Function construction method, device, equipment and medium

Abstract

The embodiments of the present application provide a function construction method, apparatus, device and medium, which obtain the total force value and position value of the atoms in the material structure; obtain the function value of the central atom symmetry function according to the position value and the fingerprint parameter value; construct a training set according to the total force value and the function value; train a machine learning model according to the training set until convergence; calculate the feature correlation value according to the converged machine learning model; determine the target fingerprint parameter value according to the feature correlation value; substitute the target fingerprint parameter value into the function to obtain the target function used to fit the machine learning force field; can use machine learning to automatically construct the central atom symmetry function of the material structure according to the data set, and automatically realize the fingerprint parameter optimization without relying on expert knowledge and manual operation; the machine learning model only focuses on the error of the training set and the results obtained by feature selection, and has the advantages of fast speed, high accuracy and low computational cost.

Description

Technical Field

[0001] The present application relates to the field of computing materials, and in particular to function construction methods, devices, equipment and media. Background Art

[0002] In the fields of computational chemistry and materials science, constructing high-precision potential energy surface functions is key to studying material phase transitions, chemical reaction path searches, global structure searches, and molecular dynamics simulations. The method of constructing potential energy surface functions through machine learning force fields is more efficient. Among them, the Behler-Parrinello (BP) method based on central atom symmetry functions is one of the most widely used methods. This method uses a series of central atom symmetry functions as structural descriptors and applies machine learning methods to establish the potential function of the atomic system. However, the BP method based on central atom symmetry functions requires certain expert knowledge, manual selection and construction of multiple distinctive structural descriptors to describe all possible local chemical environments, and has a certain dependence on expert knowledge and manual operation. Summary of the Invention

[0003] The purpose of this application is to solve one of the technical problems existing in the related art to at least a certain extent.

[0004] To this end, one purpose of the embodiments of the present application is to provide a function construction method, device, equipment and medium that can improve the construction efficiency of the central atom symmetry function.

[0005] To achieve the above objectives, an embodiment of the first aspect of the present application provides a function construction method, comprising:

[0006] Obtain the total force and position values of atoms in the material structure;

[0007] Obtaining a function value of the central atom symmetry function according to the position value and the fingerprint parameter value of the central atom symmetry function;

[0008] The total force value is used as a dependent variable and the function value is used as an independent variable to construct a training set;

[0009] Inputting the training set into a machine learning model for training until the machine learning model converges; calculating feature correlation values based on the converged machine learning model to obtain feature correlation values of the training set;

[0010] Performing feature selection on the training set according to the feature correlation value to determine a target fingerprinting parameter value;

[0011] The target fingerprint parameter value is substituted into the central atom symmetry function to obtain the target function used to fit the machine learning force field.

[0012] In certain embodiments of the first aspect of the present application, obtaining the total force value of atoms in the material structure includes:

[0013] Performing single-point calculations on the atoms of the material structure according to first principles to obtain the x-direction force component, the y-direction force component, and the z-direction force component of the atoms;

[0014] The total force value is obtained according to the x-direction force component value, the y-direction force component value and the z-direction force component value of the atom.

[0015] In certain embodiments of the first aspect of the present application, obtaining the function value of the central atom symmetry function according to the position value and the fingerprint parameter value of the central atom symmetry function includes:

[0016] Obtaining relative distances between the central atom and the adjacent atoms according to the position value of the central atom and the position values of the adjacent atoms adjacent to the central atom, wherein the central atom is any atom of the material structure;

[0017] Obtaining a truncation function value according to the relative distance value and a preset truncation radius;

[0018] Sampling the fingerprint parameter value from a preset value range of the fingerprint parameter of the central atom symmetry function;

[0019] The truncation function value and the fingerprint parameter value are substituted into the central atom symmetry function to obtain the function value.

