Compound crystal density prediction model construction method and prediction method

Abstract

The invention relates to a compound crystal density prediction model construction method and a prediction method. A multi-molecular structure third-order volume characteristic variable description method and numerical values of first-order characteristics, second-order characteristics and third-order characteristics of atomic elements are utilized to form descriptor vectors; according to the method, the crystal density of the novel single-molecule energetic compound can be rapidly predicted on the basis of the molecular SMILES code, verification shows that the method has the advantages of being simple, efficient and accurate in prediction of the crystal density, the method is not limited by the molecular variety of the energetic compound, the universality is good, the average absolute prediction error of the crystal density of various energetic molecules is 0.026 g / cm < 3 >, and the gold standard of 0.030 g / cm < 3 > is broken through. The method is of great significance to design, construction and research of energetic molecule descriptors, and can guide design and synthesis of more energetic compounds with high energy density and high detonation performance.

Description

Technical Field

[0001] The present invention designs a theoretical prediction method for the crystal density of energetic compounds based on molecular structure, which is applicable to reported or newly synthesized single-molecule energetic compounds. Background Art

[0002] Crystal density is an important indicator for measuring the energy density and detonation performance of energetic compounds, and its accurate prediction is of great scientific significance. At present, there is no efficient and accurate method for predicting the crystal density of single-molecule energetic compounds based on the molecular SMILES code as the only input.

[0003] The crystal density of energetic compounds is determined by their molecular arrangement at the crystal level, which contains three-dimensional spatial information at the crystal scale, while the molecular SMILES code only contains one-dimensional information at the molecular scale of the compound. Therefore, the crystal density prediction based on the SMILES code involves the process of dimensionality upgrading of molecular characteristic information.

[0004] At present, the prediction method with more development potential is to convert the information in the molecular SMILES into the molecular three-dimensional structure coordinates, and then extract the features related to the crystal packing arrangement in the molecular three-dimensional structure coordinates to form the energetic compound molecular descriptor associated with the crystal density, and use machine learning algorithm training to obtain the mathematical mapping law from the energetic compound molecular descriptor to the crystal density, and then build a machine learning model for prediction. However, the prediction accuracy of the reported machine learning models is generally not ideal, ranging from 0.033 to 0.040 g / cm 3 The level has not yet exceeded 0.030g / cm 3 The “gold standard”. Summary of the invention

[0005] In view of the defects or shortcomings of the prior art, the present invention provides a method for constructing a compound crystal density prediction model.

[0006] To this end, the method for constructing a compound crystal density prediction model provided by the present invention comprises the following steps:

[0007] Step 1, constructing a data set, wherein the data set includes crystal structure data and descriptor data vectors of a plurality of organic compound samples;

[0008] The crystal structure data include unit cell parameters of the corresponding compound and three-dimensional Cartesian coordinates of each atom in the unit cell;

[0009] The descriptor data vector is composed of N first-order eigenvalues, N first-order contribution eigenvalues, M second-order eigenvalues, M second-order eigenvalue contributions, L third-order eigenvalues ​​and L third-order eigenvalue contributions; N is the number of atomic species in the corresponding compound, M is the number of diatomic combinations between different atoms in the N atoms and the sum of the number of diatomic species composed of various atoms themselves, M=C N 2 + N; L is the sum of the number of types of triatomic combinations between different atoms in N atoms and the number of triatomic types composed of various atoms themselves, L = C N 3 +2C N 2 +N;

[0010] in:

[0011] Any one element of the N first-order eigenvalues ​​is: the sum of the van der Waals volumes of all atoms belonging to the n-type atoms in the sequence of atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule state of the organic compound, and the n-type atoms represent any one of the N types of atoms;

[0012] V i is the van der Waals volume occupied by the atom number i in the three-dimensional Cartesian coordinates of the isolated molecule of the organic compound;

[0013] Any one element of the N first-order contribution eigenvalues ​​is: the sum of the van der Waals volumes of all atoms belonging to the n-type atoms in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the organic compound crystal structure state, and the n-type atoms represent any one of the N types of atoms;

[0014] V i ' is the van der Waals volume occupied by the atom number i in the three-dimensional Cartesian coordinates of the organic compound crystal structure;

[0015] Any one element of the M second-order eigenvalues ​​is: the sum of the van der Waals volumes of all diatomic combinations belonging to m types of diatomic combinations in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule state of the organic compound, and the m types of diatomic combinations represent any type of diatomic combinations among the M types of diatomic combinations;

[0016]

[0017] i and j are two atomic numbers in the sequence of atoms constituting the corresponding compound, i≠j;

[0018] V ij is the van der Waals volume of the diatomic combination consisting of atoms numbered i and j in the three-dimensional Cartesian coordinates of an isolated molecule of an organic compound;

[0019] A i and A j are the first pre-exponential factors of the diatomic combined volume calculation function of the atoms with serial numbers i and j, respectively;

[0020] B i and B j are the second pre-exponential factors of the diatomic combined volume calculation function of the atoms with serial numbers i and j, respectively;

[0021] a i and a j are the first exponential factors of the diatomic combined volume calculation function of atoms numbered i and j respectively;

