Multi-metal single crystal purification optimization method based on learning model

Abstract

The invention provides a multi-metal single crystal purification optimization method based on a learning model, and relates to the technical field of data processing, and the method comprises the following steps: training according to multi-metal defect characteristics and impurity-metal repulsion force field parameters to obtain an adsorption energy prediction model; predicting the adsorption energy of the multi-metal impurities under different defect configurations by using the trained adsorption energy prediction model, and screening an optimal defect configuration according to an adsorption energy prediction result; based on the optimal defect configuration, simulating the adsorption process of each metal in the defect configuration, and analyzing the propagation and adsorption behaviors of each metal impurity in the pathway to obtain pathway simulation data; and path optimization is conducted on the path simulation data according to a purification target, a multi-metal purification scheme is obtained, and the technical problems that in the prior art, multi-metal single crystal purification efficiency is low, and impurity control precision is poor are solved. The technical effect of improving the multi-metal single crystal purification efficiency and precision by constructing the adsorption energy prediction model and combining the actual purification target is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a multi-metal single crystal purification optimization method based on a learning model. Background Art

[0002] With the widespread application of high-performance materials in high-end industries such as aerospace, microelectronics, new energy and precision manufacturing, multi-metal single crystal materials have become one of the key basic materials due to their excellent mechanical properties, electrical and thermal conductivity and chemical stability. However, multi-metal single crystal materials are easily mixed with a variety of trace impurities during the preparation process. These impurities can cause lattice distortion, decreased electron mobility, deterioration of interface performance and other problems, seriously affecting the functional performance and reliability of the materials. Traditional methods for purifying multi-metal materials, such as zone melting, electromigration purification, chemical treatment, etc., have the defects of low purification efficiency and difficulty in precise control.

[0003] The existing technology has technical problems of low purification efficiency of multi-metal single crystals and poor impurity control accuracy. Summary of the invention

[0004] The present application provides a multi-metal single crystal purification optimization method based on a learning model, which is used to solve the technical problems of low multi-metal single crystal purification efficiency and poor impurity control accuracy in the prior art.

[0005] In view of the above problems, the present application provides a multi-metal single crystal purification optimization method based on a learning model, the method comprising: training an adsorption energy prediction model according to the defect characteristics of the multi-metal and the impurity-metal repulsion field parameters; using the trained adsorption energy prediction model to predict the adsorption energy of the multi-metal impurities in different defect configurations, and screening the optimal defect configuration according to the adsorption energy prediction results; based on the optimal defect configuration, simulating the adsorption process of each metal in the defect configuration, analyzing the propagation and adsorption behavior of each metal impurity in the pathway, and obtaining pathway simulation data; optimizing the pathway simulation data according to the purification target to obtain a multi-metal purification plan.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The method provided in the embodiment of the present application is to train an adsorption energy prediction model based on the defect characteristics of multiple metals and the impurity-metal repulsion field parameters; use the trained adsorption energy prediction model to predict the adsorption energy of multiple metal impurities under different defect configurations, and select the optimal defect configuration according to the adsorption energy prediction results; based on the optimal defect configuration, simulate the adsorption process of each metal in the defect configuration, analyze the propagation and adsorption behavior of each metal impurity in the path, and obtain path simulation data; optimize the path simulation data according to the purification target to obtain a multi-metal purification plan. The technical effect of improving the purification efficiency and accuracy of multiple metal single crystals by constructing an adsorption energy prediction model and combining it with the actual purification target is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 A schematic diagram of the process flow of the multi-metal single crystal purification optimization method based on the learning model provided in this application;

[0010] Figure 2 A schematic diagram of the process of obtaining an adsorption energy prediction model in the multi-metal single crystal purification optimization method based on a learning model provided in this application. DETAILED DESCRIPTION

[0011] This application provides a multi-metal single crystal purification optimization method based on a learning model, which is used to solve the technical problems of low multi-metal single crystal purification efficiency and poor impurity control accuracy in the prior art. The technical effect of improving the purification efficiency and accuracy of multi-metal single crystals by constructing an adsorption energy prediction model and combining it with actual purification goals is achieved.

[0012] Below, the technical solutions in the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments described herein. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. It should also be noted that, for the convenience of description, only the parts related to the present invention are shown in the accompanying drawings, rather than all of them.

[0013] like Figure 1 As shown, the present application provides a multi-metal single crystal purification optimization method based on a learning model, the method comprising:

[0014] S100: Based on the defect characteristics of multiple metals and the impurity-metal repulsion field parameters, an adsorption energy prediction model is trained.