[0020] In certain embodiments of the first aspect of the present application, the method of constructing a training set by using the total force value as a dependent variable and the function value as an independent variable includes:

[0021] For each atom, the total force value corresponding to the atom is used as a dependent variable, and the function value corresponding to the atom is used as an independent variable, and a training sample is composed of the dependent variable and the independent variable;

[0022] The training set is composed of a plurality of training samples of the atoms.

[0023] In certain embodiments of the first aspect of the present application, inputting the training set into a machine learning model for training until the machine learning model converges comprises:

[0024] Inputting the training set into a machine learning model for training to obtain a prediction result for the dependent variable;

[0025] Obtaining an error value between the dependent variable and the prediction result;

[0026] When the error value is less than a preset convergence threshold, the machine learning model converges.

[0027] In certain embodiments of the first aspect of the present application, inputting the training set into a machine learning model for training until the machine learning model converges further comprises:

[0028] When the error value is greater than or equal to the convergence threshold, adjusting parameters of the machine learning model;

[0029] The training set is input into the machine learning model after parameter adjustment for training until the machine learning model converges.

[0030] In certain embodiments of the first aspect of the present application, performing feature selection on the training set according to the feature correlation value to determine a target fingerprinting parameter value includes:

[0031] Sort the independent variables in the training set according to the feature correlation values, and select the independent variables within a preset sorting range as target independent variables;

[0032] The fingerprint parameter value corresponding to the target independent variable is determined as the target fingerprint parameter value.

[0033] An embodiment of the second aspect of the present application is a function construction device, comprising:

[0034] A first acquisition unit is used to obtain the total force value of atoms in the material structure;

[0035] A second acquiring unit is used to acquire the position values of atoms in the material structure;

[0036] a function value calculation unit, configured to obtain a function value of the central atom symmetry function according to the position value and a fingerprint parameter value of the central atom symmetry function;

[0037] a training set construction unit, configured to use the total force value as a dependent variable and the function value as an independent variable to construct a training set;

[0038] A model training unit, configured to input the training set into a machine learning model for training until the machine learning model converges;

[0039] a feature correlation value determining unit, configured to calculate feature correlation values based on the converged machine learning model to obtain feature correlation values of the training set;

[0040] a target fingerprinting parameter value determining unit, configured to perform feature selection on the training set according to the feature correlation value to determine a target fingerprinting parameter value;

[0041] An objective function determination unit is used to substitute the target fingerprint parameter value into the central atom symmetry function to obtain an objective function for fitting the machine learning force field.

[0042] An embodiment of the third aspect of the present application is an electronic device, comprising a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for implementing connection and communication between the processor and the memory, wherein the program, when executed by the processor, implements the function construction method described above.

[0043] An embodiment of the fourth aspect of the present application is a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to enable a computer to execute the function construction method as described above.

[0044] The function construction method, device, equipment and medium disclosed in the embodiments of the present application obtain the total force value and position value of the atoms in the material structure; obtain the function value of the central atom symmetry function according to the position value and the fingerprint parameter value of the central atom symmetry function; use the total force value as the dependent variable and the function value as the independent variable to construct a training set; input the training set into the machine learning model for training until the machine learning model converges; calculate the feature correlation value according to the converged machine learning model to obtain the feature correlation value of the training set; perform feature selection on the training set according to the feature correlation value to determine the target fingerprint parameter value; substitute the target fingerprint parameter value into the central atom symmetry function to obtain the target function used to fit the machine learning force field; can use machine learning to automatically construct the central atom symmetry function of the corresponding material structure according to the input material structure data set, and automatically realize the fingerprint parameter optimization of the central atom symmetry function without relying on expert knowledge and manual operation; the machine learning model only focuses on the error of the training set and the target fingerprint parameter value obtained by feature selection, so it has the advantages of fast speed, high accuracy and low computational cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following introduction is made to the drawings of the embodiments of the present application or the related technical solutions in the prior art. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative work.

