Prediction method and device for high-entropy oxide material based on machine learning and medium

Abstract

The invention provides a high-entropy oxide material prediction method and device based on machine learning and a medium. The high-entropy oxide material prediction method and device based on machine learning can efficiently screen high-entropy rare earth titanate composition elements and predict heat conductivity of high-entropy rare earth titanate with other compositions. According to the method, the prediction model based on machine learning is established, a more efficient mode is provided for determining the thermal conductivity of the high-entropy rare earth titanate, and the research progress and application of the high-entropy rare earth titanate coating material can be promoted. According to the method, data of material composition and heat conductivity are extracted from a large number of published literatures, the mapping relation between each feature and the heat conductivity is explored through a machine learning model, the influence trend of each feature on the heat conductivity is defined, and the high-entropy rare earth titanate with the lower heat conductivity is designed more quickly.

Description

Technical Field

[0001] The present invention relates to the technical field of ceramic coating material screening and prediction, and in particular to a prediction method, device and medium for high entropy oxide materials based on machine learning. Background Art

[0002] Thermal protection coatings (TPCs) are widely used in high-temperature equipment such as jet engines and aircraft gas turbines to ensure the normal operation of the equipment under extremely high temperatures. Conventional coatings made from yttria-partially stabilized zirconia (YSZ) TPC materials have a low upper limit for high-temperature resistance. When the temperature exceeds 1200°C, YSZ undergoes a phase transition accompanied by a volume change, which can easily cause the coating to fail. To further increase the operating temperature of high-temperature components, a thermal protection coating material with low thermal conductivity and high stability is urgently needed.

[0003] High-entropy rare earth titanate ceramics have low thermal conductivity and excellent high-temperature stability, making them suitable as a novel thermal protection coating material. However, due to the vast potential elemental space, searching for high-entropy elements using either traditional experimental methods or first-principles calculations is time-consuming and resource-intensive. Efficiently and accurately screening the elements that make up high-entropy rare earth titanates remains a significant challenge. Currently, there is no particularly efficient method for screening the elements that make up high-entropy rare earth titanates. Summary of the Invention

[0004] In view of this, the present invention provides a prediction method, device and medium for high-entropy oxide materials based on machine learning, which can efficiently screen the constituent elements of high-entropy rare earth titanates and predict the thermal conductivity of high-entropy rare earth titanates of other compositions.

[0005] (Consistent with the claims)

[0006] Beneficial effects:

[0007] 1. The method of the present invention establishes a prediction model based on machine learning, which provides a more efficient way to determine the thermal conductivity of high-entropy rare earth titanates and can promote the research progress and application of high-entropy rare earth titanate coating materials.

[0008] 2. In the method of the present invention, data on material composition and thermal conductivity are extracted from a large number of publicly published literature, and the mapping relationship between each characteristic and thermal conductivity is explored through a machine learning model, the influence trend of each characteristic on thermal conductivity is clarified, and high-entropy rare earth titanates with lower thermal conductivity are designed more quickly.

[0009] 3. In this method, five mainstream machine learning algorithms were used to compare performance and identify the XGBoost model with the highest prediction accuracy. Thirteen candidate features were further screened using an exhaustive search method, ultimately retaining seven key features. This method reduced model complexity by 46% while maintaining prediction accuracy, significantly improving the model's interpretability and practicality.

[0010] 4. In the present invention, in the method of the present invention, by learning existing data, all data in the composition space can be predicted, which shortens the time period required to explore low thermal conductivity and high entropy rare earth titanates, reduces the trial and error cost of experiments, and improves the accuracy and reliability of low thermal conductivity material screening.

[0011] 5. In the method of the present invention, the R of the validation set 2 The prediction accuracy is 0.97, which is significantly higher than that of traditional theoretical models, providing a repeatable and verifiable methodology for the design of complex material systems.

[0012] 6. The present method, guided by a model, identified a novel high-entropy titanate system with a thermal conductivity as low as 1.54 W / (m·K). Compared to YSZ, this system combines both low thermal conductivity and high thermal stability. This provides a new path for the development of thermal barrier coatings, and is expected to further increase the service temperature of aircraft engine turbine blades and improve energy efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a workflow diagram of a method for screening and predicting high entropy rare earth titanate coating materials based on machine learning according to an embodiment of the present invention.

[0014] Figure 2 This is a data distribution diagram of thermal conductivity of an initial data set provided according to an embodiment of the present invention.

