network model gradually stabilized after 16,000 training iterations, indicating that the network had been fully trained and no further iterations were needed. After training, the electrostatic spark sensitivity prediction model for energetic materials was obtained. The trend of the loss function of the BP neural network model is shown below. Figure 6 As shown in Table 1, the specific settings for the network parameters of the BP neural network model are listed below.

[0093] Table 1. Main parameter settings for the BP neural network prediction model

[0094]

[0095]

[0096] (7) Calculate the oxygen balance, orbital band gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, detonation velocity and detonation pressure of the untrained novel energetic material 1,4,5,8-Tetranitro-1,4,5,8-tetraazadecaline (TNAD). The specific data and experimental values ​​of electrostatic spark sensitivity are in Table 2.

[0097] Table 2. Multidimensional Influence Parameter Set of TNAD

[0098]

[0099] Note: ΔE, Q nitro E ele Dipole, D, ΔH f The units for V, ρ, P, and experimental values ​​are: .au, e, .au, Debye, km / s, KJ / mol, cm 3 / mol, g / cm 3 Gpa and J.

[0100] The data from the TNAD multidimensional influence parameter set in Table 2 are input into the electrostatic spark sensitivity prediction model of energetic materials established in step (6), and the output electrostatic spark sensitivity prediction value of TNAD is 4.974J.

[0101] Example 3

[0102] Compared with Example 2, the difference is that the energetic material in step (7) is 1,3,5,7,9-Pentanitro-1,3,5,7,9-pentaazacyclodecane (DECAGEN), and the specific data of its oxygen balance, orbital band gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, detonation velocity and detonation pressure, as well as the experimental values ​​of electrostatic spark sensitivity, are shown in Table 3.

[0103] Table 3. Dataset of DECAGEN's Multidimensional Influence Parameters

[0104]

[0105] Note: ΔE, Q nitro E ele Dipole, D, ΔH f The units for V, ρ, P, and experimental values ​​are: .au, e, .au, Debye, km / s, KJ / mol, cm 3 / mol, g / cm 3 Gpa and J.

[0106] The data from the DECAGEN multidimensional influence parameter set in Table 3 were input into the electrostatic spark sensitivity prediction model for energetic materials established in step (6), and the output electrostatic spark sensitivity prediction value of DECAGEN was 2.706 J.

[0107] Example 4

[0108] Compared with Example 2, the difference is that the energetic material in step (7) is 1,3-Dinitro-1,3-diazacyclobutane (TETROGEN), and its specific data on oxygen balance, orbital gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, detonation velocity and detonation pressure, as well as the experimental values ​​of electrostatic spark sensitivity, are shown in Table 4.

[0109] Table 4. Set of Multidimensional Influence Parameters for TETROGEN

[0110]

[0111] Note: ΔE, Q nitro E ele Dipole, D, ΔH f The units for V, ρ, P, and experimental values ​​are: .au, e, .au, Debye, km / s, KJ / mol, cm 3 / mol, g / cm 3 Gpa and J.

[0112] The data from the TETROGEN multidimensional influence parameter set in Table 4 are input into the electrostatic spark sensitivity prediction model of energetic materials established in step (6), and the output electrostatic spark sensitivity prediction value of TETROGEN is 5.867J.

[0113] Example 5

[0114] Compared with Example 2, the difference is that the energetic material in step (7) is 1,9-Diacetoxy-2,4,6,8-tetranitro-2,4,6,8-tetraazanonane (ACAN), and the specific data of its oxygen balance, orbital gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, detonation velocity and detonation pressure, as well as the experimental values ​​of electrostatic spark sensitivity, are shown in Table 5.

[0115] Table 5. ACAN Multidimensional Influence Parameter Set Table

[0116]

[0117] Note: ΔE, Q nitro E ele Dipole, D, ΔHf The units for V, ρ, P, and experimental values ​​are: .au, e, .au, Debye, km / s, KJ / mol, cm 3 / mol, g / cm 3 Gpa and J.

[0118] The data from the ACAN multidimensional influence parameter set in Table 5 are input into the electrostatic spark sensitivity prediction model of energetic materials established in step (6), and the output electrostatic spark sensitivity prediction value of ACAN is 15.966J.

