Method and system for predicting stability of oxynitride based on machine learning

Abstract

The invention discloses a machine learning-based oxynitride stability prediction method. The method comprises the following steps of S1, constructing a molecular structure of an oxynitride; s2, performing structure optimization on the molecular structure by using quantum chemistry software, and performing vibration analysis to obtain a result file; s3, chemical descriptors are extracted from the result file, a pre-established stability prediction model is input, a prediction result is output, and the stability prediction model is obtained by fusing a plurality of machine learning models. According to the method, the stability of the oxynitride can be predicted in a short time, blind synthesis is avoided, the research efficiency is improved, and experimental resources are saved.

Description

Technical Field

[0001] The present invention relates to the technical field related to molecular design, and more specifically, to a method and system for predicting the stability of nitrogen oxides based on machine learning. Background Art

[0002] Nitrogen and oxygen are the two most abundant elements in the atmosphere, accounting for approximately 75% and 23% of the total mass of the atmosphere, respectively. Despite their ubiquity, stable compounds formed from just these two elements are relatively rare. New stable nitrogen oxides have the potential to serve as green fertilizers to increase crop yields and food production, facilitate environmentally friendly manufacturing of explosives for construction and mining, and serve as high-performance oxidizers in propellants for deep space flight exploration. Once developed, these compounds could drive advances in agriculture, industry, and aerospace science, making a significant contribution to human progress.

[0003] Among oxidizers, compounds such as N2O, N2O4, and N2O5 are widely used as oxidizers in rocket propellants because they can not only withstand high temperatures before decomposition, but also release a large amount of energy when extracting electrons from the fuel. On the one hand, the high density of oxygen enhances the ability to extract electrons, increases the energy release during the exothermic reaction, and produces greater external work, which can enable a longer flight distance for a given mass. For example, compared with liquid oxygen (LOX, O2), the specific impulse (I sp ) doubled to 500-700 seconds. On the other hand, the addition of nitrogen exploits the significant energy difference between the single / double N-N bonds in the reactants and the triple N≡N bonds in the products, further increasing the energy density. Recently, the thermodynamic stability of dinitramide (DN, -N(NO2)2) has been significantly improved, with a decomposition temperature (T d ) is 243°C. The successful synthesis of the nitrogen-rich oxidizer trinitroamine (TNA, N(NO2)3) has achieved the highest density impulse ever, 33% higher than LOX. Therefore, new nitrogen oxides are very promising candidates to solve the contradiction between high stability and high energy density of propellants.

[0004] However, designing and constructing new stable nitroxides is a great challenge for current experimental and theoretical work. Experimentally, the intrinsic properties of nitrogen and oxygen facilitate their separation, while the retrosynthesis of nitroxides is extremely difficult. Compared with isolated nitrogen (N2) and oxygen (O2), nitroxides exhibit significantly higher energy levels due to induced strain, with NN single bond energy being 50% higher and OO bond energy being 41% higher, indicating thermodynamic difficulties. Achieving such retrosynthesis requires rigorous experimental conditions, such as high pressure conditions, to provide the activation energy required to overcome the 50-70 kJ / mol barrier, which is kinetically challenging. In theory, current computational algorithms for virtual compound generation exhibit obvious inheritance characteristics, but the scarcity of available nitroxide building blocks severely limits the diversity of combinations. Among a small number of virtually generated structures, the number of unstable compounds far exceeds that of stable compounds, which further increases the difficulty of discovering new stable nitroxides.

[0005] Therefore, it is necessary to design a technical solution that can predict the stability of unknown compounds, identify and capture potential chemical laws, and guide actual experiments. Summary of the invention

[0006] An object of the present invention is to provide a method and system for predicting the stability of nitrogen oxides based on machine learning, which can realize the stability prediction of nitrogen oxides in a relatively short time, avoid blind synthesis, improve research efficiency, and save experimental resources.

