Determination method, application method and related device of high-throughput prediction model

Abstract

The invention discloses a determination method and application method of a high-throughput prediction model and a related device, and relates to the field of material performance prediction.The determination method of the high-throughput prediction model comprises the steps that Pearson screening is conducted on descriptors in a database; the database comprises a plurality of descriptors; the descriptors comprise inherent properties of elements, physicochemical properties of raw material precursor oxides and differentiated experimental conditions; based on an SHAP software package and the screened descriptors, determining an optimal algorithm-descriptor combination; and based on the optimal algorithm-descriptor combination, performing hyper-parameter adjustment on the optimal algorithm by adopting a grid search technology to obtain a high-throughput prediction model of the material performance. Through the model, a spinel material (Al < 2 > Cr < 0.5 > Zn < 0.1 > Mg < 0.2 > Mn < 0.2 >) O < 4 > with high breaking strength and a spinel material (Al < 2 > Fe < 0.25 > Zn < 0.25 > Mg < 0.25 > Mn < 0.25 >) O < 4 > with high hardness are found.

Description

Technical Field

[0001] The present invention relates to the field of material property prediction, and in particular to a determination method, an application method and a related device of a high-throughput prediction model. Background Art

[0002] Spinel refractory materials are mainly represented by magnesium-aluminum spinel (MgAl2O4), which is an important high-temperature structural material. They are prepared by high-temperature solid-phase reaction or electric melting to form refractory products with spinel structure. Spinel refractory materials have many advantages: high melting point enables them to work stably for a long time in high-temperature environment; excellent thermal shock resistance makes them not easy to crack in the environment of sudden temperature change; significant corrosion resistance to acids, alkalis and other chemicals; high mechanical strength and wear resistance enable them to withstand mechanical impact and wear at high temperature. These advantages make spinel refractory materials widely used in high-temperature industrial fields.

[0003] High entropy ceramics (HECs) are solid solutions of five or more cation or anion sublattices with high state entropy. They have excellent properties, including ultra-high melting point, high mechanical strength and strong corrosion resistance, and have attracted widespread attention. To date, a variety of high entropy ceramics have been synthesized, including high entropy oxides (HEOs), nitrides, carbides, borides, silicides, fluorides, etc. Among them, the local entropy increase in high entropy oxides causes the anion sublattice to be particularly susceptible to local deformation and defects. In the local area, each ion tends to isotropically distort its coordination polyhedron, causing it to deviate from the average value. The unique electronic instability of cations can also lead to anisotropic local distortion in HEO. In addition to the simple expansion, contraction or distortion of the local oxygen environment mentioned above, point defects are also the focus of affecting high entropy performance, including vacancies on cations or oxygen sublattices, elements in interstitial positions, impurities or doping elements, and antisite defects.

[0004] Whether traditional spinel refractory materials can also be optimized by introducing multi-element high entropy ideas, including mechanical properties, corrosion resistance, thermal shock resistance, etc., needs further exploration. Whether high entropy ideas can bring positive structural and performance distortion to refractory materials to achieve better refractory performance is a problem that needs to be considered at present. However, facing such a huge multi-element composition space, how to conduct efficient and feasible experimental exploration has become a serious problem. Summary of the invention

[0005] The purpose of the present invention is to provide a method for determining a high-throughput prediction model, an application method and related devices, which can quickly and accurately predict the performance of materials, improve the accuracy and speed of screening, and realize efficient and feasible experimental exploration.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] In a first aspect, the present invention provides a method for determining a high-throughput prediction model, the method for determining a high-throughput prediction model comprising:

[0008] Pearson screening is performed on each descriptor in the database to obtain a screened descriptor; the database includes several descriptors; the descriptors include: intrinsic properties from metal element atoms, physicochemical properties of the used precursor oxides and differentiated experimental conditions.

[0009] Based on the SHAP software package and the screened descriptors, the optimal algorithm-descriptor combination is determined.

[0010] Based on the optimal algorithm-descriptor combination, grid search technology is used to adjust the hyperparameters of the optimal algorithm to obtain a high-throughput prediction model for material properties.