[0119] Table 6 shows the experimental and predicted values ​​of the test set samples obtained in Examples 2, 3, 4, and 5. A direct comparison graph of the predicted and experimental values ​​is shown below. Figure 7 As shown, the data points in the test set are all closely clustered around y = x, indicating a high degree of clustering and suggesting that the prediction accuracy of the electrostatic spark sensitivity prediction model for energetic materials is very high. The residual plot of the predicted values ​​is shown below. Figure 8 As shown, it can be seen that the data points in both the training and test sets closely match y=0. The residuals of most data points in the model are randomly distributed around y=0 and are uniformly distributed, indicating that no significant systematic error was generated during the modeling process. The Williams plots for the training and prediction sets are shown below. Figure 9 As shown, Figure 9 The three lines represent y = ±3 and the lever warning value h* = 0.733, respectively. The rectangular area drawn by these three lines is the application domain of the electrostatic spark sensitivity prediction model for energetic materials. Figure 9 It can be seen that all data points in the prediction set of the energetic material electrostatic spark sensitivity prediction model are within the application domain, and no outliers appear. This proves that although the data used includes various energetic materials, the energetic material electrostatic spark sensitivity prediction model can still accurately predict them.

[0120] The specific prediction fitting parameters of the BP neural network model on the training and test sets are as follows: MSE is 0.3478 and 1.1412, RMSE is 0.5897 and 1.0683, MAE is 0.3956 and 0.7822, and the correlation coefficient R between the training and test sets of the BP neural network model is... 2 The values ​​are 0.9843 and 0.9322, respectively.

[0121] Table 6 compares the predicted values ​​and experimental values ​​of the test set in the embodiments.

[0122]

[0123] Example 6

[0124] In order to implement the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, a quantum chemistry-based electrostatic spark sensitivity prediction system for energetic materials is provided below.

[0125] like Figure 10 As shown in the figure, an embodiment of the present invention provides an electrostatic spark sensitivity prediction system for energetic materials based on quantum chemistry, comprising:

[0126] Novel energetic material sample acquisition module 1 is used to acquire novel energetic material samples.

[0127] The multidimensional influence parameter dataset construction module 2 is used to perform quantum chemical calculations on the novel energetic material sample to obtain the multidimensional influence parameter dataset corresponding to the novel energetic material sample; the multidimensional influence parameter dataset includes orbital band gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, oxygen balance and detonation parameters.

[0128] The electrostatic spark sensitivity prediction module 3 is used to input the multidimensional influence parameter dataset corresponding to the novel energetic material sample into the energetic material electrostatic spark sensitivity prediction model to obtain the electrostatic spark sensitivity value corresponding to the novel energetic material sample.

[0129] The training process for the electrostatic spark sensitivity prediction model for energetic materials is as follows:

[0130] Obtain a dataset of electrostatic spark sensitivity experimental values ​​for existing energetic material samples.

[0131] Quantum chemical calculations were performed on the existing energetic material samples to obtain a multidimensional influence parameter dataset corresponding to the existing energetic material samples.

[0132] Using the multidimensional influence parameter dataset corresponding to the existing energetic material samples as input parameters and the electrostatic spark sensitivity experimental value dataset of the existing energetic material samples as output parameters, a BP neural network is trained to obtain an energetic material electrostatic spark sensitivity value prediction model.

[0133] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0134] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present 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 the present invention.