[0007] In order to achieve these objectives and other advantages of the present invention, according to one aspect of the present invention, the present invention provides a method for predicting the stability of nitrogen oxides based on machine learning, including: S1: constructing a molecular structure of a nitrogen oxide; S2: using quantum chemistry software to optimize the molecular structure and perform vibration analysis to obtain a result file; S3: extracting chemical descriptors from the result file, inputting a pre-established stability prediction model, and outputting a prediction result, wherein the stability prediction model is obtained by fusion of multiple machine learning models.

[0008] Furthermore, in S1, the molecular structure is constructed using GaussView software or Materials studio software.

[0009] Furthermore, in S2, Gaussian software is used to perform structural optimization and vibration analysis.

[0010] Furthermore, the chemical descriptors include HOMO-LUMO energy gap difference, Laplace bond order, dipole moment, and proportion of bonds connected to oxygen.

[0011] Furthermore, the plurality of machine learning models include at least a factorization machine, a support vector machine, a linear regression, a random forest, a naive Bayes and a gradient boosting tree.

[0012] Furthermore, the method for constructing the stability prediction model includes: establishing a training set, the training set including chemical descriptors and thermodynamic stability labels corresponding to the molecular structure; using the training set to train multiple machine learning models to obtain multiple initial prediction models; evaluating multiple initial prediction models, and selecting at least two machine learning models based on the verification results; linearly fusing at least two of the machine learning models to obtain a fusion model, and training the fusion model using the training set to obtain the stability prediction model.

[0013] Furthermore, before establishing the training set, correlation analysis is performed on the chemical descriptors corresponding to the molecular structures to determine chemical descriptors that are not correlated with each other.

[0014] Furthermore, a 5-fold cross validation method is used to train and evaluate the multiple initial prediction models.

[0015] According to another aspect of the present invention, a nitrogen oxide stability prediction system based on machine learning is also provided, including: a construction unit for constructing the molecular structure of nitrogen oxides; an optimization unit for optimizing the molecular structure using quantum chemistry software, and performing vibration analysis to obtain a result file; a prediction unit for extracting chemical descriptors from the result file, inputting a pre-established stability prediction model, and outputting a prediction result, wherein the stability prediction model is obtained by fusion of multiple machine learning models.

[0016] The present invention has at least the following beneficial effects: The present invention obtains multiple chemical descriptors that can describe the molecular structure, electronic and energy characteristics through quantum chemical calculations, and then combines them with machine learning models to achieve rapid and accurate prediction of the stability of nitrogen oxides, and then accurately screen out nitrogen oxides with higher structural stability, thereby avoiding blind design and synthesis, providing an effective guiding basis for the synthesis of nitrogen oxides, improving experimental efficiency, and saving experimental resources.

[0017] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of an embodiment of the present application; Figure 2This is a correlation matrix diagram and feature importance ranking diagram of the chemical descriptors of an embodiment of the present application, wherein energy gap, dipole moment, laplacian bond order, and proportion of O-connected bond represent the HOMO-LUMO energy gap difference, Laplace bond order, dipole moment, and proportion of oxygen-connected bonds, respectively.

[0019] Figure 3 This is an example of the prediction performance of all models provided for this application, where FM, SVM, LR, NB, GBDT, and RF represent factor decomposition machine, support vector machine, linear regression, naive Bayes, gradient boosting tree, and random forest, respectively, and Customized represents a stability prediction model. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0021] It should be understood that the terms such as "having", "including" and "comprising" used in the embodiments of the present application do not exclude the existence or addition of one or more other elements or their combinations. All directional indications (such as up, down, left, right, front, back...) in the embodiments of the present application are only used to explain the relative position relationship, movement, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly. When an element is referred to as "fixed on" or "set on" another element, it can be directly on the other element or there may be a centering element at the same time. When an element is referred to as "connecting" another element, it can be directly connected to another element or it can be indirectly connected to another element through a centering element. The description of "first", "second", etc. in the embodiments of the present application is only for descriptive purposes, and cannot be understood as indicating or implying its relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" can explicitly or implicitly include at least one of the features.