[0015] Specifically, a polymetallic single crystal is a long-range ordered single-phase crystal composed of two or more metal elements. Its atoms are arranged in a strict periodic pattern throughout the material. It has high anisotropy and excellent physical properties, such as high-temperature strength and electrical properties. It is widely used in aerospace engine blades, semiconductor substrates, quantum materials and other fields. Such as nickel-based alloys, niobium-titanium superconducting single crystals, semiconductor-doped single crystal silicon, etc. Polymetallic single crystal defects refer to the non-ideal state of atomic arrangement or chemical composition in materials composed of two or more metals, just like a puzzle with a few missing parts or crooked arrangement. The defects in the crystal material are the places where the atomic arrangement is imperfect. Common defect types include vacancies, dislocations, stacking faults, etc. Vacancies are point defects at the atomic scale, such as vacancies formed by the loss of Ni atoms in Ni-based single crystals. Dislocations refer to linear defects generated when the crystal grows or is subjected to force, and stacking faults are errors in the stacking order of atomic layers. In addition, defects in polymetallic materials can adsorb impurities and are natural impurity adsorption sites. Their adsorption capacity depends on the defect type and impurity characteristics, such as impurity size, charge, chemical affinity, etc. By regulating the defect type, the capture of target impurities can be enhanced, thereby improving the purification capacity of multi-metal single crystals. The impurity-metal repulsion field parameter refers to the repulsive force between metal atoms and impurity atoms due to the overlap of electron clouds. Through the repulsive force field, impurities can be pushed from the inside of the metal lattice to specific areas, such as defect channels. For example, in nickel-based alloys, the strong repulsive force between lead (Pb) and nickel (Ni) forces lead atoms to move to the grain boundaries.

[0016] Based on this, common defects in multiple groups of different multi-metal monomer systems, such as vacancies, dislocations, stacking faults and other structural information, are collected and extracted from literature databases or large databases, and defect features are extracted, including defect category, size, position, defect electron density, etc., as well as impurity-metal repulsive force field parameters. Impurity-metal repulsive force field parameters include the repulsive force field strength and range of action in different metal matrices. Defect features and impurity-metal repulsive force field parameters are used as training data. Based on machine learning methods, defect features and force field parameters are input for model training to obtain an adsorption energy prediction model. The optional models include extreme random forests, support vector machines and neural network models. Adsorption energy refers to the energy required or released by impurity atoms to be adsorbed at a defect site, usually expressed in electron volts (eV). The lower the adsorption energy (the larger the negative number), the more stable the adsorption. For example, if the adsorption energy of lead (Pb) at a vacancy is -1.2eV, it means that it particularly likes to stick to the vacancy and is more reliable than sticking to the normal position. By training, an adsorption energy prediction model is obtained, which can quickly and accurately predict the adsorption energy of a certain impurity in the environment. Then, by designing defects, the adsorption energy of the defects is very low, impurities are adsorbed, and the purification efficiency and accuracy of multi-metal single crystals are improved.

[0017] S200: using the trained adsorption energy prediction model to predict the adsorption energy of the multi-metal impurities under different defect configurations, and selecting the optimal defect configuration according to the adsorption energy prediction results.

[0018] Specifically, the trained adsorption energy prediction model can output the adsorption energy value under the combination by inputting defect characteristics and impurity-metal repulsion force field parameters. Then, the trained adsorption energy prediction model is used to batch predict multi-metal impurities under all alternative defect configurations, quickly obtain the adsorption energy of each impurity-defect combination, and then evaluate the adsorption capacity of each defect structure for different impurities, and screen out the optimal defect configuration in the purification process. For example, prepare a set of defect configuration data sets that may exist in multi-metals, each data record includes characteristics such as defect type, structure size, electron density, and corresponding force field parameters. For example, the following sample configurations are generated in a Ni-Cu alloy: Configuration A: vacancy (3.5Å), electron density 0.38e / ų, Configuration B: dislocation (4.2Å), electron density 0.40e / ų, then, for each impurity, such as Zn and Pb, it is combined with each defect configuration to form a sample pair, and input into the model for adsorption energy prediction. For example, the prediction results are as follows:

[0019]

[0020] Next, according to the screening strategy, the adsorption energy prediction results are used to select the configuration with the lowest (most negative) adsorption energy value, or according to comprehensive indicators such as the average adsorption capacity of multiple impurities and the total adsorption amount, the optimal defect configuration suitable for the purification path is obtained. The optimal defect configuration refers to the defect structure with the best adsorption capacity for multiple metal impurities. It is usually manifested as the configuration with the most negative adsorption energy or the highest comprehensive adsorption capacity score. For example, if Pb removal is the main focus, configuration A is selected as the optimal configuration, representing the site where impurities are most easily captured. If Zn needs to be taken into account, a mixed configuration optimization is used for comprehensive judgment. By predicting and screening the adsorption energy under different defect configurations, the defect configuration with the strongest adsorption capacity for specific impurities can be quickly screened out, providing a clear direction for subsequent adsorption pathway modeling and crystal structure regulation, improving the adaptability and intelligence level of the purification scheme, and thus improving the purification efficiency and quality, and reducing impurity residues in multi-metal single crystals.

[0021] S300: Based on the optimal defect configuration, simulate the adsorption process of each metal in the defect configuration, analyze the propagation and adsorption behavior of each metal impurity in the channel, and obtain channel simulation data.