[0022] b i and b j are the second exponential factors of the diatomic combined volume calculation function of the atoms with serial numbers i and j, respectively;

[0023] α is the exponential factor of the diatomic combination volume calculation function;

[0024] R ij is the geometric space distance between atom number i and atom number j;

[0025] R cut2，i and R cut2，j are the diatomic combination cutoff radii of atoms numbered i and j respectively;

[0026] Any one of the M second-order contribution eigenvalues ​​is: the sum of the van der Waals volumes of all diatomic combinations belonging to m types of diatomic combinations in the sequence of atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the organic compound crystal structure state, and the m types of diatomic combinations represent any type of diatomic combination among the M types of diatoms;

[0027] V ij ' =V ij -V i '-V j ' (2)

[0028] V ij ' is the van der Waals volume of the diatomic combination of atoms numbered i and j in the three-dimensional Cartesian coordinates of the organic compound crystal structure;

[0029] V i ' is the van der Waals volume occupied by the atom number i in the three-dimensional Cartesian coordinates of the organic compound crystal structure;

[0030] V j ' is the van der Waals volume occupied by the atom with serial number j in the three-dimensional Cartesian coordinates of the crystal structure of the organic compound;

[0031] Any one of the L third-order eigenvalues ​​is: the sum of the van der Waals volumes of all the three-atom combinations belonging to the l-type three-atom combination in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule state of the organic compound, and the l-type three-atom combination is any one of the L types of three-atom combinations;

[0032]

[0033] i, j and k are the three atomic numbers of the sequential atoms constituting the corresponding compound, i≠j≠k;

[0034] V ijk is the van der Waals volume of the three-atom combination consisting of atoms numbered i, j and k in the three-dimensional Cartesian coordinates of the isolated molecule state of the organic compound;

[0035] C i , C j and C k Calculate the first pre-exponential factor for the volume of the three-atom combination of atom number i, atom number j and atom number k respectively;

[0036] c i 、c j and c k Calculate the second pre-exponential factor for the volume of the three-atom combination of atom number i, atom number j and atom number k respectively;

[0037] D i , D j and D k They are the intercept terms for calculating the volume of the three-atom combination of atom number i, atom number j and atom number k respectively;

[0038] β is the exponential factor of the volume calculation function of the three-atom combination;

[0039] R ij is the geometric distance between atoms numbered i and j; R ik is the geometric distance between atom number i and atom number k; R kj is the geometric space distance between atom number k and atom number j;

[0040] R cut3，i , R cut3，j and R cut3，k are the three-atom combination cutoff radius of atom number i, the three-atom combination cutoff radius of atom number j, and the three-atom combination cutoff radius of atom number k respectively;

[0041] Any one of the L third-order contribution eigenvalues ​​is: the sum of the van der Waals volumes of all the three-atom combinations belonging to the l-type three-atom combination in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule state of the organic compound, and the l-type three-atom combination is any one of the L types of three-atom combinations;

[0042] V ijk '=V ijk -V ij ' -V ik ' -V jk ' (4)

[0043] V ijk ' is the van der Waals volume occupied by the three-dimensional Cartesian coordinates of the three-atom combination of atoms numbered i, j and k in the crystal structure state of the organic compound;

[0044] V ij ' is the van der Waals volume of the diatomic combination of atoms numbered i and j in the three-dimensional Cartesian coordinates of the organic compound crystal structure;

[0045] V ik ' is the van der Waals volume of the diatomic combination of atoms numbered i and k in the three-dimensional Cartesian coordinates of the organic compound crystal structure;

[0046] V jk ' is the van der Waals volume of the diatomic combination of atoms numbered k and j in the three-dimensional Cartesian coordinates of the organic compound crystal structure;

[0047] Step 2, fitting the crystal density and descriptor data vectors of several organic compounds to obtain a prediction model for the crystal density of energetic compounds, where the crystal density is the dependent variable and the descriptor data is the independent variable.

[0048] Optionally, the organic compound is an energetic compound.

[0049] An optional solution is that the elements contained in the organic compound are a combination of any two or more of the four elements of carbon, hydrogen, nitrogen and oxygen.

[0050] An optional solution is that the crystal density of the organic compound samples ranges from 1.40 to 2.05 g / cm 3 .

[0051] An optional solution is that the three-dimensional Cartesian coordinates of the organic compound are obtained by processing the SMILES code of the energetic compound using the Rdkit program to obtain an initial three-dimensional structure, and then optimizing the initial three-dimensional structure using the GFN2-xTB method to obtain the three-dimensional Cartesian coordinates.

[0052] An optional solution is that the fitting in step 2 adopts a kernel ridge regression algorithm, and the kernel function in the kernel ridge regression algorithm parameters adopts a radial basis function.

[0053] An optional solution is that the crystal structure data of the organic compound in step 1 is obtained by using an X-ray diffractometer measurement method, and the crystal density is calculated based on the crystal structure data as a standard value for model training.

[0054] The present invention also provides a method for predicting compound crystal density, which comprises: using the above method to obtain characteristic data of organic compounds, using the characteristic data as input of a model constructed by the above method, and calculating the crystal density of the organic compound to be predicted.