[0011] Optionally, the expression of the Pearson filter is:

[0012]

[0013] Where r is the Person coefficient, n is the number of data points, ∑xy is the sum of the product of the index values ​​of the corresponding element descriptors of X and Y, ∑x is the sum of the index values ​​of all element descriptors in X, ∑y is the sum of the index values ​​of all element descriptors in Y, ∑x 2 is the sum of the squares of the descriptor index values ​​of all elements in X, ∑y 2 It is the sum of the squares of the descriptor index values ​​of all elements in Y.

[0014] Optionally, based on the SHAP software package and the screened descriptors, determining an optimal algorithm-descriptor combination specifically includes:

[0015] Based on the filtered descriptors, a plurality of temporary databases are formed; the temporary databases are combinations of different numbers of filtered descriptors.

[0016] Based on the SHAP software package, combinations of different algorithms and different temporary databases were determined.

[0017] The combination of different algorithms and different temporary databases with the highest R 2 The combination of the value and the RMSE value below the preset threshold is taken as the optimal algorithm-descriptor combination.

[0018] Optionally, the R 2 The value is calculated as:

[0019]

[0020] Among them, SS resis the residual sum of squares, that is, the sum of squares of the difference between the true observation value and the algorithm prediction value. The true observation value is the experimental data value of the material performance in the temporary database, and the algorithm prediction value is the predicted performance value of the algorithm. SS tot is the total sum of squares, that is, the sum of the squares of the differences between the true observations and the mean of the true observations.

[0021] Optionally, the RMSE value is calculated as follows:

[0022]

[0023] Among them, m is the total number of true observations, y i is the ith true observation value, which is the experimental data value of the material properties in the temporary database, is the ith algorithm prediction value, which is the prediction performance value of the algorithm.

[0024] Optionally, the method for determining the high-throughput prediction model further includes:

[0025] The ten-fold cross validation method was combined with R 2 The prediction accuracy of different algorithms was evaluated by using the RMSE value.

[0026] In a second aspect, the present invention provides an application method of a high-throughput prediction model, the application method of the high-throughput prediction model comprising:

[0027] Get the descriptor of the material to be tested.

[0028] The descriptor of the material to be tested is input into a high-throughput prediction model of material properties to obtain a predicted performance value corresponding to the material to be tested; the high-throughput prediction model of material properties is a model trained according to any of the high-throughput prediction model determination methods described above.

[0029] Optionally, a single-phase high-entropy spinel refractory material with a microhardness of 1770.6 HV1 is obtained by applying the application method of the high-throughput prediction model, wherein the single-phase high-entropy spinel refractory material is (Al2Fe 0.25 Zn 0.25 Mg 0.25 Mn 0.25 )O4, the microhardness of the single-phase high entropy spinel refractory material is 235% higher than that of the magnesia-alumina spinel prepared by the same experimental process, and is 1585% higher than that of the magnesia-chromium spinel prepared by the same experimental process.

[0030] Optionally, the application method of the high-throughput prediction model is used to obtain a multiphase composite refractory material with a flexural strength of 161.24 MPa, wherein the multiphase composite refractory material is (Al2Cr 0.5 Zn0.1 Mg 0.2 Mn 0.2 )O4, the room-temperature flexural strength of the multi-phase composite refractory material is 40% higher than that of the magnesia-alumina spinel prepared by the same experimental process, and is 312% higher than that of the magnesia-chromium spinel prepared by the same experimental process.

[0031] In a third aspect, the present invention provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for determining the high-throughput prediction model described above or the method for applying the high-throughput prediction model described above.

[0032] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, it implements the method for determining the high-throughput prediction model described above or the method for applying the high-throughput prediction model described above.

[0033] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, characterized in that when the computer program is executed by a processor, it implements the method for determining the high-throughput prediction model described above or the method for applying the high-throughput prediction model described above.