[0022] Specifically, after determining the optimal defect configuration, the adsorption process of each metal in the defect configuration is simulated based on the optimal defect configuration, and the propagation path and adsorption behavior of each impurity in the crystal are analyzed. Adsorption process simulation refers to the use of computational methods to reconstruct the entire process of impurities entering the crystal from the material surface, propagating in the defect area and finally being adsorbed at the atomic scale. This process requires simulating the behavior of impurities when crossing different energy barriers, that is, energy barriers, in the lattice. Molecular dynamics simulation or transition state search algorithm can be used for dynamic path analysis. During the simulation process, the propagation trajectory, jump path, residence time and final adsorption position of each impurity atom in the defect area are recorded, and the adsorption energy change and migration energy barrier along the path are calculated, and the pathway simulation data are composed. The pathway simulation data includes but is not limited to: initial position, end position, path coordinate sequence, adsorption energy, residence point distribution, etc. Through simulation analysis, the diffusion trend and adsorption capacity differences of different impurities are revealed, which helps to identify the types of impurities that are preferentially removed or easily residual, and provides a data basis based on real behavior simulation for subsequent pathway optimization and structural regulation, which significantly improves the accuracy and purification reliability of the multi-metal single crystal purification path.

[0023] S400: Optimizing the pathway simulation data according to the purification target to obtain a multi-metal purification solution.

[0024] Furthermore, the purification target includes a single crystal structure parameter target and a purification threshold target.

[0025] Specifically, after completing the pathway simulation, in order to achieve effective purification of multi-metal impurities, it is necessary to further analyze and optimize the acquired pathway simulation data in combination with the established purification goals, so as to construct a multi-metal purification scheme that meets the expected performance requirements. First, clarify the purification goals. The purification goals include single crystal structure parameter goals and purification threshold goals. Among them, the single crystal structure parameter goals refer to the structural property requirements that the crystal should achieve after the final purification, such as crystal geometry, lattice stability, surface energy control, etc., which are key indicators for measuring structural quality. The purification threshold target refers to the target value to which the impurity concentration needs to be reduced. For example, the content of a certain metal impurity must be less than 0.01at.% (atomic percentage) or the maximum tolerable number of impurities per unit volume, which is the core standard for measuring purification efficiency and completion. Then, using the purification goals as constraints, the pathway simulation data is further analyzed and optimized to obtain a multi-metal purification scheme. For example, in the scenario of purifying lead (Pb) impurities in nickel metal, since lead particularly likes vacancies and has an adsorption energy of -1.2eV, the purification scheme is obtained by optimizing the propagation adsorption pathway, which is to bombard nickel with ions, create a bunch of vacancies by creating defects, and use adsorption energy to aggregate and adsorb lead impurities on the created vacancy defects to capture impurities. Finally, the impurities are discharged by heating the material and using the characteristics of the reduced Pb desorption energy barrier at high temperature to discharge lead from the vacancies, and then polishing to remove the surface enrichment layer to achieve the removal of lead impurities. By introducing the purification target quantification into the optimization process, the conversion of data to application scenarios is realized, and multi-metal purification strategies can be adaptively generated for different impurity types and concentration requirements to avoid blindly adjusting the structure, ensuring that each step of regulation serves the ultimate purification goal, reducing impurity residues, and improving purification efficiency and quality.

[0026] Further, such as Figure 2 As shown, S100 is described in which the adsorption energy prediction model is trained based on the defect characteristics of multiple metals and the impurity-metal repulsion field parameters, including: S110: collecting defect characteristic data of multiple metals and impurity-metal repulsion field parameter data, the defect characteristic data including defect category, size, position, defect electron density, and the impurity-metal repulsion field parameter data including force field strength and action distance; S120: performing adsorption characteristic correlation analysis based on the defect characteristic data and the impurity-metal repulsion field parameter data, and constructing a training data set; S130: using the training data set to perform model training convergence to obtain the adsorption energy prediction model.

[0027] Specifically, first, comprehensive training data is collected, including two types of core parameters closely related to adsorption behavior: one is the defect characteristic data of multiple metals, and the other is the repulsive force field parameter data between impurities and metals. Among them, the defect characteristic data include the type of defect (such as vacancies, dislocations, stacking faults, etc.), size (i.e., the defect influence range, the unit can be Å), spatial position (three-dimensional coordinates in the unit cell) and defect electron density (referring to the electron cloud distribution concentration in the area, the unit is usually e / ų), these characteristics directly determine the adsorption possibility and energy state of impurities at the defect. The impurity-metal repulsive force field parameters mainly measure the repulsive effect between impurity atoms and matrix atoms, specifically including the force field strength (i.e., the potential energy of the impurity repelled) and the action distance (the effective radius affected by the force field). The force field strength and action distance of the repulsive force field can be calculated by molecular dynamics simulation tools such as LAMMPS, and modeled using potential function models such as Lennard-Jones potential or Morse potential. Then, through statistical methods such as Pearson correlation coefficient, the defect feature data and impurity-metal repulsion field parameter data are analyzed for adsorption feature correlation, the relationship between defect attributes and impurity adsorption behavior is quantified, and a training data set is constructed. The training data set includes input feature defect type, impurity-metal repulsion field parameter data and label value, that is, the measured or calculated adsorption energy value, which can be obtained through first principle calculation software such as VASP, or extracted from the literature database. Then, a machine learning algorithm is used for model training. For example, an extreme random forest model is selected. During the training process, the data set is divided into a training set and a test set, such as an 8:2 ratio. Model fitting and hyperparameter adjustment are performed on the training set, and the model performance is evaluated on the test set to ensure its prediction accuracy and generalization ability. When the loss function of the model no longer decreases significantly with the increase in the number of iterations and reaches a stable state, it means that the model training converges and a trained adsorption energy prediction model is obtained. By constructing an adsorption energy prediction model, the speed and accuracy of adsorption energy prediction can be improved, providing a basis for subsequent multi-metal purification path simulation and optimization, and achieving fast and accurate purification effects.