[0046] Figure 1 This is a step diagram of the function construction method provided in the embodiment of the present application;

[0047] Figure 2 It is a sub-step diagram of step S100 of an embodiment of the present application;

[0048] Figure 3 It is a sub-step diagram of step S200 of an embodiment of the present application;

[0049] Figure 4 It is a sub-step diagram of step S300 of an embodiment of the present application;

[0050] Figure 5 It is a sub-step diagram of step S600 of an embodiment of the present application;

[0051] Figure 6 is a structural diagram of a function construction device provided in an embodiment of the present application;

[0052] Figure 7 is a structural diagram of an electronic device provided in an embodiment of the present application;

[0053] Figure 8 This is a schematic diagram of the characteristic importance analysis results of the central atom symmetry function of the Ge material structure;

[0054] Figure 9 This is a schematic diagram of the weighted importance analysis results of the central atom symmetry function of the Ge material structure;

[0055] Figure 10 This is a comparison chart of the prediction accuracy of the test set by the function construction method provided in the embodiment of the present application and the traditional traversal method. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0057] It should be noted that although the device schematics illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the device or the sequence in the flowcharts. The terms "first," "second," and the like in the specification, claims, or accompanying drawings are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.

[0058] The embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0059] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0060] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0061] The embodiments of the present application provide a function construction method, apparatus, device and medium, which obtain the total force value and position value of the atoms in the material structure; obtain the function value of the central atom symmetry function according to the position value and the fingerprint parameter value of the central atom symmetry function; use the total force value as the dependent variable and the function value as the independent variable to construct a training set; input the training set into the machine learning model for training until the machine learning model converges; calculate the feature correlation value according to the converged machine learning model to obtain the feature correlation value of the training set; perform feature selection on the training set according to the feature correlation value to determine the target fingerprint parameter value; substitute the target fingerprint parameter value into the central atom symmetry function to obtain the target function used to fit the machine learning force field; can use machine learning to automatically construct the central atom symmetry function of the corresponding material structure according to the input material structure data set, and automatically realize the fingerprint parameter optimization of the central atom symmetry function without relying on expert knowledge and manual operation; the machine learning model only focuses on the error of the training set and the target fingerprint parameter value obtained by feature selection, so it has the advantages of fast speed, high accuracy and low computational cost.

[0062] The embodiments of the present application are further described below with reference to the accompanying drawings.

[0063] The embodiment of the present application provides a function construction method, which includes but is not limited to steps S100, S200, S300, S400, S500, S600, and S700.

[0064] Reference Figure 1 , function construction methods include:

[0065] Step S100, obtaining the total force value and position value of atoms in the material structure;

[0066] Step S200, obtaining a function value of the central atom symmetry function according to the position value and the fingerprint parameter value of the central atom symmetry function;

[0067] Step S300 , using the total force value as the dependent variable and the function value as the independent variable to construct a training set;

[0068] Step S400: inputting the training set into the machine learning model for training until the machine learning model converges;

[0069] Step S500, calculating feature correlation values based on the converged machine learning model to obtain feature correlation values of the training set;

[0070] Step S600, performing feature selection on the training set according to the feature correlation value to determine the target fingerprint parameter value;

[0071] Step S700: Substitute the target fingerprint parameter value into the central atom symmetry function to obtain the target function used to fit the machine learning force field.

[0072] Regarding step S100 , on the one hand, according to the material system to be studied, data on the total force values and position values of atoms in the material structure can be directly obtained from an existing database to form a data set.

[0073] On the other hand, the total force and position values of the atoms in the material structure can be obtained by measuring and calculating the material structure.

[0074] According to the material system to be studied, a computer model of the material structure is selected. The computer model of the material structure is a model constructed on a computer based on the characteristics of the material structure.

[0075] Material results can include but are not limited to the following structures: crystal bulk structure of space group, surface model of high Miller index surface, surface model of low Miller index surface, cluster model and amorphous bulk and surface structure, subcluster model and amorphous bulk and surface structure, etc.