[0034] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0035] The present invention provides a method for determining a high-throughput prediction model, an application method, and related devices. The present invention first performs Pearson screening on each descriptor in the database, that is, uses the Pearson coefficient to calculate the degree of correlation between each descriptor, and screens out the corresponding descriptors. Then, based on the SHAP software package and the screened descriptors, the optimal algorithm-descriptor combination is determined, so that the screening process is more targeted, and the grid search technology is used to adjust the hyperparameters of the optimal algorithm to obtain a high-throughput prediction model for material properties. The high-throughput prediction model uses descriptors as input and predicted performance values ​​as output. It can quickly and accurately predict the performance of materials, thereby improving the accuracy and speed of screening and realizing efficient and feasible experimental exploration. Components with good mechanical properties at a sintering temperature of 1600°C were screened out from the prediction space for experimental verification, and a single-phase high-entropy spinel refractory (Al2Fe 0.25 Zn 0.25 Mg 0.25 Mn 0.25)O4, which is 235% higher than the microhardness of magnesium aluminum spinel (MgAl2O4) prepared by the same experimental process, and 1585% higher than the microhardness of magnesium chromium spinel (MgCr2O4) prepared by the same experimental process; it is also found that the multiphase composite refractory (Al2Cr 0.5 Zn 0.1 Mg 0.2 Mn 0.2 )O4, which has a room-temperature flexural strength that is 40% higher than that of magnesium-aluminum spinel (MgAl2O4) prepared by the same experimental process, and a room-temperature flexural strength that is 312% higher than that of magnesium-chromium spinel (MgCr2O4) prepared by the same experimental process. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. 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 paying creative work.

[0037] Figure 1 This is an application environment diagram of a method for determining a high-throughput prediction model or a method for applying a high-throughput prediction model in one embodiment of the present invention.

[0038] Figure 2 A schematic flow chart of a method for determining a high-throughput prediction model provided by one embodiment of the present invention.

[0039] Figure 3 A 3D bar chart of the RMSE values ​​of various models and descriptors for flexural strength training provided in an embodiment of the present invention.

[0040] Figure 4 A 3D histogram of the RMSE values ​​of various models-descriptors for microhardness training provided by an embodiment of the present invention.

[0041] Figure 5a The 1600°C fired (Al2Cr 0.5 Zn x Mg y Mn z )Ternary heat map for prediction of flexural strength of O4 composition system.

[0042] Figure 5b The 1600°C sintering (Al2Fe 0.25 Zn x Mg y Mn z )Ternary heat map of microhardness prediction of O4 composition system.

[0043] Figure 6 A schematic flow chart of a method for applying a high-throughput prediction model provided by one embodiment of the present invention.

[0044] Figure 7 A schematic diagram of the structure of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. 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.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Nowadays, with the growth of computer computing power, machine learning technology is gradually intersecting with materials science. Using experimental data or simulation calculation data as a database to train machine learning models, simulate the correlation distribution between descriptors and target performance, and thus quickly reduce the experimental exploration space of descriptors and target performance is a proven and feasible experimental method. However, the descriptors used in the training of different machine learning models are different. There are more accurate machine learning models that use a large number of simulation results calculated by first principles as descriptors; there are also training databases formed by weighted combinations of elemental properties or precursor physical and chemical properties, which can perform high-throughput predictions. The accuracy may not be as good as the calculation database, but the advantage is that the descriptors are easy to obtain and the huge search space of descriptors and target performance can be quickly screened without involving simulation calculation processes.

[0048] The method for determining the high-throughput prediction model provided by the embodiment of the present invention can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the descriptors in the database to the server 104. After the server 104 receives the descriptors in the database, the server 104 performs Pearson screening on each descriptor in the database to obtain the screened descriptors; the database includes several descriptors; the descriptors include: inherent properties from metal element atoms, physical and chemical properties of the precursor oxides used, and differentiated experimental conditions; based on the SHAP software package and the screened descriptors, the optimal algorithm-descriptor combination is determined; based on the optimal algorithm-descriptor combination, the grid search technology is used to adjust the hyperparameters of the optimal algorithm to obtain a high-throughput prediction model of material properties. The server 104 can feedback the obtained high-throughput prediction model of material properties to the terminal 102.

[0049] In addition, in some embodiments, the method for determining the high-throughput prediction model can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly train the high-throughput prediction model for the descriptors in the database, or the server 104 can obtain the descriptors in the database from the data storage system and train the high-throughput prediction model for the descriptors in the database.