[0028] Furthermore, an adsorption feature correlation analysis is performed based on the defect feature data and the impurity-metal repulsive force field parameter data to construct a training data set, including: performing adsorption characteristic analysis of metal impurities and defect configurations based on the defect feature data to determine the impurity-defect configuration adsorption energy; performing a fusion analysis based on the impurity-metal repulsive force field parameter data and the impurity-defect configuration adsorption energy to determine the synergistic adsorption coefficient of the repulsive force field parameters on the impurity-defect configuration adsorption energy; constructing a training data set based on the impurity-defect configuration adsorption energy and the synergistic adsorption coefficient of the repulsive force field parameters, wherein the data label in the training data set is adsorption energy.

[0029] Specifically, using first-principles software or molecular simulation tools, such as VASP and LAMMPS, based on defect feature data, the adsorption behaviors of different metal impurities in different defect configurations, such as adsorption tendency and adsorption energy change, are calculated and simulated, and which defect structures are more friendly or repulsive to impurities are identified to obtain the corresponding impurity-defect configuration adsorption energy. The impurity-defect configuration adsorption energy refers to the energy released when the impurity atom combines with a specific defect, and a negative value indicates spontaneous adsorption. Then, a fusion analysis of force field parameters and adsorption energy is performed to analyze the regulatory effect of the impurity atom on the adsorption behavior after being affected by the metal matrix repulsive force field. Specifically, a sample group of defect feature comparison is constructed, that is, under the same defect configuration, the impurity-metal force field strength or action distance is adjusted, and the adsorption energy change value is recorded respectively, and the adsorption energy difference (ΔE) is calculated. Based on this difference and the basic adsorption energy, the synergistic adsorption coefficient is obtained through normalization function or ratio calculation. And according to the synergistic adsorption coefficient of the impurity-defect configuration adsorption energy and the repulsive force field parameters, a training data set is constructed, and the data label in the training set is adsorption energy. By fusing and analyzing structural features with physical interaction data, the model training data set covers real defect structures and actual repulsive force field changes, thereby improving the model's generalization ability and prediction accuracy.

[0030] Furthermore, the adsorption characteristics of metal impurities and defect configurations are analyzed according to the defect characteristic data to determine the impurity-defect configuration adsorption energy, including: analyzing the adsorption probability of different metal impurities between different defect configurations based on the defect characteristic data; experimentally determining the measured adsorption energies of different metal impurities in different defect configurations; aligning the adsorption probability and measured adsorption energy according to metal impurities and defect configurations to determine the impurity-defect configuration adsorption energy.

[0031] Specifically, firstly, the collected defect characteristic data are used to analyze the adsorption tendency of various metal impurities in various defect configurations through first-principles calculations or molecular simulation tools. Among them, the defect characteristic data include the type, size, position and electron density of the defect. According to the defect characteristic data, the adsorption tendency of different metal impurities in different defect configurations, such as vacancies and dislocations, is analyzed to predict the impurity adsorption probability. The adsorption probability reflects the influence trend of structural factors on the adsorption behavior. For example, for copper impurities in aluminum matrix, it is found through calculation that the adsorption probability at vacancy defects is as high as 0.85, while the adsorption probability at the dislocation core is 0.32, indicating that vacancy defects are more likely to adsorb copper impurities. At this stage, high-throughput computing platforms, such as Materials Project, can be used to achieve batch simulation to ensure coverage of major defect configurations. Then, the actual adsorption energy of the above impurity and defect combinations is measured by experimental means. The measurement method can be obtained by first-principles calculations, such as VASP simulation of the energy difference before and after adsorption, surface science experiments, such as scanning tunneling microscope STM combined with XPS energy spectrum analysis. After the adsorption probability and adsorption energy are obtained, alignment and integration are performed, with metal impurities and defect configurations as the main keys, and the adsorption probability is paired with the corresponding measured adsorption energy to ensure that each training sample not only has a real physical adsorption energy label, but also retains the tendency information of its structure-induced adsorption behavior. In this way, the adsorption energy of each group of impurity-defect configurations is finally determined and used for subsequent model training. By combining experimental data with structural feature data to form high-quality data labels, accurate and reliable adsorption energy labels are provided for model training, which helps the model better learn the causal relationship between defect structure and adsorption energy, thereby improving the accuracy and generalization ability of the entire adsorption energy prediction model, improving the accuracy and reliability of adsorption energy prediction, and thus improving the efficiency and accuracy of multi-metal single crystal purification optimization.