[0076] Reference Figure 2 , wherein obtaining the total force value of atoms in the material structure includes but is not limited to the following steps:

[0077] Step S110, performing single-point calculation on the atoms of the material structure according to first principles to obtain the x-direction force component value, the y-direction force component value, and the z-direction force component value of the atoms;

[0078] In step S120 , a total force value is obtained according to the x-direction force component value, the y-direction force component value, and the z-direction force component value of the atom.

[0079] For step S110, the first principle is a calculation from scratch, which does not require any experimental parameters. Only some basic physical constants are needed to obtain the principles of the basic properties of the ground state of the system. According to the principles of interaction between atomic nuclei and electrons and their basic laws of motion, the principles of quantum mechanics are applied, starting from specific requirements, and the algorithm that directly solves the Schrödinger equation after some approximate processing is the first principle algorithm. Molecules are composed of atoms, and atoms are composed of atomic nuclei and electrons. Quantum mechanics calculation is to calculate the molecular structure and molecular energy (or ions) based on the principle of interaction between atomic nuclei and electrons, and then the various properties of the substance can be calculated. According to the first principle, the force analysis of each atom in the material structure by single-point calculation can obtain the x-direction component force value, y-direction component force value and z-direction component force value of the atom.

[0080] In step S120, the total force value of the atom is obtained by synthesizing the x-direction force component value, the y-direction force component value, and the z-direction force component value of the atom. Specifically, the total force value of the atom can be expressed by the following formula: Among them, F x Indicates the x-direction force value of the atom; F y Indicates the y-direction force value of the atom; F z Represents the z-direction component of the atom.

[0081] By constructing a suitable coordinate system and placing the material structure into the coordinate system, the position value of each atom in the material structure can be obtained.

[0082] Reference Figure 3 For step S200, obtaining the function value of the central atom symmetry function according to the position value and the fingerprint parameter value of the central atom symmetry function includes but is not limited to the following steps:

[0083] Step S210, obtaining relative distance values between the central atom and the adjacent atoms according to the position value of the central atom and the position values of the adjacent atoms adjacent to the central atom;

[0084] Step S220, obtaining a truncation function value according to the relative distance value and a preset truncation radius;

[0085] Step S230, sampling from a preset value range of the fingerprint parameter of the central atom symmetry function to obtain a fingerprint parameter value;

[0086] Step S240: Substitute the truncation function value and the fingerprint parameter value into the central atom symmetry function to obtain the function value.

[0087] In this embodiment, the central atom symmetry function is divided into: a central atom symmetry function containing distance information and a central atom symmetry function containing angle information.

[0088] Specifically, the central atom symmetry function containing distance information can be expressed by the following formula: The central atom symmetry function containing angle information can be expressed as follows: Among them, η, R s , R c ,ζ,λ,θ s are all fingerprint parameters of the central atom symmetry function.

[0089] f c is the cutoff function. The cutoff function value can be calculated as follows. Take any atom as the central atom, and calculate the relative distance between the central atom and the adjacent atoms based on the position value of the central atom and the position value of the adjacent atoms. The relative distance between the two atoms is generally calculated based on the Euclidean distance. Then, the cutoff function value is obtained based on the relative distance value and the preset cutoff radius; for example, taking the i-th atom as the central atom, the cutoff function can be expressed by the following formula: Among them, R ij represents the relative distance between the i-th atom and the j-th neighboring atom, R c Represents the cutoff radius.

[0090] The fingerprint parameter value is sampled from the preset value range of the fingerprint parameter of the central atom symmetry function; for example, for the fingerprint parameter R s , fingerprint parameter R s The preset value range is From the fingerprint parameter R s The preset value range is sampled equidistantly to obtain the fingerprint parameter R s Multiple fingerprint parameter values.

[0091] Of course, in other embodiments, other sampling methods may be used to sample the fingerprint parameters, such as random sampling.

[0092] Substituting the constant form of the truncation function value and the fingerprint parameter value into the central atom symmetry function for calculation, the function value of the central atom symmetry function corresponding to the constant form of the truncation function value and the fingerprint parameter value can be obtained.