[0050] At the same time, the application method of the high-throughput prediction model provided by the embodiment of the present invention can also be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, or it can be integrated on the server 104, or it can be placed on the cloud or other servers. The terminal 102 can send the descriptor of the material to be tested to the server 104. After the server 104 receives the descriptor of the material to be tested, the server 104 inputs the descriptor of the material to be tested into the high-throughput prediction model of the material performance to obtain the predicted performance value corresponding to the material to be tested. The server 104 can feedback the obtained predicted performance value corresponding to the material to be tested to the terminal 102.

[0051] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, and tablet computers. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0052] In an exemplary embodiment, Figure 2As shown, a method for determining a high-throughput prediction model is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In an embodiment of the present invention, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate the method, which includes the following steps S1 to S3. Among them:

[0053] S1: Perform Pearson screening on each descriptor in the database to obtain screened descriptors; the database includes several descriptors; the descriptors include: intrinsic properties from metal element atoms, physicochemical properties of the precursor oxide used, and differentiated experimental conditions.

[0054] S2: Determine the optimal algorithm-descriptor combination based on the SHAP software package and the screened descriptors.

[0055] S3: Based on the optimal algorithm-descriptor combination, grid search technology is used to adjust the hyperparameters of the optimal algorithm to obtain a high-throughput prediction model for material properties.

[0056] By implementing the above-mentioned steps S1 to S3, the present invention places the importance ranking method in the SHAP software package in advance in feature engineering, which not only provides the direction of descriptor screening, but also can provide different descriptor combinations for different algorithms, making the screening process more targeted and improving the accuracy and speed of screening.

[0057] In step S1, the establishment of a database specifically includes:

[0058] The database contains two parts of descriptors. One part of the descriptors comes from the inherent properties of metal element atoms and the physical and chemical properties of the precursor oxides used. They not only provide basic information on material properties, but also reveal the composite influence of various components of the material through the calculation of weighted averages and weighted standard deviations. The element and atomic data are obtained from recognized free databases (including but not limited to Materials Project (MP), AFLOW and Royal Society of Chemistry, etc.). The other part of the descriptors is differentiated experimental conditions, such as sintering temperature and holding time for different results.

[0059] The calculation formula of weighted average is as follows:

[0060]

[0061] The formula for calculating the weighted standard deviation is as follows:

[0062]

[0063] in, is the weighted average, x i is the value of the i-th group, w i is the weight corresponding to the i-th group of values, s w is the weighted standard deviation, n is the number of data points. The weighted standard deviation calculation formula is the formula used when the weights are normalized.

[0064] With these easily accessible descriptors, high-throughput predictions have a foundation. However, before the actual application of machine learning models, these descriptors must be screened to ensure the effectiveness and accuracy of the models.

[0065] In step S1, the correlation between each descriptor is calculated using the Pearson coefficient (Person coefficient), which is a statistical indicator for measuring the linear correlation between two variables, and its value is between -1 and +1. A positive value indicates a positive correlation, a negative value indicates a negative correlation, and a zero value indicates no linear correlation.

[0066] The expression of the Pearson filter is:

[0067]

[0068] Where r is the Person coefficient, n is the number of data points, ∑xy is the sum of the product of the index values ​​of the corresponding element descriptors of X and Y, ∑x is the sum of the index values ​​of all element descriptors in X, ∑y is the sum of the index values ​​of all element descriptors in Y, ∑x 2 is the sum of the squares of the descriptor index values ​​of all elements in X, ∑y 2 It is the sum of the squares of the descriptor index values ​​of all elements in Y.

[0069] In another exemplary embodiment of the present invention, in step S2, specifically including:

[0070] S21: Based on the filtered descriptors, a plurality of temporary databases are formed; the temporary databases are combinations of different numbers of filtered descriptors.

[0071] S22: Based on the SHAP software package, determine the combination of different algorithms and different temporary databases.

[0072] S23: Combining different algorithms with different temporary databases to obtain the highest R 2 The combination of the value and the RMSE value below the preset threshold is taken as the optimal algorithm-descriptor combination.