[0032] Furthermore, a fusion analysis is performed based on the impurity-metal repulsive force field parameter data and the impurity-defect configuration adsorption energy, including: constructing comparative defect feature data based on the impurity-metal repulsive force field parameter data, the comparative defect feature data including defect record data of the same metal impurity and the same defect configuration under different impurity-metal repulsive force field parameters; performing adsorption energy experimental measurement based on the comparative defect feature data to obtain the adsorption energy difference of the comparative defect feature data; and calculating the synergistic adsorption coefficient based on the adsorption energy difference and the corresponding impurity-defect configuration adsorption energy without a repulsive force field.

[0033] Specifically, based on the existing impurity-metal repulsion force field parameter data, a set of comparative defect feature data is constructed. The construction principle of the comparative defect feature data is: under the same impurity element and defect configuration conditions, different force field strength and action distance parameters are set to form multiple experimental groups to observe the effect of force field changes on adsorption energy. For example, for Zn impurities in the Ni-Cu alloy grain boundary defect configuration, three sets of force field parameters are set: (1) force field strength = 2.0eV, action distance = 2.0Å; (2) strength = 2.5eV, action distance = 2.2Å; (3) strength = 3.0eV, action distance = 2.5Å. The three groups constitute the comparative samples of the impurity-defect combination. Then, based on the comparative defect feature data, the adsorption energy is experimentally determined, that is, under each set of different force field parameter settings, the first principle or molecular simulation calculation is used to respectively determine the adsorption energy value of the impurity at the defect site, and then the adsorption energy difference of the comparative defect feature data under different force fields is obtained. For example, the adsorption energies measured in the above three settings are -0.74eV, -0.69eV, and -0.65eV, respectively. The adsorption energy difference can be calculated as follows: ΔE1=-0.69-(-0.74)=+0.05eV, ΔE2=-0.65-(-0.74)=+0.09eV, where -0.74eV is the basic adsorption energy of the impurity-defect combination without an external repulsive force field (or minimum force field). Finally, based on the above adsorption energy difference and the basic adsorption energy in the absence of a force field, the synergistic adsorption coefficient is calculated. The synergistic adsorption coefficient is used to quantify the intensity of the force field's regulation of the adsorption energy. The formula is: synergistic adsorption coefficient=ΔE / |Ebasic|, substituting the data into: synergistic adsorption coefficient 1=0.05 / 0.74≈0.068, synergistic adsorption coefficient 2=0.09 / 0.74≈0.122. The synergistic adsorption coefficient is a dimensionless indicator. The larger the value, the stronger the regulation of the force field on the adsorption behavior. The synergistic adsorption coefficient is used as one of the input features to train the adsorption energy prediction model. It not only enhances the multi-dimensional expression ability of the model training data, but also effectively makes up for the prediction blind spot caused by relying only on static structural features, so that the adsorption energy prediction model has the ability to sensitively respond to changes in complex physical force fields, thereby achieving more accurate and reliable predictions, and providing reliable data support for the subsequent purification and optimization of multi-metal single crystals.

[0034] Furthermore, the training data set is used to perform model training convergence to obtain the adsorption energy prediction model, including: segmenting the training data set according to the segmentation ratio to construct a training set and a test set; selecting an extreme random forest algorithm to train the adsorption energy prediction model, wherein the training set is used to train the adsorption energy prediction model and adjust the model hyperparameters, and the trained model is verified and optimized using the test set until the convergence target is reached, thereby determining the adsorption energy prediction model.

[0035] Specifically, after the training data set is constructed, the constructed training data set is segmented. Each sample data in the training data set contains multiple input features, such as defect type, defect size, defect electron density, impurity-metal repulsion field parameters, cooperative adsorption coefficient, etc., as well as a corresponding output label, namely adsorption energy. In order to ensure the generalization ability of the model on new data, the data set can be divided into a training set and a test set according to a certain ratio, such as a training set: test set ratio of 8:2. Among them, the training set is used for the learning process of the model, and the test set is used to verify the prediction performance of the model for unseen samples. The training set is used to fit the extreme random forest model. Extreme random forest is an integrated model based on decision trees. By constructing multiple trees and introducing more randomness in the feature and sample dimensions, the generalization ability of the model is improved and overfitting is prevented. During the training process, the key hyperparameters of the model are adjusted, such as the number of trees, the maximum tree depth, the minimum number of split samples, etc. Hyperparameter optimization can be achieved by grid search or random search. After training, the model is applied to the test set for prediction, and the predicted value is compared with the true adsorption energy label, and the model performance is evaluated using the mean square error. If the model does not reach the preset convergence target, that is, the prediction accuracy is not enough, then return to adjust the hyperparameters, continue training and optimization, until the model converges stably, and obtain the adsorption energy prediction model. For example, using a set of parameters in the test set, the input features are: defect type = vacancy, size = 4.0Å, electron density = 0.32e / ų, force field strength = 1.8eV, synergy coefficient = 0.45, the model predicts the adsorption energy of -0.82eV, and the error is only 0.02eV compared with the true value of -0.80eV, indicating that the model has good accuracy. By establishing a nonlinear mapping relationship from structure and force field characteristics to adsorption energy, the adsorption behavior of impurities and defect combinations can be quickly and accurately predicted, thereby improving the automation and intelligence level of the overall purification path design, and improving the purification efficiency and accuracy.