Claims

1. A method for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemistry, characterized in that, include: Obtain samples of novel energetic materials; Quantum chemical calculations were performed on the novel energetic material sample to obtain a multidimensional influence parameter dataset corresponding to the novel energetic material sample; the multidimensional influence parameter dataset includes orbital band gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, oxygen balance and detonation parameters; the quantum chemical calculation process is as follows: in Gaussian09 software, the DFT calculation method of 6-31G* basis set and B3LYP density functional is selected to perform quantum chemical calculations on the energetic material sample, and output .out file and .chk file, specifically including: (1) using PubChem, Chemical The CASDatabase database checks and verifies the molecular structure information of the selected energetic materials to ensure that all molecular structure information is accurate; (2) Use ChemDraw to draw the molecular structure to be calculated, and input the .cdx format file output by ChemDraw into Chem3D. Chem3D is used to check and modify the three-dimensional spatial structure model of the molecule and outputs it as a .gjf format file that can be read by Gaussian calculation software; (3) Input the constructed three-dimensional spatial structure model of the molecule into the DFT-B3LYP method set in Gaussian09 software, and in the 6-31G* base At the group level, the input molecular three-dimensional spatial structure is globally optimized, and the optimized structure is analyzed by frequency analysis and there is no imaginary frequency, ensuring that the obtained geometric configuration is in a minimum energy state; (4) When using the supercomputing center, first set the calculation level and the running memory and number of cores used when calling the supercomputing in the Gaussian09 software, and set 10GB of memory and 64 cores for calculation when performing structure optimization and frequency calculation; input the calculation parameters into the supercomputing platform in the form of .gjf example file; in the supercomputing platform, after selecting the output folder and the supercomputing partition to be called, submit the job and perform calculation to obtain the output .out file and .chk file; The multidimensional influence parameter dataset corresponding to the novel energetic material sample is input into the electrostatic spark sensitivity value prediction model of energetic material to obtain the electrostatic spark sensitivity value corresponding to the novel energetic material sample. The training process for the electrostatic spark sensitivity prediction model for energetic materials is as follows: Obtain a dataset of electrostatic spark sensitivity experimental values ​​for existing energetic material samples; Quantum chemical calculations were performed on the existing energetic material samples to obtain a multidimensional influence parameter dataset corresponding to the existing energetic material samples; Using the multidimensional influence parameter dataset corresponding to the existing energetic material samples as input parameters and the electrostatic spark sensitivity experimental value dataset of the existing energetic material samples as output parameters, a BP neural network is trained to obtain an energetic material electrostatic spark sensitivity value prediction model; wherein, the training function adopts the trainlm function of the Levenberg-Marquardt algorithm, the steepest gradient descent algorithm traingda with a variable learning rate, and the traingdm function with momentum BP algorithm to correct the weights and thresholds.

2. The method for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemistry according to claim 1, characterized in that, The multidimensional influence parameter dataset corresponding to the novel energetic material sample is input into the electrostatic spark sensitivity prediction model for energetic materials to obtain the electrostatic spark sensitivity value corresponding to the novel energetic material sample, specifically including: Pearson correlation analysis and cleaning of similar redundant variables were performed on the data in the multidimensional influence parameter dataset corresponding to the novel energetic material sample. The cleaned data was normalized using the min-max method to obtain the normalized multidimensional influence parameter dataset corresponding to the novel energetic material sample. The normalized multidimensional influence parameter dataset corresponding to the novel energetic material sample is input into the electrostatic spark sensitivity prediction model of the energetic material to obtain the electrostatic spark sensitivity value corresponding to the novel energetic material sample.

3. The method for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemistry according to claim 1, characterized in that, Using the multidimensional influence parameter dataset corresponding to the existing energetic material samples as input parameters and the electrostatic spark sensitivity experimental value dataset of the existing energetic material samples as output parameters, a BP neural network is trained to obtain an energetic material electrostatic spark sensitivity value prediction model, specifically including: Pearson correlation analysis and cleaning of similar redundant variables were performed on the data in the multidimensional influence parameter dataset corresponding to the existing energetic material samples. The min-max method was used to normalize the data in the cleaned multidimensional influence parameter dataset corresponding to the existing energetic material sample and the data in the electrostatic spark sensitivity experimental value dataset of the existing energetic material sample, so as to obtain the normalized multidimensional influence parameter dataset and the normalized electrostatic spark sensitivity experimental value dataset of the existing energetic material sample. Using the normalized multidimensional influence parameter dataset corresponding to the existing energetic material samples as input parameters and the normalized electrostatic spark sensitivity experimental value dataset of the existing energetic material samples as output parameters, a BP neural network is trained to obtain an energetic material electrostatic spark sensitivity value prediction model.

4. The method for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemistry according to claim 3, characterized in that, Before the step of training the BP neural network to obtain the electrostatic spark sensitivity prediction model for energetic materials using the normalized multidimensional influence parameter dataset corresponding to the existing energetic material samples as input parameters and the normalized electrostatic spark sensitivity experimental value dataset of the existing energetic material samples as output parameters, the following steps are also included: A BP neural network is constructed based on the normalized multidimensional influence parameter dataset corresponding to the existing energetic material samples and the normalized electrostatic spark sensitivity experimental value dataset of the existing energetic material samples.

5. The method for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemistry according to claim 1, characterized in that, The formula for calculating the oxygen balance is: OB = 1600(z - 2x - 0.5y) / MW; In the formula, OB represents oxygen balance, x, y, and z represent the number of C, H, and O atoms in the molecular formula, and MW represents the molecular weight.