[0073] Preliminary screening of multiple descriptors with a Pearson coefficient less than 0.9. Then, these descriptors are sorted by independent SHAP importance using multiple different algorithms (the type of algorithm can be adjusted by the user). Each algorithm takes out the top 4, 8, 12, 16, 20, 24 descriptors in its own SHAP importance ranking, and forms multiple temporary databases for training and screening. Finally, a combination of an algorithm and a temporary database with the highest R 2 The RMSE value with a value lower than the preset threshold is used for the subsequent hyperparameter adjustment of the selected model and the high-throughput prediction of material properties.

[0074] It should be noted that when training different algorithms, descriptors are used as input and target performance is used as output; and the temporary database used to train each algorithm is a combination of descriptors arranged according to the SHAP importance of the algorithm itself, and the training database corresponding to each algorithm is different. This makes descriptor screening algorithm-specific in database feature engineering.

[0075] The above methods for descriptor screening combined with the SHAP software package all use the ten-fold cross-validation method combined with R 2 The prediction accuracy of different algorithms was evaluated by using the RMSE value.

[0076] Among them, the R 2 The value is calculated as:

[0077]

[0078] Among them, SS res is the residual sum of squares, that is, the sum of squares of the difference between the true observation value and the algorithm prediction value. The true observation value is the experimental data value of the material performance in the temporary database, and the algorithm prediction value is the predicted performance value of the algorithm. SS tot is the total sum of squares, that is, the sum of the squares of the differences between the true observations and the mean of the true observations.

[0079] The calculation formula of the RMSE value is:

[0080]

[0081] Among them, m is the total number of true observations, y i is the ith true observation value, which is the experimental data value of the material properties in the temporary database, is the ith algorithm prediction value, which is the prediction performance value of the algorithm.

[0082] Write a Python program for the machine learning model with the optimal hyperparameters determined above, and encapsulate the model into a function so that it can perform component-performance (descriptor-target performance) calculation prediction. The specific prediction function code is as follows:

[0083] import pandas as pd

[0084] import pickle

[0085] import numpy as np

[0086] df11 = pd.read_csv('Element Information Table 25.csv', index_col = 0, header = 0) # Read in element and precursor property data

[0087] df22 = pd.read_csv('Table of oxide properties 25.csv', index_col = 0, header = 0) # read in element and precursor property data

[0088] df33 = pd.read_csv('Element Information Table SD25.csv', index_col = 0, header = 0) # Read in element and precursor property data

[0089] df44 = pd.read_csv('Table of oxide properties SD25.csv', index_col = 0, header = 0) # read in element and precursor property data

[0090] with open('rf_kz_model.pkl','rb')as file:#Package model read

[0091] model = pickle.load(file)

[0092] def SDform(p,df):#Define the weighted standard deviation function of the database

[0093] a=[]

[0094] weights = np.array(p)

[0095] row_names = df.index.tolist()

[0096] for in range(len(df)):

[0097] datap.array(df.iloc[i])

[0098] weighted_mean_np.average(data, weights_weights)

[0099] weighted_varianceNp.average((data-weighted_mean)**2, weights_weights) .

[0100] weighted_stdnp.sqrt(weighted_variance);

[0101] a.append(weighted_std)

[0102] seriesPd.Series(a,index!row_names)

[0103] return series

[0104] def predict_ternary(new_p):#If you have a ternary, the prefix is ​​not a ternary type

[0105] df1 df11*new_p

[0106] df2 df22*new_p

[0107] concatenated_seriesPd.concat([df1.sum(axis1),df2.sum(axis1),SDform(new_p,df33),SDform(new_p,df44)],axis1);

[0108] final_pd.DataFrame(concatenated_series.fill(0).sum(axis1)).T

[0109] final [ 'T ' ] 1200

[0110] KZ_num model.predict(final)

[0111] result,float(KZ_num[0]);

[0112] return result

[0113] The above function can be combined with the Ternary library in Python to draw a ternary heat map of component-performance prediction, thus achieving the purpose of high-throughput prediction.

[0114] The application of the above method is introduced below.

[0115] The process of high entropy optimization of the mechanical properties of spinel refractory materials using this method is as follows:

[0116] (1) Establish a database and filter out descriptors.