[0093] It can be understood that different function values can be obtained by calculating the truncation function values and the central atom symmetry function of the fingerprint parameter values with different numerical values.

[0094] In addition, each atom in the material structure needs to be calculated as a central atom to obtain the function value of each atom in the corresponding material structure.

[0095] Reference Figure 4For step S300, the total force value is used as the dependent variable and the function value is used as the independent variable to construct a training set, including but not limited to the following steps:

[0096] Step S310: For each atom, the total force value of the corresponding atom is used as the dependent variable, and the function value of the corresponding atom is used as the independent variable, and the dependent variable and the independent variable constitute a training sample;

[0097] Step S320: A training set is formed from a plurality of atomic training samples.

[0098] It can be understood that for a training sample of an atom, the training sample has a dependent variable, which is the total force value of the corresponding atom; and the training sample has multiple independent variables, which are the function values of the central atomic symmetry function corresponding to the atom, and these independent variables correspond to the dependent variables.

[0099] The training samples of all atoms of the material structure are combined into a data set, and a data set of a certain size is randomly sampled without replacement. Each row of data represents a sample of a central atom to construct a training set.

[0100] For step S400, a suitable machine learning model is selected and the parameters of the machine learning model are initialized.

[0101] Machine learning models include but are not limited to neural networks, random forests, gradient boosting regression trees, support vector machines, etc.

[0102] For example, when the machine learning model is a neural network algorithm, parameters such as the number of neurons, the number of neural network layers, and the weight of the neural network are initialized; when the machine learning model is a gradient boosting regression tree, parameters such as the number of trees, tree branches, and learning rate are initialized.

[0103] The training set is fed into the machine learning model after initialization to train the model and obtain predictions for the dependent variable. The error between the dependent variable and the prediction is then calculated. The machine learning model converges when the error is less than a preset convergence threshold. When the error is greater than or equal to the convergence threshold, the machine learning model parameters are adjusted and the training set is fed into the adjusted model to train the model until convergence.

[0104] Specifically, the error value is the root mean square error, which can be expressed by the following formula: Where N represents the number of training samples, Y i represents the i-th dependent variable, f(x i ) represents the prediction result corresponding to the i-th dependent variable.

[0105] The convergence criterion for model training is that the root mean square error (RMSE) of the training set is less than the convergence threshold.

[0106] With reference to the accuracy of force calculation based on the first principle, the convergence threshold can be set as Of course, in other embodiments, the convergence threshold may be other values, such as wait.

[0107] In addition, the parameters of the machine learning model can be adjusted by gradually increasing the parameter values of the machine learning model according to the principle of equidistant increments.

[0108] Step S500 calculates feature correlation values based on the converged machine learning model to obtain feature correlation values for the training set. It is understood that feature correlation values are related to the number of times a feature is used and the gain associated with its use; gain can be understood as the improvement in the loss function or a similar metric. Generally speaking, the more times a feature is split, the greater the gain it achieves after the split, and the more important the feature is.

[0109] Specifically, methods for calculating feature correlation values include, but are not limited to, principal component analysis, covariance analysis, P-value analysis, feature importance analysis, etc. That is, the feature correlation value can be the principal component corresponding to principal component analysis, the covariance corresponding to covariance analysis, the P-value corresponding to P-value analysis, or the feature importance corresponding to feature importance analysis.

[0110] Feature importance analysis also includes methods based on SHapley Additive exPlanation (SHAP), permutation, distillation learning, and machine learning interpretability. The distillation learning method trains a linear regression model to fit the predictions from the machine learning model, and then assesses the importance of the features by combining their standard deviations and the weights of the linear regression model. The machine learning interpretability method calculates the gradient of the input layer and assesses feature importance based on the gradient and the size of the eigenvalues.