[0036] Furthermore, the optimal defect configuration is screened according to the adsorption energy prediction results, including: obtaining the adsorption energy prediction results of the multi-metal impurities under each defect configuration; with the goal of maximizing the adsorption amount of the multi-metal impurities, searching for defect configurations according to the adsorption energy prediction results to obtain the optimal defect configuration.

[0037] Specifically, after completing the prediction of the adsorption energy of multi-metal impurities under different defect configurations, the adsorption performance of each defect configuration is further analyzed and compared based on the prediction results to screen out the optimal defect configuration that meets the purification requirements. First, the prediction results of the adsorption energy of multi-metal impurities under various defect configurations are sorted out. Adsorption energy refers to the energy change of a certain impurity at a specific defect position, which is usually a negative value. The more negative the value, the stronger the adsorption. The prediction results come from the adsorption energy prediction model trained in the previous stage. By inputting various defect parameters and impurity-metal repulsion field parameters, the corresponding adsorption energy can be quickly output. Then, on the basis of obtaining multiple adsorption energy prediction results, all defect configurations are screened with the maximization of adsorption amount as the optimization goal. The adsorption amount refers to the comprehensive performance of the adsorption capacity of a certain defect configuration for multiple impurities, which is usually quantified in two ways: the sum of adsorption energy and the average adsorption energy of impurities. The sum of adsorption energy is the sum of the absolute values ​​of the adsorption energy of all impurities under this configuration. The larger the value, the stronger the adsorption capacity. The average adsorption energy of impurities refers to the calculation of the average adsorption energy of each impurity under this configuration, reflecting the universal adsorption of this configuration for impurities. In the actual screening process, the adsorption energy prediction table can be aggregated, sorted and screened with the help of the pandas library in Python. Through defect configuration search and screening, the defect structure that is most conducive to the adsorption of various impurities can be efficiently located, providing high-quality initial structure selection for subsequent adsorption path simulation and purification strategy optimization, achieving precise and directional purification, significantly enhancing the capture efficiency of target impurities, and improving the quality and efficiency of impurity purification.

[0038] Furthermore, the pathway simulation data is optimized according to the purification target, including: obtaining pathway simulation data of multiple metal impurities, including adsorption position, adsorption energy, and propagation migration path; taking maximization of the purification target as the evaluation value, searching for geometric shape, size, and surface property solutions for the single crystal based on the pathway simulation data, and optimizing the propagation adsorption pathway.

[0039] Specifically, the pathway simulation data of multi-metal impurities are first obtained, including adsorption position, adsorption energy and propagation migration path. The adsorption position refers to the final stop point or high adsorption probability area of ​​the metal impurity in the crystal structure, which is usually the coordinate point near a specific defect. Adsorption energy is used to describe the energy change of impurities adsorbed at a specific position. The more negative the value, the stronger the adsorption. The propagation migration path refers to the path trajectory of impurities from the surface or internal distribution point into the structure, after diffusion, and finally arriving at the adsorption point, as well as the energy barriers, jump frequencies, etc. Then the purification target is maximized as the evaluation value. Based on the existing pathway simulation data, an adjustable parameter space is constructed. The geometric shape size and surface properties of the single crystal are used as optimization variables. A multi-objective optimization algorithm is used to continuously search and evaluate the contribution of different structural combinations to the purification target during the iteration process, so as to achieve the overall optimization of the propagation adsorption pathway and obtain a multi-metal purification scheme with excellent performance and stable structure. By optimizing the propagation adsorption pathway, the purification efficiency and directional control ability can be improved, so that impurities can be gathered in the controllable area faster and more concentratedly, thereby greatly improving the efficiency and controllability of metal purification.

[0040] Furthermore, the propagation adsorption pathway is optimized, including: performing a quantitative analysis of multi-impurity characteristics based on the pathway simulation data of the multi-metal impurities to obtain a quantitative list of multi-impurity adsorption; performing an external field compensation gradient analysis based on the multi-impurity adsorption quantitative list, performing multi-level division according to the compensation gradient, and constructing a multi-level defect compensation network, wherein the multi-level defect compensation network includes multi-level defect capture parameters; obtaining dynamic external field collaborative control parameters; based on the dynamic external field collaborative control parameters, performing parameter control on the multi-level defect compensation network layer by layer to obtain an external field collaborative control strategy; performing target adjustment on the evaluation value according to the external field collaborative control strategy; based on the adjustment target, searching for geometric shape, size, and surface property solutions for the single crystal to optimize the propagation adsorption pathway.