[0117] The database contains 48 descriptors. The first 46 descriptors come from the intrinsic properties of metal atoms and the physicochemical properties of the precursor oxides used; the 47th descriptor in the database is the thermodynamic mixing entropy ΔS conf ; The 48th descriptor is the sintering temperature of the sample after pressing, which is one of the key process parameters affecting the material properties.

[0118] The 48 original descriptors were removed from those with an absolute value of the Pearson coefficient greater than 0.9, and the remaining 24 basic descriptors are shown in Table 1.

[0119] Table 1 24 descriptors with absolute values ​​of Pearson coefficients less than 0.9

[0120]

[0121] (2) Screened algorithms.

[0122] After obtaining 24 less relevant descriptors, eight algorithms (Adaboost, DecisionTreeRegressor, Elastic Net Regression, Lasso Regression, NeuralNetworks, Random Forest, Ridge Regression, XGBoost) were screened based on the SHAP software package and combined with databases with different numbers of descriptors. In the training of fold strength, the Random Forest algorithm and its database composed of the top 8 descriptors ranked by SHAP had the highest R 2 The value is 0.571, and the specific data are shown in Table 2. There is also a lower RMSE value of 17.67MPa. Figure 3 As shown; therefore, the Random Forest algorithm is selected to train the next flexural strength prediction model.

[0123] Table 2 R of 8 algorithms for training and evaluation of flexural strength 2value

[0124]

[0125]

[0126] In the training of microhardness, the Neural Networks algorithm and its database consisting of the top 8 descriptors ranked by SHAP have the highest R 2 The value is 0.574, the specific data are shown in Table 3, and there is a lower RMSE value of 136.79HV1 Figure 4 As shown; therefore, the neural network algorithm (NN) is used to train the next microhardness prediction model.

[0127] Table 3 R of 8 algorithms for microhardness training and evaluation 2 value

[0128]

[0129] (3) Prediction model with good performance.

[0130] For the above high R 2 We used the grid search technique to optimize the hyperparameters of the neural network algorithm and the random forest algorithm with low RMSE values ​​and ten-fold cross-validation. 2 As the model evaluation standard, a large hyperparameter search space was defined, and the number of calculation iterations exceeded 10 million times. Finally, the random forest model R trained with the 112-group sample flexural strength database was used. 2 The value is 0.9326 and the RMSE value is 8.92MPa. The neural network model R trained using the microhardness database of 112 groups of samples 2 The value is 0.8516 and the RMSE value is 111.75HV1. The specific hyperparameter selections of the two models are shown in Table 4.

[0131] Table 4 Algorithm hyperparameter selection

[0132] Random Forest Hyperparameters NN Hyperparameters n_estimators = 54 hidden_layer_sizes=(148,148) max_depth=8 activation='relu' min_samples_split=2 alpha = 0.061 min_samples_leaf=1 learning_rate = 'constant' max_features = 'log2' learning_rate_init=0.001 criterion = 'friedman_mse' max_iter=1026

[0133] High-throughput prediction of components with optimal mechanical properties.

[0134] After obtaining two models with excellent performance, we explored the flexural strength and microhardness of ceramics sintered at 1600℃ for the consideration of refractory applications. The random forest model predicting flexural strength showed that high entropy spinel (Al2Cr 0.5 Zn x Mg y Mn z)O4 has a room temperature flexural strength of more than 130MPa, which is more than 50% higher than that of magnesium aluminum spinel (MgAl2O4) and magnesium chromium spinel (MgCr2O4) prepared by the same experimental process. The predicted ternary thermal map is as follows Figure 5a The neural network model predicting microhardness shows that high entropy spinel (Al2Fe 0.25 Zn x Mg y Mn z )O4 has a microhardness of more than 1500HV1, which is more than 300% higher than the room temperature flexural strength of magnesium aluminum spinel (MgAl2O4) and magnesium chromium spinel (MgCr2O4) prepared by the same experimental process. Figure 5b shown.

[0135] In summary, this method achieves high-throughput prediction of material mechanical properties, opening up ideas for optimizing the mechanical properties of spinel refractory materials.