[0111] Reference Figure 5 For step S600, feature selection is performed on the training set according to the feature correlation value to determine the target fingerprint parameter value, including but not limited to the following steps:

[0112] Step S610, sorting the independent variables in the training set according to the feature correlation values, and selecting the independent variables within a preset sorting range as target independent variables;

[0113] Step S620: Determine the fingerprint parameter value corresponding to the target independent variable as the target fingerprint parameter value.

[0114] In this embodiment, the independent variables of the training samples in the training set are sorted according to the feature correlation values. For example, a training sample in the training set has 5 independent variables, and these 5 independent variables are sorted from large to small according to the feature importance. In addition, weighted importance can be considered. The weighted importance is specifically the feature importance averaged over each hyperparameter.

[0115] The first or most pre-selected independent variables are selected as target independent variables. The function value of the target independent variable is obtained by substituting the fingerprint parameter value into the central atom symmetry function. The fingerprint parameter value corresponding to the function value of the target independent variable is determined as the target fingerprint parameter value. The target fingerprint parameter value is therefore the optimal fingerprint parameter value.

[0116] In step S700, the target fingerprint parameter value is substituted for the fingerprint parameter of the central atom symmetry function to obtain the objective function used to fit the machine learning force field, that is, the objective function is the central atom symmetry function with the value of the fingerprint parameter determined. For example, for the central atom symmetry function containing distance information η, R s 、R c The value of has been determined; for the central atom symmetry function containing angle information ζ, λ, θ s The value of has been determined.

[0117] The following further explains the central atom symmetry function of the germanium Ge material structure as an example.

[0118] Obtain a dataset of Ge material structures; collect various crystal structures and amorphous structures of Ge, including surface structures and cluster structures, totaling 5589; use first-principles software to perform single-point calculations on each structure, obtain the force value of each Ge atom in all structures, and then calculate the total force value; save the position value and total force value of each Ge atom in all structures.

[0119] The central atom symmetry function containing distance information is used for the Ge material structure. Fingerprinting parameters η and R s Sampling is performed to obtain fingerprint parameters η and R s The fingerprint parameter value is calculated based on the fingerprint parameters η and R s Fingerprint parameter value calculation Get the function value.

[0120] The total force value is used as the dependent variable and the function value is used as the independent variable. The sets of total force values and function values are sampled without replacement to obtain training samples of 1000 Ge atoms, and a training set of Ge atoms is created.

[0121] Initialize the model of the gradient boosted regression tree algorithm with 100 trees, 2 branches, 4 depths, and a learning rate of 0.05. Input the training set into the model of the gradient boosted regression tree algorithm for training.

[0122] When judging whether the trained gradient boosting regression tree algorithm model has reached the convergence standard, the root mean square error RMSE of the force on the central atom of the training set is calculated. If the root mean square error is less than the error threshold The model of the gradient boosting regression tree algorithm converges; otherwise, the model parameters are continuously adjusted, for example, the number of regression trees is increased by 50, and the training set is input into the model of the gradient boosting regression tree algorithm after the adjustment to train until the model of the gradient boosting regression tree algorithm converges.

[0123] Reference Figure 8 and Figure 9 , the feature importance of the training set is calculated based on the converged gradient boosting regression tree algorithm; and the average value of each R s The weighted importance of the ...

[0124] use Central atom symmetry function generated by equidistant sampling As a structural descriptor, it is used for subsequent machine learning force field training and prediction with excellent performance.

[0125] Reference Figure 10 Comparing the function construction method of this embodiment with the traditional traversal method, it can be seen that the traditional traversal method requires sequentially examining different fingerprinting parameters and then judging the quality of the results based on the performance of subsequent machine learning force field training, which is computationally expensive. The function construction method of this embodiment, based on the optimization method of automatically constructing the central atom symmetry function based on machine learning, can automatically, quickly, and efficiently obtain relatively good training and prediction results without the need for a posterior distribution.

[0126] In this embodiment, machine learning can be used to automatically construct the central atom symmetry function of the corresponding material structure based on the input material structure data set, and automatically realize the fingerprint parameter optimization of the central atom symmetry function without relying on expert knowledge and manual operation. It can be widely used in the pre-processing process of various machine learning force field development software packages based on the BP method.