[0055] The present invention utilizes the third-order volume characteristic variable description method of multi-molecular structure and the numerical values ​​of the first-order characteristics, second-order characteristics and third-order characteristics of each atomic element to form a descriptor vector, which can realize the rapid prediction of the crystal density of new single-molecule energetic compounds based on the molecular SMILES code. It has been verified that the invention has the characteristics of simple, efficient and accurate prediction of crystal density, is not limited to the type of molecular compounds, has good universality, and is of great significance to the design and construction research of energetic molecular descriptors, and can guide the design and synthesis of more energetic compounds with high energy density and high detonation performance. The machine learning prediction model established based on a relatively small sample data set (less than 5,000 energetic material data samples) has an average absolute prediction error of 0.026g / cm for the crystal density of various types of energetic molecules. 3 , breaking through 0.030g / cm 3 The “gold standard”.

[0056] The present invention calculates the first-order characteristics, second-order characteristics and third-order characteristics of the molecular structure of energetic materials through analytical formulas, evaluates the linearly correlated components of the volume of energetic molecules, and uses them as molecular descriptors to quickly and accurately predict the crystal density. In the feature descriptor construction stage, sufficient data that is highly numerically correlated with the crystal density of energetic compounds is obtained, which effectively reduces the amount of training sample data required to establish an accurate prediction model, and greatly improves the training accuracy and prediction ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 The initial three-dimensional molecular structure of the energetic material TNT.

[0058] Figure 2 The optimized three-dimensional molecular structure of the energetic material TNT.

[0059] Figure 3 Initial three-dimensional molecular structure of energetic material CL-20.

[0060] Figure 4Optimized three-dimensional molecular structure of energetic material CL-20. DETAILED DESCRIPTION

[0061] Unless otherwise specified, the scientific and technical terms used herein are understood according to the knowledge of ordinary technicians in the relevant fields.

[0062] The present invention evaluates the space occupation and arrangement of the molecular volume in the unit cell of energetic compounds in the crystal state based on the concept of molecular volume. The parameters of the volume occupation and volume contribution characteristics of the first order (single atom), second order (i.e., diatomic combination, the diatomic combination of the present invention does not consider whether the diatomic atoms are chemically bonded in the related compounds), and third order (i.e., triatomic combination, the triatomic combination of the present invention does not consider whether the related atoms are chemically bonded in the related compounds) are calculated by statistical methods and first principle simulations and form structural descriptors. A new machine learning prediction model is constructed using the descriptors. The crystal density prediction error of energetic compounds is reduced to 0.030 g / cm 3 the following.

[0063] It should be noted that the several energetic compound samples in step 1 of the scheme of the present invention are selected according to the object to be predicted. Specifically, multiple compounds can be selected according to the constituent atoms and crystal characteristics of the object to be predicted. To ensure the accuracy of the prediction results, the series of organic compounds are selected as much as possible to have relevant atomic types and reasonable crystal density parameter values ​​with the object to be predicted.

[0064] In the solution of the present invention, V i and V i ' are all statistical parameters, that is, they are obtained through statistics on a series of organic compounds. The series of organic compounds are multiple compounds selected according to the constituent atoms and crystal characteristics of the object to be predicted. In order to ensure the accuracy of the prediction results, the series of organic compounds are selected as much as possible to have relevant atomic types and reasonable crystal density parameter values ​​with the object to be predicted.

[0065] Specifically, V i The value depends on the type of atom to which the number i atom belongs. The van der Waals volume occupied by different types of atoms in the three-dimensional Cartesian coordinates of isolated molecules of organic compounds can be calculated by first-principles simulation and statistical methods (the statistical range can be flexibly determined according to the prediction object, in order to obtain accurate and reliable prediction results. For example, to predict the crystal density of energetic compounds containing C, H, N, and O atoms, the relevant statistical range can be selected from the Cambridge University CCDC crystal database with a crystal density greater than 1.40 g / cm 3 , the theoretical detonation velocity is greater than 6000m / s for about 1000 kinds of energetic compounds).

[0066] V iThe value of ' depends on the type of atom to which the number i belongs. The van der Waals volume occupied by different types of atoms in the three-dimensional Cartesian coordinates of the crystal structure of organic compounds can be calculated by first-principles simulation and statistical methods (the statistical range can be the crystal density greater than 1.40g / cm in the CCDC crystal database of Cambridge University). 3 , the theoretical detonation velocity is greater than 6000m / s for about 1000 kinds of energetic compounds).

[0067] The values ​​of the pre-exponential factor, exponential factor, intercept term, diatomic combination cutoff radius and triatomic combination cutoff radius involved in formulas (1) and (3) of the present invention depend on the atomic species to which the corresponding numbered atoms belong. The values ​​of the relevant parameters of different atomic species can be obtained by first-principles simulation calculation and statistical methods (the statistical range can be flexibly determined according to the prediction object, so as to obtain accurate and reliable prediction results. For example, to predict the crystal density of energetic compounds containing C, H, N, and O atoms, the relevant statistical range can be selected from the crystal density greater than 1.40 g / cm in the Cambridge University CCDC crystal database. 3 , the theoretical detonation velocity is greater than 6000m / s for about 1000 energetic compounds). The exponential factors α and β can be selected based on experience. According to the inventor's experience, the reasonable value range of α is 0.5-10.0, and the reasonable value range of β is 0.5-10.0.