[0022] It should be noted that the technical solutions between the various embodiments of the present application can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0023] like Figure 1 As shown, the embodiment of the present application provides a method for predicting nitrogen oxide stability based on machine learning, including: S1: Construct the molecular structure of nitrogen oxides; Exemplarily, the molecular structure is constructed using GaussView software and Materials studio software, and software such as chem3d can also be used; Exemplarily, the following steps can be used for molecular structure design: Step 1: Design unit blocks - planar and cyclic fragments, including three-membered rings, four-membered rings, five-membered rings and six-membered rings. Step 2: Stack the unit blocks in sequence, connecting them directly and by bridging; Step 3: Decoration of the framework.

[0024] S2: Optimizing the molecular structure using quantum chemistry software and performing vibration analysis to obtain a result file; For example, Gaussian software is used for structural optimization and vibration analysis, and NWChem, VASP, Q-Chem and other software may also be used; Exemplarily, the result file (log file) output by Gaussian software includes chemical descriptors such as energy and vibration frequency, as well as HOMO-LUMO energy gap difference, Laplace bond order, dipole moment, proportion of bonds connected to oxygen, etc.; S3: extracting chemical descriptors from the result file, inputting a pre-established stability prediction model, and outputting prediction results, wherein the stability prediction model is obtained by integrating multiple machine learning models.

[0025] Exemplarily, the plurality of machine learning models include at least a factorization machine, a support vector machine, a linear regression, a random forest, a naive Bayes, and a gradient boosting tree; Exemplarily, the method for constructing the stability prediction model includes: establishing a training set, the training set including chemical descriptors and thermodynamic stability labels corresponding to the molecular structure; using the training set to train multiple machine learning models to obtain multiple initial prediction models; evaluating the multiple initial prediction models, and selecting at least two machine learning models according to the verification results; linearly fusing at least two machine learning models to obtain a fusion model, and training the fusion model using the training set to obtain the stability prediction model; using feature descriptors as input variables for machine learning; and using thermodynamic stability labels (1 or 0) as output variables for machine learning; Exemplarily, before establishing the training set, the chemical descriptors corresponding to the molecular structures are subjected to correlation analysis to determine the chemical descriptors that are not correlated with each other; the correlation between the feature descriptors is analyzed, and the degree of correlation between the two variables is measured by calculating the Pearson correlation coefficient between the descriptors. If the Pearson correlation coefficient between the two features is less than 0.6, it indicates that the correlation between the selected features is weak; the most important factor for regulating the thermodynamic stability of the oxidant is determined by feature importance analysis, and its correlation matrix is ​​as follows: Figure 2 As shown; through feature importance analysis, the most important factor regulating the thermodynamic stability of the oxidant was determined; the Pearson correlation coefficient calculation formula is as follows: In the above formula, x i ,y i Represent two different chemical descriptors, , represents the average of two different chemical descriptors; Exemplarily, a 5-fold cross validation method is used to train and evaluate the multiple initial prediction models; that is, the data set is divided into 5 parts, 4 of which are used as training sets and 1 as a test set in turn, the model is trained on the training set, and the model is evaluated on the test set; finally, the 5 results are averaged to obtain the final result, and compared according to their prediction performance (such as accuracy, recall rate, F1 value, etc.) on the given data set; based on the evaluation of the single model, the model with relatively good performance is selected and constructed into a fusion customized model; the fusion model can combine the advantages of each other to improve the overall prediction effect; for nitrogen oxides, the fusion model of factor decomposition machine, support vector machine and linear regression has the best effect; In order to further improve the prediction performance of the customized model, an adaptive weight optimization strategy is introduced. By dynamically adjusting the weight distribution of the model, the optimal weight combination is found and applied to the fusion model. The data set is trained to obtain the corresponding model parameters, and the customized model trained based on the existing data set is output. For example, the selected machine model is first given an initial weight, and then the weights are adjusted one by one. After the weights are adjusted, the fusion model is trained and the prediction results are evaluated (accuracy, recall rate and F1 value). According to the evaluation results, the optimal weight is determined. In summary, this embodiment obtains multiple chemical descriptors that can describe the molecular structure, electronic and energy characteristics through quantum chemical calculations, and then combines the machine learning model to construct a stability prediction model, which can achieve rapid and accurate prediction of the stability of nitrogen oxides, and then accurately screen out nitrogen oxides with higher structural stability, thereby avoiding blind design and synthesis, providing an effective guiding basis for the synthesis of nitrogen oxides, improving experimental efficiency, and saving experimental resources.