[0127] Furthermore, the machine learning model used in this function construction method intelligently optimizes the parameters of the central atom symmetry function, avoiding the inefficiency of previous ergodic searches for these parameters and enabling efficient search for suitable central atom symmetry functions. Because the machine learning model focuses solely on the errors in the training set and the target fingerprint parameter values obtained through feature selection, it offers the advantages of high speed, high accuracy, and low computational cost.

[0128] Another embodiment of the present application provides a function construction device.

[0129] Reference Figure 6 The function construction device includes a first acquisition unit 110, a second acquisition unit 120, a function value calculation unit 130, a training set construction unit 140, a model training unit 150, a feature correlation value determination unit 160, a target fingerprinting parameter value determination unit 170 and a target function determination unit 180.

[0130] Among them, the first acquisition unit 110 is used to obtain the total force value of the atoms in the material structure; the second acquisition unit 120 is used to obtain the position value of the atoms in the material structure; the function value calculation unit 130 is used to obtain the function value of the central atom symmetry function based on the position value and the fingerprint parameter value of the central atom symmetry function; the training set construction unit 140 is used to use the total force value as the dependent variable and the function value as the independent variable to construct a training set; the model training unit 150 is used to input the training set into the machine learning model for training until the machine learning model converges; the feature correlation value determination unit 160 is used to calculate the feature correlation value according to the converged machine learning model to obtain the feature correlation value of the training set; the target fingerprint parameter value determination unit 170 is used to perform feature selection on the training set according to the feature correlation value to determine the target fingerprint parameter value; the objective function determination unit 180 is used to substitute the target fingerprint parameter value into the central atom symmetry function to obtain the objective function used to fit the machine learning force field.

[0131] In this embodiment, the function construction device can use machine learning to automatically construct the central atom symmetry function of the corresponding material structure based on the input material structure data set, and automatically realize the fingerprint parameter optimization of the central atom symmetry function without relying on expert knowledge and manual operation. It can be widely used in the pre-processing process of various machine learning force field development software packages based on the BP method.

[0132] It can be understood that the contents in the function construction method embodiment are applicable to the function construction device embodiment. The functions specifically implemented by the function construction device embodiment are the same as those in the function construction method embodiment, and the beneficial effects achieved are also the same as those achieved by the function construction method embodiment.

[0133] Another embodiment of the present application provides an electronic device. Figure 7 The electronic device includes a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory. When the program is executed by the processor, the function construction method is implemented, i.e., steps S100 to S700 are executed.

[0134] Memory 220, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer executable programs, such as the function construction in the above-mentioned embodiments of the present invention. Processor 210 implements the function construction in the above-mentioned embodiments of the present invention by running the non-transitory software programs and programs stored in memory 220.

[0135] The memory 220 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data required to execute the function construction in the above-mentioned embodiment of the present invention, etc. In addition, the memory 220 may include a high-speed random access memory 220, and may also include a non-volatile memory 220, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 220 may optionally include a memory 220 remotely arranged relative to the processor 210, and these remote memories 220 may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0136] Another embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the function construction method, namely, to execute steps S100 to S700.

[0137] Those skilled in the art will appreciate that all or some of the steps and systems in the method disclosed above can be implemented as software, firmware, hardware, and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include computer storage media (or non-transitory media) and communication media (or temporary media). As known to those skilled in the art, the term computer storage media is included in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data) and is volatile and non-volatile, removable, and non-removable. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory, or other memory technology, CD-ROM, digital versatile disks (DVD), or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage, or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media generally contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and may include any information delivery medium. In the above description of this specification, the reference terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" and the like are intended to mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0138] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and intent of the present application, and that the scope of the present application is defined by the claims and their equivalents.