6. The method for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemistry according to claim 1, characterized in that, The calculation process for the orbital energy gap is as follows: Input the output .chk file into GaussianView to determine the highest occupied molecular orbital energy and the lowest unoccupied molecular orbital energy in the molecular structure information of energetic materials; According to the formula ΔE gap =|E HOMO -E LUMO |Calculate the orbital bandgap; In the formula, ΔE gap For the orbital band gap, E HOMO E is the highest occupied molecular orbital energy. LUMO This represents the energy of the lowest unoccupied molecular orbital.

7. The method for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemistry according to claim 1, characterized in that, The calculation process for the nitro charge is as follows: Find the Mulliken charge of all atoms corresponding to any nitro group in the energetic material in the .out file; According to formula Q Nitro =Q N +Q O1 +Q O2 Calculate the sum of the Mulliken charges of all atoms corresponding to any nitro group in the energetic material; The sum of the Mulliken charges of all atoms corresponding to the nitro group is compared and screened, and the minimum sum of Mulliken charges is determined as the nitro charge of the energetic material; In the formula, Q Nitro Q represents the sum of the Mulliken charges of the three atoms in the nitro group of an energetic material molecule. N Q O1 Q O2 These are the Mulliken charges of the nitrogen atom and the two oxygen atoms in the nitro group, respectively.

8. The method for predicting the electrostatic spark sensitivity of energetic materials based on quantum chemistry according to claim 1, characterized in that, The electron energy data comes from the Summary in the Gaussian View window; the dipole moment data and the enthalpy of generation data both come from the .out file.

9. A quantum chemistry-based electrostatic spark sensitivity prediction system for energetic materials, characterized in that, include: A novel energetic material sample acquisition module is used to acquire samples of novel energetic materials. The multidimensional influence parameter dataset construction module is used to perform quantum chemical calculations on the new energetic material sample to obtain the multidimensional influence parameter dataset corresponding to the new energetic material sample; the multidimensional influence parameter dataset includes orbital band gap, nitro charge, electron energy, dipole moment, enthalpy of formation, molecular volume, theoretical density, oxygen balance and detonation parameters; the quantum chemical calculation process is as follows: in Gaussian09 software, the DFT calculation method of 6-31G* basis set and B3LYP density functional is selected to perform quantum chemical calculations on the energetic material sample and output .out file and .chk file, specifically including: (1) using PubChem, Chemical The CASDatabase database checks and verifies the molecular structure information of the selected energetic materials to ensure that all molecular structure information is accurate; (2) Use ChemDraw to draw the molecular structure to be calculated, and input the .cdx format file output by ChemDraw into Chem3D. Chem3D is used to check and modify the three-dimensional spatial structure model of the molecule and outputs it as a .gjf format file that can be read by Gaussian calculation software; (3) Input the constructed three-dimensional spatial structure model of the molecule into the DFT-B3LYP method set in Gaussian09 software, and in the 6-31G* base At the group level, the input molecular three-dimensional spatial structure is globally optimized, and the optimized structure is analyzed by frequency analysis and there is no imaginary frequency, ensuring that the obtained geometric configuration is in a minimum energy state; (4) When using the supercomputing center, first set the calculation level and the running memory and number of cores used when calling the supercomputing in the Gaussian09 software, and set 10GB of memory and 64 cores for calculation when performing structure optimization and frequency calculation; input the calculation parameters into the supercomputing platform in the form of .gjf example file; in the supercomputing platform, after selecting the output folder and the supercomputing partition to be called, submit the job and perform calculation to obtain the output .out file and .chk file; The electrostatic spark sensitivity prediction module is used to input the multidimensional influence parameter dataset corresponding to the novel energetic material sample into the energetic material electrostatic spark sensitivity prediction model to obtain the electrostatic spark sensitivity value corresponding to the novel energetic material sample. The training process for the electrostatic spark sensitivity prediction model for energetic materials is as follows: Obtain a dataset of electrostatic spark sensitivity experimental values ​​for existing energetic material samples; Quantum chemical calculations were performed on the existing energetic material samples to obtain a multidimensional influence parameter dataset corresponding to the existing energetic material samples; Using the multidimensional influence parameter dataset corresponding to the existing energetic material samples as input parameters and the electrostatic spark sensitivity experimental value dataset of the existing energetic material samples as output parameters, a BP neural network is trained to obtain an energetic material electrostatic spark sensitivity value prediction model; wherein, the training function adopts the trainlm function of the Levenberg-Marquardt algorithm, the steepest gradient descent algorithm traingda with a variable learning rate, and the traingdm function with momentum BP algorithm to correct the weights and thresholds.

Images