[0068] Taking C, H, N, and O atoms as examples, the process of determining the values ​​of the above parameters and their specific values ​​are explained:

[0069] N = 4, through first principle simulation calculation and statistical method (the statistical range can be the crystal density greater than 1.40g / cm in the Cambridge University CCDC crystal database). 3 , about 1000 kinds of energetic compound molecules with theoretical detonation velocity greater than 6000m / s), the V value and V' value of each type of atom are shown in Table 1.

[0070] Table 1 Values ​​of V and V' for HCNO elements

[0071] element V V＇ H 3.99 7.31 C 12.81 11.78 N 12.10 11.22 O 12.05 11.03

[0072] The process of determining the values ​​of the parameters in Table 1 specifically takes H atoms as an example. The process of determining the parameter V is as follows: the electronic distribution states of molecules of about 1000 energetic compounds in the Cambridge University CCDC crystal database in an isolated state are calculated by first principles, and the spatial volume occupation of all H atoms in these compounds is counted, and the probability of H atoms appearing in different structures and functional groups is considered, and the weighted average of the spatial volume occupation of H atoms is obtained as V;

[0073] The process of determining V' is as follows: by calculating the electron distribution state of the molecules of the above-mentioned approximately 1,000 energetic compounds in the crystal structure through first-principles calculations, the spatial volume occupied by all H atoms in these compounds is counted, and the probability of H atoms appearing in different structures and functional groups is considered, and the weighted average of the spatial volume occupied by H atoms is obtained as V';

[0074] Furthermore, M = 10, where 10 second-order eigenvalues ​​and 10 second-order eigencontribution values ​​are the values ​​corresponding to the diatomic combinations of HH, CC, NN, OO, HC, HN, HO, CN, CO, and NO, which can be calculated according to equations (1) and (2). The empirical value of α in equation (1) is 4.0, and the other parameters A, B, a, b, and R cut2 The values ​​of etc. are shown in Table 2.

[0075] Table 2 HCNO element pairs A, B, a, b and R cut2 The value of

[0076] element A a B b <![CDATA[R cut2 ]]> H 19.15 0.73 1.95 -0.80 0.375 C 16.21 0.72 0.73 -5.66 0.770 N 18.74 0.62 1.74 -7.06 0.735 O 21.52 0.51 1.08 -8.40 0.745

[0077] The process of determining the values ​​of the parameters in Table 2 is to take H atom as an example. The process of determining the parameters A and a is as follows: the electron distribution state of about 1000 energetic compounds in the Cambridge University CCDC crystal database in an isolated state is calculated by first principles, and the atomic distances of these compounds are less than The spatial volume occupied by H atoms in a diatomic system containing H atoms varies with the interatomic distance. The pre-exponential factor of the exponential function is obtained by statistical fitting as A and the exponential factor is a;

[0078] The process of determining parameters B and b is as follows: first-principles calculations are used to determine the electron distribution states of about 1,000 energetic compounds in the crystal database, and the atomic distances of these compounds are less than The spatial volume occupied by H atoms in a diatomic system containing H atoms varies with the interatomic distance. The pre-exponential factor of the exponential function is obtained by statistical fitting as B and the exponential factor is b;

[0079] R cut2 The determination process is: first-principles calculation of the electron distribution state of about 1000 energetic compounds in the above crystal database in an isolated state, and the atomic distances of these compounds are less than The atomic radius of the H atom in the diatomic system containing the H atom, and considering the probability of H atoms appearing in different structures and functional groups, the weighted average of the spatial volume occupied by the H atom is obtained as R cut2 ;

[0080] The corresponding L = 20, where 20 third-order eigenvalues ​​and 20 third-order eigencontribution values ​​are the values ​​corresponding to various triatomic combinations of HHH, CCC, NNN, OOO, HHC, HHN, HHO, CCH, CCN, CCO, NNH, NNC, NNO, OOH, OOC, OON, HCN, HCO, HNO, and CNO, which are calculated according to equations (3) and (4). The empirical value of β in equation (3) is 4.0. The other parameters C, D, c, and R cut3 The values ​​of etc. are shown in Table 3.

[0081] Table 3 HCNO elements for C, D, C and R cut3 The value of

[0082] element C c D <![CDATA[R cut3 ]]> H 4.73 0.72 21.87 0.375 C 51.11 0.15 35.89 0.770 N 9.01 0.46 15.92 0.735 O 14.94 0.33 14.26 0.745

[0083] The process of determining the values ​​of the parameters in Table 3 is to take H atom as an example. The process of determining the parameters C and c is as follows: the electron distribution state of about 1000 energetic compounds in the Cambridge University CCDC crystal database in an isolated state is calculated by first principles, and the distance between atoms in these compounds is less than The spatial volume occupied by H atoms in a three-atom system containing H atoms varies with the interatomic distance between two atoms. The pre-exponential factor of the exponential function is obtained by statistical fitting as C and the exponential factor is c.