[0026] In another embodiment, 168 nitrogen oxide structures are screened from known nitrogen oxides, of which 120 are thermodynamically stable compounds and 48 are thermodynamically unstable. The molecular structure of the new structure is modeled. First, GaussView 6.0 and Materials studio 2019 are used to construct the initial 3D molecular structure; then, the B3LYP / 6-311++G(d,p) theoretical method in Gaussian16 software is used for geometry optimization and vibration analysis, and the CCSD(T) / jun-cc-pvtz theoretical method is used to calculate the energy of the optimized structure. Then, Multiwfn software and Python code are used to analyze and collect the output result files to obtain 4 relevant chemical descriptors, including HOMO-LUMO energy gap difference, Laplace bond order, dipole moment, and the proportion of bonds connected to oxygen.

[0027] The above characteristics are analyzed for correlation, such as Figure 2 As shown, the correlation coefficients between the features are all lower than 0.6, indicating that there is no obvious correlation between the features.

[0028] The above-mentioned characteristic variables are used as the input variables of machine learning, and the labels of thermodynamic stability (molecules with stable structures are 1 or molecules with unstable structures are 0) are used as the output variables of machine learning.

[0029] Center standardization is used to normalize all sample eigenvalues. The calculation formula of center standardization is as follows: Among them, x i,j is the normalized value of the 𝑗th feature in the 𝑖th sample, is the original value of the 𝑗th feature in the 𝑖th sample, μ j is the mean of the 𝑗th feature, σ j is the standard deviation of the 𝑗th feature.

[0030] FM (factorization machine), SVM (support vector machine), LR (linear regression), NB (naive Bayes), GBDT (gradient boosting tree), and RF (random forest) were selected for testing. The input of the model is the characteristic data of all samples (4 chemical descriptors), and the output is the label of the sample (thermodynamically stable structure is 1, and thermodynamically unstable structure is 0). First, hyperparameter tuning was performed through grid search, and the optimal parameter combination of the support vector machine was determined to be: L1 regularization coefficient 0.15; maximum number of iterations 300; the optimal parameters of the factorization machine are: learning rate 0.05, number of factors 2, L2 regularization coefficient 0.3, initialization standard deviation coefficient 0.05, number of training rounds 10; the optimal parameters of linear regression are L1 regularization coefficient 0.12; maximum number of iterations 300. Then, 5-fold cross validation is used for data partitioning, model training and evaluation. The entire data set is randomly divided into 5 folds of equal size. Four of them are used as training sets and one as test sets in turn. The model is trained on the training set, and the model performance is evaluated on the test set. The relevant indicators (accuracy, recall rate and F1 value) are calculated. For details, see Table 1 and Figure 3 .

[0031] Accuracy, recall and F1 value are used to quantitatively evaluate the model prediction performance, and the best prediction model is saved based on the model performance. The calculation formula is as follows: Table 1 Explanation of the parameters in the above formula Table 2 Evaluation results of each model In order to further improve the prediction performance of the model, a fusion customized model based on factor decomposition machine, support vector machine and linear regression was selected to obtain a stability prediction model. The weight distribution of the model was dynamically adjusted through an adaptive weight optimization strategy to find the optimal weight combination. It was found that the prediction effect was best when the weight of the factor decomposition machine was 0.4, the weight of the support vector machine was 0.3, and the weight of the linear regression was 0.3. Repeat the above process and average the accuracy, recall rate and F1 value obtained multiple times. In the end, the best performing stability prediction model had an accuracy of 0.93, an F1 value of 0.95, and a recall rate of 0.98, with good prediction performance.