[0136] From the above predicted ternary heat map, we screened out components with good mechanical properties at a sintering temperature of 1600°C and conducted experimental verification to obtain the following:

[0137] 1. Single-phase high entropy spinel refractory (Al2Fe 0.25 Zn 0.25 Mg0. 25 Mn 0.25 )O4, the microhardness is 235% higher than that of magnesium aluminum spinel (MgAl2O4) prepared by the same experimental process, and the microhardness is 1585% higher than that of magnesium chromium spinel (MgCr2O4) prepared by the same experimental process;

[0138] 2. Multiphase composite refractory (Al2Cr 0.5 Zn 0.1 Mg 0.2 Mn 0.2 )O4, which has a room-temperature flexural strength that is 40% higher than that of magnesium-aluminum spinel (MgAl2O4) prepared by the same experimental process, and a room-temperature flexural strength that is 312% higher than that of magnesium-chromium spinel (MgCr2O4) prepared by the same experimental process.

[0139] The present invention places the importance ranking method in the SHAP software package in the feature engineering, which not only provides the direction of descriptor screening, but also provides different descriptor combinations for different algorithms, making the screening process more targeted and improving the accuracy and speed of screening. Through this method, a database was established using experimental data for machine learning model training, and a neural network model and a random forest model with good generalization ability for predicting mechanical properties were obtained, outputting a huge search space for the relationship between composition and mechanical properties. And from the prediction space, components with good mechanical properties at a sintering temperature of 1600°C were screened for experimental verification, and a single-phase high-entropy spinel refractory (Al2Fe 0.25 Zn 0.25 Mg 0.25 Mn 0.25 )O4, which is 235% higher than the microhardness of magnesium aluminum spinel (MgAl2O4) prepared by the same experimental process, and 1585% higher than the microhardness of magnesium chromium spinel (MgCr2O4) prepared by the same experimental process; it is also found that the multiphase composite refractory (Al2Cr 0.5 Zn 0.1 Mg 0.2 Mn 0.2 )O4, which has a room-temperature flexural strength that is 40% higher than that of magnesium-aluminum spinel (MgAl2O4) prepared by the same experimental process, and a room-temperature flexural strength that is 312% higher than that of magnesium-chromium spinel (MgCr2O4) prepared by the same experimental process.

[0140] In an exemplary embodiment, Figure 6 As shown, a method for applying a high-throughput prediction model is provided, and the method for applying the high-throughput prediction model comprises:

[0141] A1: Get the descriptor of the material to be tested.

[0142] A2: Input the descriptor of the material to be tested into a high-throughput prediction model of material properties to obtain a predicted performance value corresponding to the material to be tested; the high-throughput prediction model of material properties is a model trained according to any of the high-throughput prediction model determination methods described above.

[0143] The present invention also provides an application scenario, which applies the above-mentioned method for determining a high-throughput prediction model and the method for applying a high-throughput prediction model. Specifically: The method for determining a high-throughput prediction model and the method for applying a high-throughput prediction model provided in this embodiment can be applied in a material performance prediction scenario. The material performance prediction scenario includes the acquisition and screening of descriptors, the determination of the optimal algorithm-descriptor combination, the hyperparameter adjustment link, and the prediction link; after the descriptors in the database are screened, based on the SH AP software package, the combination of different algorithms and different temporary databases can be determined; the combination of different algorithms and different temporary databases with the highest value and the value below the preset threshold is used as the optimal algorithm-descriptor combination, and then the grid search technology is used to adjust the hyperparameters of the optimal algorithm, so as to obtain a high-throughput prediction model of material performance, and the descriptor of the material to be tested is input into the high-throughput prediction model of material performance, so as to obtain the predicted performance value corresponding to the material to be tested.

[0144] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store descriptors in the database or descriptors of the material to be tested. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for determining a high-throughput prediction model or an application method of a high-throughput prediction model is implemented.

[0145] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0146] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the above-mentioned method embodiments are implemented when the processor executes the computer program.

[0147] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the above-mentioned method embodiments are implemented.

[0148] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the above-mentioned method embodiments are implemented.