[0084] The process of determining D is to calculate the electron distribution state of about 1000 energetic compounds in the above crystal database in an isolated state by first principle, and to examine the distance between atoms in these compounds less than The trend of the spatial volume occupied by H atoms in a three-atom system containing H atoms changing with the interatomic distance between each other. Under the premise that the values ​​of the pre-exponential factor C and the exponential factor c of the exponential function are known, an intercept is further introduced to increase the function description accuracy. When the function obtains the highest fitting accuracy, the intercept value is D.

[0085] R cut3 The determination process is: first-principles calculation of the electron distribution state of about 1000 energetic compounds in the above crystal database in an isolated state, and the determination ... an isolated state, and the determination of the electron distribution state of about 1000 energetic compounds in an isolated state, and the determination of the electron distribution state of about 10 The atomic radius of the H atom in the three-atom system containing the H atom, and considering the probability of H atoms appearing in different structures and functional groups, the weighted average of the spatial volume occupied by the H atom is obtained as R cut3 .

[0086] In a specific solution, the organic compounds or energetic compounds used in constructing the model of the present invention can be selected from compounds that have been synthesized and publicly reported in the prior art. The specific implementation process can be achieved with the help of existing relevant software.

[0087] The following are specific examples of the present invention. It should be noted that the following examples are preferred examples of the present invention. Those skilled in the art can optimize the features, formulas, parameters, fitting methods and other technical factors in the special data within the scope of the present invention to obtain other implementation plans within the scope of the present invention. The present invention is not limited to the following examples. The methods are conventional methods unless otherwise specified.

[0088] Embodiment 1:

[0089] This embodiment is a method for constructing a theoretical prediction model of crystal density of a specific compound. The energetic compounds used in constructing the model are selected from about 1,000 energetic compounds that have been synthesized and publicly reported in the prior art, and the types of elements contained are limited to carbon, hydrogen, nitrogen, and oxygen. The compound types include nitrates, ammonium nitrates, cubanes, azoles, azides, etc. The crystal density of the energetic compounds involved ranges from 1.40 to 2.05 g / cm 3 The crystal density is measured by X-ray diffraction (XRD) and other methods. The specific method is as follows:

[0090] The three-dimensional Cartesian coordinates of each energetic compound are obtained by processing the SMILES code of the energetic compound using the Rdkit program to obtain the initial three-dimensional structure, and then optimizing the initial three-dimensional structure using the GFN2-xTB method;

[0091] Then, the descriptor data vector of each molecule is calculated, and the values ​​of each parameter in the relevant formula adopt the values ​​of the relevant parameters in the above exemplary description of C, H, N, and O;

[0092] The crystal density and descriptor data vector of the compound are fitted by the kernel ridge regression algorithm; the machine learning software used in the construction is the open source program scikit-learn. According to the specific settings of the scikit-learn program, the construction of the kernel ridge regression algorithm model requires setting six parameters: kernel, alpha, gamma, degree, coef0 and kernel_params. In this embodiment, kernel = 'rbf', alpha and gamma values ​​are obtained by grid search according to the content of the training set, degree = 3.0, coef0 = 1.0, kernel_params = 'None';

[0093] The model validation method was 5-fold cross validation, and the training set was the crystal density greater than 1.40 g / cm in the Cambridge CCDC crystal database. 3 , 1200 energetic compounds with theoretical detonation velocity greater than 6000m / s; the validation set is the crystal density greater than 1.40g / cm in the Cambridge University CCDC crystal database3 , 300 energetic compounds with theoretical detonation velocity greater than 6000m / s; the average absolute error of the training set is 0.0251g / cm 3 The mean absolute error of the validation set is 0.0259 g / cm 3 The average absolute prediction error of the crystal density of various energetic molecules is 0.026 g / cm 3 level.

[0094] Embodiment 2:

[0095] In this example, the energetic material TNT is selected as the target molecule, and the crystal density measured by XRD experiment is 1.654 g / cm 3 The model constructed in Example 1 is further used to predict the crystal density of TNT, and the method is as follows:

[0096] The SMILES code of the TNT molecule is: "CC1=C(C=C(C=C1[N+](=O)[O-])[N+](=O)[O-])[N+](=O)[O-]", and the SMILES code is converted into a three-dimensional molecular structure using the EmbedMolecule module in the RDKit software, and the conversion result is saved in an XYZ format file;

[0097] The TNT molecule was analyzed by the gfn2 method of the xtb software. Figure 1 The initial three-dimensional structure shown in the figure was optimized, and the optimized results are shown in Figure 2 As shown, the optimization result is saved in an XYZ format file so that the new structure data covers the original structure data;

[0098] According to the optimized structure of TNT molecule recorded in the XYZ format file, the four first-order eigenvalues ​​of TNT molecule are calculated as follows: ['H': 36.530, 'C': 82.474, 'N': 36.630, 'O': 66.186];