[0032] The embodiment of the present application also provides a nitrogen oxide stability prediction system based on machine learning, including: a construction unit, used to construct a molecular structure of a nitrogen oxide; an optimization unit, used to optimize the molecular structure using quantum chemistry software, and perform vibration analysis to obtain a result file; a prediction unit, used to extract chemical descriptors from the result file, input a pre-established stability prediction model, and output a prediction result, wherein the stability prediction model is obtained by fusion of multiple machine learning models; The embodiments of the present application utilize computer programs to establish a construction unit, an optimization unit, and a prediction unit. The construction unit has a user operation interface for constructing the desired nitrogen oxides. The optimization unit calls quantum chemistry software for structural optimization to obtain a result file. The prediction unit inputs the chemical descriptors extracted from the result file into a pre-established stability prediction model to obtain a result on whether the constructed nitrogen oxides are stable, thereby providing an effective guiding basis for the synthesis of nitrogen oxides, avoiding blind design and synthesis, improving experimental efficiency, and saving experimental resources.

[0033] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A method for predicting the stability of nitrogen oxides based on machine learning, characterized in that: include: S1: Construct the molecular structure of nitrogen oxides; S2: Optimizing the molecular structure using quantum chemistry software and performing vibration analysis to obtain a result file; S3: extracting chemical descriptors from the result file, inputting a pre-established stability prediction model, and outputting prediction results, wherein the stability prediction model is obtained by integrating multiple machine learning models.

2. The method for predicting nitrogen oxide stability based on machine learning according to claim 1, characterized in that: In S1, the molecular structure is constructed using GaussView software or Materials studio software.

3. The method for predicting nitrogen oxide stability based on machine learning according to claim 1, characterized in that: In the S2, Gaussian software is used to perform structural optimization and vibration analysis.

4. The method for predicting nitrogen oxide stability based on machine learning according to claim 1, characterized in that: The chemical descriptors include HOMO-LUMO energy gap difference, Laplace bond order, dipole moment, and the proportion of bonds connected to oxygen.

5. The method for predicting nitrogen oxide stability based on machine learning as claimed in claim 1, characterized in that: The multiple machine learning models include at least factorization machine, support vector machine, linear regression, random forest, naive Bayes and gradient boosting tree.

6. The method for predicting nitrogen oxide stability based on machine learning according to claim 5, characterized in that: The method for constructing the stability prediction model comprises: Establishing a training set, wherein the training set includes chemical descriptors and thermodynamic stability labels corresponding to the molecular structures; Using the training set to train the plurality of machine learning models to obtain a plurality of initial prediction models; Evaluate the multiple initial prediction models, and select at least two of the machine learning models based on the verification results; At least two of the machine learning models are linearly fused to obtain a fusion model, and the fusion model is trained using the training set to obtain the stability prediction model.

7. The method for predicting nitrogen oxide stability based on machine learning according to claim 6, characterized in that: Before establishing the training set, correlation analysis is performed on the chemical descriptors corresponding to the molecular structures to determine chemical descriptors that are not correlated with each other.

8. The method for predicting nitrogen oxide stability based on machine learning as claimed in claim 6, characterized in that: The multiple initial prediction models are trained and evaluated using a 5-fold cross validation method.

9. A nitrogen oxide stability prediction system based on machine learning, characterized in that: include: Building blocks for constructing the molecular structure of nitrogen oxides; An optimization unit, used for optimizing the molecular structure using quantum chemistry software and performing vibration analysis to obtain a result file; A prediction unit is used to extract chemical descriptors from the result file, input a pre-established stability prediction model, and output a prediction result, wherein the stability prediction model is obtained by integrating multiple machine learning models.

Images