[0015] Figure 3 The five machine learning models R provided according to the embodiment of the present invention are 2 , and RMSE comparison chart.

[0016] Figure 4 This is a graph of the optimal feature subset under the optimal model provided by an embodiment of the present invention.

[0017] Figure 5 This is a comparison chart of the predicted value and the actual value of the thermal conductivity of the material based on the extreme gradient boosting model (XGB) provided according to an embodiment of the present invention.

[0018] Figure 6 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0020] The rapid development of machine learning technology has provided a new opportunity for the efficient screening of high-entropy rare earth titanate constituent elements. Machine learning can quickly discover implicit and effective information from a large amount of data, which provides effective help for the screening of high-entropy rare earth titanate constituent elements. For example, the extreme gradient boosting tree model (eXtreme Gradient Boosting, XGB) works in an integrated learning manner, constructs multiple weak learners (usually decision trees), and trains them based on the residuals of the previous learners, with the aim of reducing the value of the loss function. Its core lies in the use of a second-order Taylor expansion to approximate the objective function, which can perform gradient descent optimization more accurately. It performs well in both classification and regression problems. The present invention proposes a method for efficiently screening high-entropy rare earth titanate constituent elements from the perspective of machine learning.

[0021] This embodiment provides a method for screening and predicting high entropy rare earth titanate coating materials based on machine learning. The process is as follows: Figure 1 As shown, the following steps are included:

[0022] Step 1: Based on the high entropy rare earth titanate system studied in this example, data is collected from published literature and laboratories, including material composition and thermal conductivity information, to establish an initial data set.

[0023] Specifically, building a complete initial dataset includes the following sub-steps:

[0024] Step 1-1: The perovskite structure has the advantages of high symmetry, high bonding strength, and the ability to withstand extreme temperatures. It is not prone to phase transitions and decomposition at high temperatures and is very stable. The single-phase structure avoids defects and stress concentrations at the grain boundaries of multiphase materials, ensuring thermal expansion consistency and chemical stability. Select appropriate keywords and search and collect literature related to the single-phase structure of perovskites in the literature database;

[0025] Step 1-2: perform text and image mining on the collected literature data with perovskite single-phase structure, and set the exclusion threshold to an average ion radius less than The triple criterion of radius deviation δr ≥ 0.118 and mass deviation δm ≥ 0.357, where xi is the atomic fraction of the element, ri is the ionic radius, and mi is the atomic mass. Finally, 253 groups of material composition and thermal conductivity information that meet the conditions are extracted;

[0026] In steps 1-3, the selected features include 13 features: average ionic radius of the A site in high entropy rare earth titanate ATiO3, average atomic mass, mass deviation, radius deviation, tolerance factor, average atomic number, average electron concentration, average electronegativity, average first ionization energy, mass range, radius range, first ionization energy range, and temperature.

[0027] In steps 1-4, the characteristics and thermal conductivity data are organized to create the initial data set.

[0028] Step 2: Clean and preprocess the dataset based on machine learning, and then divide the dataset into training set, test set, and prediction set.

[0029] Specifically, step 2 includes the following sub-steps:

[0030] Step 2-1: Based on machine learning, check whether the initial data set contains null values ​​or outliers, and perform data normalization to obtain a data set for the machine learning model. The data set includes material characteristics and corresponding thermal conductivity information;

[0031] The outliers are defined as values ​​less than Q1-1.5×IQR or greater than Q3+1.5×IQR, where Q1 is the first quartile, Q3 is the third quartile, and IQR=Q3-Q1. The data are normalized using the linear normalization method, x norm =(xx min ) / (x max -x min ), where x is the original data, x min is the minimum value of each feature, x max is the maximum value of each feature, x norm is the normalized data.

[0032] Figure 2 This is the data distribution of the thermal conductivity of the initial data set in Example 1 of the present invention. There are no outliers, the data distribution is relatively concentrated, and the degree of dispersion is low.

[0033] In step S2-2, based on machine learning, the data set is divided into training set, test set, and validation set in a ratio of 7:2:1.

[0034] Step 3: Perform model training and evaluation on the dataset, screen out the optimal model and optimal feature subset, and perform five-fold cross-validation to ensure that the model has good generalization performance.

[0035] Specifically, step S3 includes the following sub-steps:

[0036] In step 3-1, the Pearson correlation coefficient is calculated for the features of the data set obtained in S2, and features with a correlation coefficient greater than 0.9 are deleted to obtain a low-correlation feature set, which consists of 11 features in total.