[0099] According to the optimized structure of TNT molecule recorded in the XYZ format file, the four first-order contribution eigenvalues ​​of TNT molecule are calculated by formula (2): ['H': 19.950, 'C': 90.167, 'N': 33.660, 'O': 72.306];

[0100] According to the optimized structure of TNT molecule recorded in the XYZ format file, the 10 second-order eigenvalues ​​of TNT molecule are calculated by equations (1) and (2): ['HH': 1.367, 'HC': 57.709, 'HN': 0.450, 'HO': 1.309, 'CC': 98.483, 'CN': 39.983, 'CO': 9.500, 'NN': 1.36*10-4 ,'NO':100.914,'OO':5.902]; 10 second-order contribution eigenvalues ​​are: ['HH':0.993,'HC':9.533,'HN':0.241,'HO':0.689,'CC':-19.981,'CN':-2.705,'CO':2.026,'NN':6.68*10 -5 ,'NO':-6.878,'OO':1.400];

[0101] According to the optimized structure of TNT molecule recorded in the XYZ format file, the 20 third-order eigenvalues ​​of TNT molecule are calculated by equations (3) and (4): ['HHH': 0.0507, 'HHC': 2.605, 'HHN': 0.00679, 'HHO': 0.0310, 'HCC': 7.586, 'HCN': 1.103, 'HCO': 1.197, 'HNN': 2.37*10 -4 ,'HNO':0.532,'HOO':0.0256,'CCC':13.588,'CCN':8.128,'CCO':2.662,'CNN':0.00454,'CNO':12.967,'COO':0.892,'NNN':6.71*10 -8 ,'NNO':1.87*10 -4 ,'NOO':8.289,'OOO':8.00*10 -6 ]; 20 third-order contribution eigenvalues ​​are: ['HHH':-0.0550,'HHC':-2.109,'HHN':-0.00585,'HHO':-0.0252,'HCC':-3.844,'HCN':-0.655,'HCO':-0.690,'HNN':1.49*10 -4 ,'HNO':-0.315,'HOO':-0.315,'CCC':-2.630,'CCN':-1.799,'CCO':-0.819,'CNN':-0.00189,'CNO':-2.673,'COO':-0.252,'NNN':-2.79*10 -8 ,'NNO':-8.92*10 -4 ,'NOO':-1.583,'OOO':-3.23*10 -6 ];

[0102] The descriptor data vector of the TNT molecule is: [36.530,82.474,36.630,66.186,19.950,90.167,33.660,72.306,1.367,57.709,0.450,1.309,98.483,39.983,9.500,1.36*10 -4 ,100.914,5.902,0.993,9.533,0.241,0.689,-19.981,-2.705,2.026,6.68*10 -5 ,-6.878,1.400,0.0507,2.605,'0.00679,0.0310,7.586,1.103,1.197,2.37*10 -4 ,0.532,0.0256,13.588,8.128,2.662,0.00454,12.967,0.892,6.71*10 -8 ,1.87*10 -4 ,8.289,8.00*10 -6 ,-0.0550,-2.109,-0.00585,-0.0252,-3.844,-0.655,:-0.690,:1.49*10 -4 ,-0.315,-0.315,-2.630,:-1.799,-0.819,-0.00189,-2.673,-0.252,

[0103] -2.79*10 -8 ,-8.92*10 -4 ,-1.583,-3.23*10 -6 ];

[0104] The Python script uses the descriptor vector as input and calls the machine learning model generated by the kernel ridge regression algorithm in Example 1 to predict the crystal density of the target molecule. The prediction result is: 1.668 g / cm 3 The absolute error between the predicted result and the experimental value is: |1.668-1.654|g / cm 3 =0.014g / cm 3 .

[0105] Embodiment 3:

[0106] This example uses energetic material CL-20 as the target molecule, and the crystal density measured by XRD experiment is 2.044 g / cm 3 The model constructed in Example 1 is further used to predict the crystal density of TNT, and the method is as follows:

[0107] The SMILES code of the CL-20 molecule is: "C12C3N(C4C(N3[N+](=O)[O-])N(C(N1[N+](=O)[O-])C(N2[N+](=O)[O-])N4[N+](=O)[O-])[N+](=O)[O-])[N+](=O)[O-]", and the SMILES code is converted into a three-dimensional molecular structure using the EmbedMolecule module in the RDKit software, and the conversion result is saved in an XYZ format file;

[0108] The CL-20 molecule was identified by the gfn2 method of the xtb software. Figure 3 The initial three-dimensional structure shown in FIG. 1 is optimized, and the optimized structure is as follows: Figure 4 As shown, the optimization result is saved in an XYZ format file so that the new structure data covers the original structure data;

[0109] According to the optimized structure of CL-20 molecule recorded in the XYZ format file, the four first-order eigenvalues ​​of CL-20 molecule are: ['H': 43.836, 'C': 70.692, 'N': 134.640, 'O': 132.372]; the four first-order contribution eigenvalues ​​are: ['H': 23.940, 'C': 77.286, 'N': 146.520, 'O': 144.612];