[0149] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0150] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0151] The database involved in each embodiment provided by the present invention may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided by the present invention may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0152] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0153] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for determining a high-throughput prediction model, characterized in that: The method for determining the high-throughput prediction model comprises: Performing Pearson screening on each descriptor in the database to obtain screened descriptors; the database includes a plurality of descriptors; the descriptors include: intrinsic properties from metal element atoms, physicochemical properties of precursor oxides used, and differentiated experimental conditions; Determine the optimal algorithm-descriptor combination based on the SHAP software package and the screened descriptors; Based on the optimal algorithm-descriptor combination, grid search technology is used to adjust the hyperparameters of the optimal algorithm to obtain a high-throughput prediction model for material properties.

2. The method for determining a high-throughput prediction model according to claim 1, characterized in that: The expression of the Pearson filter is: Where r is the Pearson coefficient, n is the number of data points, ∑xy is the sum of the product of the index values ​​of the element descriptors corresponding to X and Y, ∑x is the sum of the index values ​​of all element descriptors in X, ∑y is the sum of the index values ​​of all element descriptors in Y, ∑x 2 is the sum of the squares of the descriptor index values ​​of all elements in X, ∑y 2 It is the sum of the squares of the descriptor index values ​​of all elements in Y.

3. The method for determining a high-throughput prediction model according to claim 1, characterized in that: Based on the SHAP software package and the screened descriptors, the optimal algorithm-descriptor combination is determined, specifically including: Based on the filtered descriptors, a plurality of temporary databases are formed; the temporary databases are combinations of different numbers of filtered descriptors; Based on the SHAP software package, determining the combination of different algorithms and different temporary databases; The combination of different algorithms and different temporary databases with the highest R 2 The combination of the value and the RMSE value below the preset threshold is taken as the optimal algorithm-descriptor combination.

4. The method for determining a high-throughput prediction model according to claim 3, characterized in that: The R 2 The value is calculated as: Among them, SS res is the residual sum of squares, that is, the sum of squares of the difference between the true observation value and the algorithm prediction value. The true observation value is the experimental data value of the material performance in the temporary database, and the algorithm prediction value is the predicted performance value of the algorithm. SS tot is the total sum of squares, that is, the sum of the squares of the differences between the true observations and the mean of the true observations.

5. The method for determining a high-throughput prediction model according to claim 3, characterized in that: The calculation formula of the RMSE value is: Among them, m is the total number of true observations, y i is the ith true observation value, which is the experimental data value of the material properties in the temporary database, is the ith algorithm prediction value, which is the prediction performance value of the algorithm.

6. The method for determining a high-throughput prediction model according to claim 3, characterized in that: The method for determining the high-throughput prediction model also includes: The ten-fold cross validation method was combined with R 2 The prediction accuracy of different algorithms was evaluated by using the RMSE value.

7. A method for applying a high-throughput prediction model, characterized in that: The application method of the high-throughput prediction model includes: Get the descriptor of the material to be tested; The descriptor of the material to be tested is input into a high-throughput prediction model of material properties to obtain a predicted performance value corresponding to the material to be tested; the high-throughput prediction model of material properties is a model trained according to the method for determining a high-throughput prediction model according to any one of claims 1-6.

8. The application method of the high-throughput prediction model according to claim 7, characterized in that: A single-phase high-entropy spinel refractory material with a microhardness of 1770.6 HV1 is obtained by applying the application method of the high-throughput prediction model. The single-phase high-entropy spinel refractory material is (Al2Fe 0.25 Zn 0.25 Mg 0.25 Mn 0.25 )O4, the microhardness of the single-phase high entropy spinel refractory material is 235% higher than that of the magnesia-alumina spinel prepared by the same experimental process, and is 1585% higher than that of the magnesia-chromium spinel prepared by the same experimental process.

9. The application method of the high-throughput prediction model according to claim 7, characterized in that: The application method of the high-throughput prediction model is used to obtain a multiphase composite refractory material with a flexural strength of 161.24 MPa. The multiphase composite refractory material is (Al2Cr 0.5 Zn 0.1 Mg 0.2 Mn 0.2 )O4, the room-temperature flexural strength of the multi-phase composite refractory material is 40% higher than that of the magnesia-alumina spinel prepared by the same experimental process, and is 312% higher than that of the magnesia-chromium spinel prepared by the same experimental process.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for determining the high-throughput prediction model described in any one of claims 1 to 6 or the method for applying the high-throughput prediction model described in claim 7.

Images