[0041] Specifically, based on the pathway simulation data, the adsorption behavior of various metal impurities is quantitatively analyzed, and the diffusion, adsorption and other behaviors of different metal impurities in the crystal are quantitatively described to form a comparable and sortable index system, including key features such as adsorption energy, migration path length, energy barrier height, residence time, etc., to form a multi-impurity adsorption quantitative list. The multi-impurity adsorption quantitative list is a sorted output table of multi-impurity behavior characteristics, which records the adsorption capacity and response characteristics of each impurity under specific structures or external field conditions, and is used to describe the adsorption difficulty and behavior trend of each impurity in different defect areas. According to the adsorption quantitative data in the multi-impurity adsorption quantitative list, the external field compensation gradient analysis is performed, that is, the degree of change of the adsorption efficiency of different regions under the conditions of no external field and external field is evaluated, so as to divide multiple gradient level regions, such as high response area, medium response area and low response area. According to the gradient division results, different control priorities are assigned respectively, and a multi-level defect compensation network is constructed. The multi-level defect compensation network is a structure-response mapping model that organizes regions with different response levels in the crystal into a multi-layer structure. Each level contains corresponding capture defect parameters, such as target adsorption energy range, path energy barrier threshold, structural coupling parameters, etc. The multi-level defect compensation network includes multi-level capture defect parameters, that is, the external field control conditions and response thresholds adapted to different regions. On this basis, the dynamic external field coordinated control parameters suitable for each region are designed and obtained. The dynamic external field coordinated control parameters refer to dynamically adjustable physical parameters, such as magnetic field strength, electric field frequency, electric field direction, etc. The magnetic field strength adjustment range is 0.5T to 2T, and the application direction is along the crystal growth direction. The electric field frequency adjustment range is 10Hz to 100Hz, and the application direction is perpendicular to the crystal surface. The external field parameters are dynamically adjusted according to the migration path and adsorption behavior of impurities. The goal is to flexibly configure in different regions to achieve the optimal impurity guiding effect. For example, the first layer preferentially applies an axial static magnetic field (such as 2.0±0.3 T) to enhance impurity capture; the second layer simultaneously applies a rotating magnetic field (1.8 T, 500 rpm) and a low-frequency electric field (5 kV / cm, 10kHz) to promote the diffusion of impurities to the target defect area; the third layer uses a high-frequency pulse electric field (12 kV / cm, 100 kHz) to suppress the re-dissolution of impurities and ensure the continuity of purification. Then, based on the dynamic external field collaborative control parameters, the multi-level defect compensation network is parameter-controlled layer by layer, the external field is applied in stages, and the migration path and adsorption behavior of impurities are optimized step by step to form an external field collaborative control strategy. The external field collaborative control strategy uses a variety of external physical fields to adjust the impurity behavior inside the crystal in layers to achieve the goal of multi-impurity purification. This intelligent control method can achieve precise intervention in impurity propagation paths, adsorption efficiency, back-expansion risks, etc.Finally, the evaluation value is updated and the target is adjusted according to the external field coordinated control strategy, and the geometry, size, and surface properties of the single crystal are searched and optimized based on the adjusted target. For example, the geometry, size, and surface properties of the single crystal are searched and optimized using a genetic algorithm. The genetic algorithm steps include: Initializing the population: generating a set of random single crystal geometry, size, and surface property parameters. Fitness function: using the adjusted target as the fitness function to evaluate the fitness of each individual. Selection operation: selecting individuals according to fitness, and individuals with high fitness have a higher probability of being selected. Crossover and mutation: performing crossover and mutation operations on the selected individuals to generate a new population. Iterative optimization: repeating the selection, crossover, and mutation operations until the preset number of iterations is reached or the fitness converges. Through optimization search, the impurity guidance and adsorption efficiency are further improved, and the optimal solution for the overall propagation adsorption pathway is obtained. Through the combination of data quantification, external field response, and hierarchical control, multi-dimensional factors are fully integrated to further improve the adaptability and accuracy of the purification strategy, enhance the adsorption of impurities and metals, and improve the purification effect.

[0042] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0043] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A multi-metal single crystal purification optimization method based on a learning model, characterized in that: include: According to the defect characteristics of multi-metal and the impurity-metal repulsion field parameters, the adsorption energy prediction model is trained and obtained; The adsorption energy of multi-metal impurities under different defect configurations is predicted using the trained adsorption energy prediction model, and the optimal defect configuration is selected according to the adsorption energy prediction results; Based on the optimal defect configuration, the adsorption process of each metal in the defect configuration is simulated, the propagation and adsorption behavior of each metal impurity in the pathway is analyzed, and pathway simulation data is obtained; The pathway simulation data is optimized according to the purification target to obtain a multi-metal purification solution.