[0110] The 10 second-order eigenvalues ​​of CL-20 molecule are ['HH':0.122,'HC':67.056,'HN':8.073,'HO':3.856,'CC':37.287,'CN':166.722,'CO':4.058,'NN':85.384,'NO':221.046,'OO':11.662]; the 10 second-order contribution eigenvalues ​​are: ['HH':0.0939,'HC':10.485,'HN':3.877,'HO':1.964,'CC':-3.129,'CN':-13.292,'CO':1.096,'NN':1.592,'NO':-10.444,'OO':2.801];

[0111] The 20 third-order eigenvalues ​​of the optimized structure of CL-20 molecule are: ['HHH':1.00*10 -5,'HHC':0.354,'HHN':0.0404,'HHO':0.00488,'HCC':4.895,'HCN':15.352,'HCO':1.918,'HNN':2.299,'HNO':2.651,'HOO':0.104,'CCC':2.344,'CCN':27.680,'CCO':0.434,'CNN':31.475,'CNO':10.099,'COO':0.265,'NNN':1.754,NNO:25.895,'NOO':18.588,'OOO':0.00799]; 20 third-order contribution eigenvalues ​​are: ['HHH':- 1.196,'HHC':-0.319,'HHN':-0.0353,'HHO':-0.00421,'HCC':-2.534,'HC N':-8.002,'HCO':-1.096,'HNN':-1.312,'HNO':-1.505,'HOO':-0.0603,'C CC':-0.624,'CCN':-6.097,'CCO':-0.148,'CNN':-7.485,'CNO':-2.967,'C OO':-0.0845,'NNN':-0.590,NNO:-5.262,'NOO':-3.756,'OOO':-0.00283];

[0112] The descriptor data vector is: [43.836,70.692,134.640,132.372,23.940,77.286,146.520,144.612,0.122,67.056,8.073,3.856,37.287,166.722,4.058,85.384,221.046,11.662,0.0939,10.485,3.877,1.964,:-3.129,-13.292,1.096,1.592,-10.444,2.801,1.00*10 -5 ,0.354,0.0404,0.00488,4.895,15.352,1.918,2.299,2.651,0.104,2.344,27.680,0.434,31.475,10.099,0.265,1.754,25.895,18.588,0.00799]

[0113] The Python script uses the descriptor vector as input and calls the machine learning model generated by the kernel ridge regression algorithm in Example 1 to predict the crystal density of the target molecule. The prediction result is: 2.018 g / cm3 ; The absolute error between the predicted result and the experimental value is: |2.018-2.044|g / cm 3 =0.026g / cm 3 .