[0139] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present application, and these equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A function construction method, characterized in that: include: Obtain the total force and position values of atoms in the material structure; Obtaining a function value of the central atom symmetry function according to the position value and the fingerprint parameter value of the central atom symmetry function; The total force value is used as a dependent variable and the function value is used as an independent variable to construct a training set; Inputting the training set into a machine learning model for training until the machine learning model converges; Calculating feature correlation values based on the converged machine learning model to obtain feature correlation values of the training set; Performing feature selection on the training set according to the feature correlation value to determine a target fingerprinting parameter value; The target fingerprint parameter value is substituted into the central atom symmetry function to obtain the target function used to fit the machine learning force field.

2. A function construction method according to claim 1, characterized in that: Get the total force value of the atoms in the material structure, including: Performing single-point calculations on the atoms of the material structure according to first principles to obtain the x-direction force component, the y-direction force component, and the z-direction force component of the atoms; The total force value is obtained according to the x-direction force component value, the y-direction force component value and the z-direction force component value of the atom.

3. A function construction method according to claim 1, characterized in that: Obtaining the function value of the central atom symmetry function according to the position value and the fingerprint parameter value of the central atom symmetry function includes: Obtaining relative distances between the central atom and the adjacent atoms according to the position value of the central atom and the position values of the adjacent atoms adjacent to the central atom, wherein the central atom is any atom of the material structure; Obtaining a truncation function value according to the relative distance value and a preset truncation radius; Sampling the fingerprint parameter value from a preset value range of the fingerprint parameter of the central atom symmetry function; The truncation function value and the fingerprint parameter value are substituted into the central atom symmetry function to obtain the function value.

4. A function construction method according to claim 1, characterized in that: The method of using the total force value as a dependent variable and the function value as an independent variable to construct a training set includes: For each atom, the total force value corresponding to the atom is used as a dependent variable, and the function value corresponding to the atom is used as an independent variable, and a training sample is composed of the dependent variable and the independent variable; The training set is composed of a plurality of training samples of the atoms.

5. A function construction method according to claim 1, characterized in that: Inputting the training set into the machine learning model for training until the machine learning model converges includes: Inputting the training set into a machine learning model for training to obtain a prediction result for the dependent variable; Obtaining an error value between the dependent variable and the prediction result; When the error value is less than a preset convergence threshold, the machine learning model converges.

6. A function construction method according to claim 5, characterized in that: Inputting the training set into the machine learning model for training until the machine learning model converges further includes: When the error value is greater than or equal to the convergence threshold, adjusting parameters of the machine learning model; The training set is input into the machine learning model after parameter adjustment for training until the machine learning model converges.

7. A function construction method according to claim 4, characterized in that: The performing feature selection on the training set according to the feature correlation value to determine the target fingerprinting parameter value includes: Sort the independent variables in the training set according to the feature correlation values, and select the independent variables within a preset sorting range as target independent variables; The fingerprint parameter value corresponding to the target independent variable is determined as the target fingerprint parameter value.

8. A function construction device, characterized in that: include: A first acquisition unit is used to obtain the total force value of atoms in the material structure; A second acquiring unit is used to acquire the position values of atoms in the material structure; a function value calculation unit, configured to obtain a function value of the central atom symmetry function according to the position value and a fingerprint parameter value of the central atom symmetry function; a training set construction unit, configured to use the total force value as a dependent variable and the function value as an independent variable to construct a training set; A model training unit, configured to input the training set into a machine learning model for training until the machine learning model converges; a feature correlation value determining unit, configured to calculate feature correlation values based on the converged machine learning model to obtain feature correlation values of the training set; a target fingerprinting parameter value determining unit, configured to perform feature selection on the training set according to the feature correlation value to determine a target fingerprinting parameter value; An objective function determination unit is used to substitute the target fingerprint parameter value into the central atom symmetry function to obtain an objective function for fitting the machine learning force field.

9. An electronic device, characterized in that: The electronic device includes a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory. When the program is executed by the processor, the function construction method according to any one of claims 1 to 7 is realized.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the function construction method according to any one of claims 1 to 7.

Images