2. The multi-metal single crystal purification optimization method based on the learning model according to claim 1 is characterized in that: The method of training and obtaining an adsorption energy prediction model based on the defect characteristics of multiple metals and the impurity-metal repulsion field parameters includes: Collect defect feature data of multiple metals and impurity-metal repulsion field parameter data, wherein the defect feature data includes defect type, size, position, and defect electron density, and the impurity-metal repulsion field parameter data includes field strength and action distance; Perform adsorption feature correlation analysis based on the defect feature data and impurity-metal repulsion field parameter data to construct a training data set; The training data set is used to perform model training convergence to obtain the adsorption energy prediction model.

3. The multi-metal single crystal purification optimization method based on the learning model according to claim 2 is characterized in that: According to the defect feature data and the impurity-metal repulsion field parameter data, adsorption feature correlation analysis is performed to construct a training data set, including: Analyzing the adsorption characteristics of metal impurities and defect configurations according to the defect characteristic data to determine the impurity-defect configuration adsorption energy; According to the impurity-metal repulsive force field parameter data and the impurity-defect configuration adsorption energy, a synergistic adsorption coefficient of the repulsive force field parameter on the impurity-defect configuration adsorption energy is determined; A training data set is constructed according to the impurity-defect configuration adsorption energy and the synergistic adsorption coefficient of the repulsive force field parameters, wherein the data label in the training data set is the adsorption energy.

4. The multi-metal single crystal purification optimization method based on the learning model according to claim 3 is characterized in that: The adsorption characteristics of metal impurities and defect configurations are analyzed according to the defect characteristic data to determine the impurity-defect configuration adsorption energy, including: Based on the defect characteristic data, analyzing the adsorption probability of different metal impurities between different defect configurations; The adsorption energies of different metal impurities in different defect configurations were experimentally determined; The adsorption probability is performed according to the metal impurities and defect configurations, and the adsorption energy is aligned to determine the impurity-defect configuration adsorption energy.

5. The multi-metal single crystal purification optimization method based on the learning model according to claim 4 is characterized in that: A fusion analysis is performed based on the impurity-metal repulsion field parameter data and the impurity-defect configuration adsorption energy, including: According to the impurity-metal repulsion field parameter data, constructing comparison defect feature data, the comparison defect feature data including defect record data of the same metal impurity and the same defect configuration under different impurity-metal repulsion field parameters; According to the compared defect characteristic data, an adsorption energy experimental measurement is performed to obtain an adsorption energy difference of the compared defect characteristic data; The synergistic adsorption coefficient is calculated based on the adsorption energy difference and the impurity-defect configuration adsorption energy under the corresponding non-repulsive force field.

6. The multi-metal single crystal purification optimization method based on learning model according to claim 3, characterized in that: The training data set is used to perform model training convergence to obtain the adsorption energy prediction model, including: Splitting the training data set according to the splitting ratio to construct a training set and a test set; The extreme random forest algorithm is selected to train the adsorption energy prediction model, wherein the adsorption energy prediction model is trained and the model hyperparameters are adjusted using the training set, and the trained model is verified and optimized using the test set until the convergence target is reached to determine the adsorption energy prediction model.

7. The multi-metal single crystal purification optimization method based on learning model according to claim 1, characterized in that: Screen the optimal defect configuration based on the adsorption energy prediction results, including: Obtain the adsorption energy prediction results of multi-metal impurities under various defect configurations; With the goal of maximizing the adsorption amount of the multi-metal impurities, a defect configuration search is performed based on the adsorption energy prediction result to obtain the optimal defect configuration.

8. The multi-metal single crystal purification optimization method based on learning model according to claim 1, characterized in that: Optimizing the pathway simulation data according to the purification target, including: Obtain pathway simulation data of multi-metal impurities, including adsorption position, adsorption energy, and propagation migration path; Taking the maximization of the purification target as the evaluation value, the geometric shape, size, and surface property scheme of the single crystal are searched based on the pathway simulation data to optimize the propagation adsorption pathway.

9. The multi-metal single crystal purification optimization method based on learning model according to claim 8, characterized in that: Optimize the propagation adsorption pathway, including: According to the channel simulation data of the multi-metal impurities, a multi-impurity feature quantitative analysis is performed to obtain a multi-impurity adsorption quantitative list; Based on the multi-impurity adsorption quantification list, an external field compensation gradient analysis is performed, and a multi-level division is performed according to the compensation gradient to construct a multi-level defect compensation network, wherein the multi-level defect compensation network includes multi-level capture defect parameters; Obtain dynamic external field coordinated control parameters; Based on the dynamic external field coordinated control parameters, the parameters of the multi-level defect compensation network are controlled layer by layer to obtain an external field coordinated control strategy; Performing target adjustment on the evaluation value according to the external field coordinated control strategy; Based on the adjustment target, the geometric shape, size and surface property solutions of the single crystal are searched to optimize the propagation adsorption path.

10. The multi-metal single crystal purification optimization method based on learning model according to claim 1, characterized in that: The purification targets include single crystal structure parameter targets and purification threshold targets.

Images