Claims

1. A method for constructing a compound crystal density prediction model, characterized in that: The method comprises the following steps: Step 1, constructing a data set, wherein the data set includes crystal structure data and descriptor data vectors of a plurality of organic compound samples; The crystal structure data include unit cell parameters of the corresponding compound and three-dimensional Cartesian coordinates of each atom in the unit cell; The descriptor data vector is composed of N first-order eigenvalues, N first-order contribution eigenvalues, M second-order eigenvalues, M second-order eigenvalue contributions, L third-order eigenvalues ​​and L third-order eigenvalue contributions; N is the number of atomic species in the corresponding compound, M is the number of diatomic combinations between different atoms in the N atoms and the sum of the number of diatomic species composed of various atoms themselves, M=C N 2 + N; L is the sum of the number of types of triatomic combinations between different atoms in N atoms and the number of triatomic types composed of various atoms themselves, L = C N 3 +2C N 2 +N; in: Any one element of the N first-order eigenvalues ​​is: the sum of the van der Waals volumes of all atoms belonging to the n-type atoms in the sequence of atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule state of the organic compound, and the n-type atoms represent any one of the N types of atoms; V i is the van der Waals volume occupied by the atom number i in the three-dimensional Cartesian coordinates of the isolated molecule of the organic compound; Any one element of the N first-order contribution eigenvalues ​​is: the sum of the van der Waals volumes of all atoms belonging to the n-type atoms in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the organic compound crystal structure state, and the n-type atoms represent any one of the N types of atoms; V i ' is the van der Waals volume occupied by the atom number i in the three-dimensional Cartesian coordinates of the organic compound crystal structure; Any one element of the M second-order eigenvalues ​​is: the sum of the van der Waals volumes of all diatomic combinations belonging to m types of diatomic combinations in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule state of the organic compound, and the m types of diatomic combinations represent any type of diatomic combinations among the M types of diatomic combinations; i and j are two atomic numbers in the sequence of atoms constituting the corresponding compound, i≠j; V ij is the van der Waals volume of the diatomic combination consisting of atoms numbered i and j in the three-dimensional Cartesian coordinates of an isolated molecule of an organic compound; A i and A j are the first pre-exponential factors of the diatomic combined volume calculation function of the atoms with sequence numbers i and j, respectively; B i and B j are the second pre-exponential factors of the diatomic combined volume calculation function of the atoms with serial numbers i and j, respectively; a i and a j are the first exponential factors of the diatomic combined volume calculation function of atoms numbered i and j respectively; b i and b j are the second exponential factors of the diatomic combined volume calculation function of the atoms with serial numbers i and j, respectively; α is the exponential factor of the diatomic combination volume calculation function; R ij is the geometric space distance between atom number i and atom number j; R cut2，i and R cut2，j are the diatomic combination cutoff radii of atoms numbered i and j respectively; Any one of the M second-order contribution eigenvalues ​​is: the sum of the van der Waals volumes of all diatomic combinations belonging to m types of diatomic combinations in the sequence of atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the organic compound crystal structure state, and the m types of diatomic combinations represent any type of diatomic combination among the M types of diatoms; V ij ' =V ij -V i '-V j ' (2) V ij ' is the van der Waals volume of the diatomic combination of atoms numbered i and j in the three-dimensional Cartesian coordinates of the organic compound crystal structure; V i ' is the van der Waals volume occupied by the atom number i in the three-dimensional Cartesian coordinates of the organic compound crystal structure; V j ' is the van der Waals volume occupied by the atom with serial number j in the three-dimensional Cartesian coordinates of the crystal structure of the organic compound; Any one of the L third-order eigenvalues ​​is: the sum of the van der Waals volumes of all the three-atom combinations belonging to the l-type three-atom combination in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule state of the organic compound, and the l-type three-atom combination is any one of the L types of three-atom combinations; i, j and k are the three atomic numbers of the sequential atoms constituting the corresponding compound, i≠j≠k; V ijk is the van der Waals volume of the three-atom combination consisting of atoms numbered i, j and k in the three-dimensional Cartesian coordinates of the isolated molecule state of the organic compound; C i , C j and C k Calculate the first pre-exponential factor for the volume of the three-atom combination of atom number i, atom number j and atom number k respectively; c i 、c j and c k Calculate the second pre-exponential factor for the volume of the three-atom combination of atom number i, atom number j and atom number k respectively; D i , D j and D k They are the intercept terms for calculating the volume of the three-atom combination of atom number i, atom number j and atom number k respectively; β is the exponential factor of the volume calculation function of the three-atom combination; R ij is the geometric distance between atoms numbered i and j; R ik is the geometric distance between atom number i and atom number k; R kj is the geometric space distance between atom number k and atom number j; R cut3，i , R cut3，j and R cut3，k are the three-atom combination cutoff radius of atom number i, the three-atom combination cutoff radius of atom number j, and the three-atom combination cutoff radius of atom number k respectively; Any one of the L third-order contribution eigenvalues ​​is: the sum of the van der Waals volumes of all the three-atom combinations belonging to the l-type three-atom combination in the sequence atoms constituting the corresponding compound in the three-dimensional Cartesian coordinates of the isolated molecule state of the organic compound, and the l-type three-atom combination is any one of the L types of three-atom combinations; V ijk '=V ijk -V ij ' -V ik ' -V jk ' (4) V ijk ' is the van der Waals volume occupied by the three-dimensional Cartesian coordinates of the three-atom combination of atoms numbered i, j and k in the crystal structure state of the organic compound; V ij ' is the van der Waals volume of the diatomic combination of atoms numbered i and j in the three-dimensional Cartesian coordinates of the organic compound crystal structure; V ik ' is the van der Waals volume of the diatomic combination of atoms numbered i and k in the three-dimensional Cartesian coordinates of the organic compound crystal structure; V jk ' is the van der Waals volume of the diatomic combination of atoms numbered k and j in the three-dimensional Cartesian coordinates of the organic compound crystal structure; Step 2, fitting the crystal density and descriptor data vectors of several organic compounds to obtain a prediction model for the crystal density of energetic compounds, where the crystal density is the dependent variable and the descriptor data is the independent variable.

2. The method for constructing a compound crystal density prediction model according to claim 1, characterized in that: The organic compound is an energetic compound.

3. The method for constructing a compound crystal density prediction model according to claim 1 or 2, characterized in that: The elements contained in the organic compound sample are any combination of two or more of the four elements of carbon, hydrogen, nitrogen and oxygen.

4. The method for constructing a compound crystal density prediction model according to claim 1, characterized in that: The crystal density of the organic compound samples ranges from 1.40 to 2.05 g / cm 3 .

5. The method for constructing a compound crystal density prediction model according to claim 1, characterized in that: The three-dimensional Cartesian coordinates of the organic compound are obtained by processing the SMILES code of the energetic compound using the Rdkit program to obtain an initial three-dimensional structure, and then optimizing the initial three-dimensional structure using the GFN2-xTB method to obtain the three-dimensional Cartesian coordinates.

6. The method for constructing a compound crystal density prediction model according to claim 1, characterized in that: The fitting in step 2 adopts a kernel ridge regression algorithm, and the kernel function in the kernel ridge regression algorithm parameters adopts a radial basis function.

7. The method for constructing a compound crystal density prediction model according to claim 1, characterized in that: The crystal structure data of the organic compound in step 1 is obtained by using an X-ray diffractometer measurement method, and the crystal density is calculated based on the crystal structure data as a standard value for model training.

8. A method for predicting compound crystal density, characterized in that: The method comprises: The method described in step 1 of claim 1 is used to obtain characteristic data of the organic compound to be predicted, and the characteristic data is used as input of the model constructed by the method described in claim 1 to calculate the crystal density of the organic compound to be predicted.

Images