[0037] Step 3-2: For the low correlation feature set, the gradient boosting decision tree model (GBDT), Gaussian process regression model (GPR), support vector regression model (SVR), random forest model (RF), and extreme gradient boosting model (XGB) are used to train the models respectively. 2 (coefficient of determination) and RMSE (root mean square error) are used as evaluation indicators to screen the model and determine the model to be used in the next step.

[0038] Figure 3 The five machine learning models in Example 1 of the present invention calculate R through step 3-2 2 , RMSE comparison chart, such as Figure 3 As shown, the extreme gradient boosting model (XGB) has a larger 2 and smaller RMSE, so the extreme gradient boosting model (XGB) is selected for subsequent operations.

[0039] Step 3-3, for the selected models, use the exhaustive method to filter the low correlation feature set. First, store the indexes of all features in a list, and then define best_R 2 The two variables best_feature_combination are used to record the currently found optimal R 2 The initial values ​​of the two variables are set to 0 and None respectively. Set up a for loop, starting from selecting 1 feature, gradually increasing to selecting 11 features, and calling the combinations function in the Python standard library itertools to loop to generate all non-repeated feature index combinations, select features by index, use the data under the current feature combination to train the model to obtain the prediction results, and then use the R of the current feature combination to calculate the prediction results. 2 Compared with the previously recorded optimal R 2 (best_R 2 ) for comparison. If the current R 2 Higher, update best_R 2 For the current R 2 And update best_feature_combination to the current feature combination. Finally, output the optimal R 2 Combined with corresponding features.

[0040] Figure 4This is the result of feature screening using the exhaustive method in step 3-3 in Example 1 of the present invention. When the number of features is 7 (temperature, average ion radius, average atomic mass, tolerance factor, average electron concentration, mass range, first ionization energy range), the model effect is the best and is the optimal feature subset. The effect of the optimized extreme gradient boosting model (XGB) is further improved, and the R on the validation set is 0. 2 It increased from 0.95 to 0.97, and RMSE decreased from 0.12 to 0.09.

[0041] In steps 3-4, the model is subjected to five-fold cross-validation. The cross-validation scores are 0.98823085, 0.77590104, 0.9123058, 0.94294739, and 0.88175508, respectively. The variance of the five-fold cross-validation scores is 0.005, indicating that the model has good generalization performance.

[0042] Figure 5 It is the fitting effect of the data set in Example 1 of the present invention on the extreme gradient boosting model (XGB).

[0043] In steps 3-4, the SHAP method is used to examine the relationship between the features and thermal conductivity. Negative SHAP values ​​are more frequent for temperature and poor quality. In most cases, increases in these two features may reduce thermal conductivity.

[0044] Step 4: Use the trained model to search in the component space to complete the screening and prediction of new materials. Material composition, elements, normalization processing

[0045] Specifically, step S4 includes the following sub-steps:

[0046] Step 4-1: In order to adjust the charge balance, an alkali metal element is introduced into the A position of the high-entropy rare earth titanate, and the rare earth element and the alkali metal element are arranged and combined to create a theoretical high-entropy rare earth titanate database.

[0047] Step 4-2, calculate the radius deviation, mass deviation and A-site average ionic radius of the material in the theoretical high entropy rare earth titanate database to ensure that a single-phase perovskite structure can be generated and the average ionic radius is less than Data with a radius deviation greater than or equal to 0.118 and a mass deviation greater than or equal to 0.357 were deleted.

[0048] The radius deviation calculation formula is: The formula for calculating mass deviation is: The calculation formula for the average ionic radius at the A site is: where x i is the atomic fraction of the ith element, r iis the ionic radius of the i-th element, m i is the ionic radius of the ith element.

[0049] In step 4-3, the screened high-entropy rare earth titanate data is used as the input of the optimal model obtained in S3 to obtain the thermal conductivity predicted by the model, thereby completing the screening and prediction of the high-entropy rare earth titanate coating material.

[0050] The embodiment of the present application also provides an electronic device, Figure 6 The structure of an electronic device provided by an embodiment of the present invention is shown. For example, the electronic device 60 may include a processor 61, a memory 62, and a transmission device 63. The processor 61 is used to execute the prediction method of high entropy oxide materials based on machine learning mentioned in the above embodiment, wherein the processor and the memory can be connected by a bus or other means, taking a bus connection as an example. The transmission device can be connected to the processor and the memory by wire or wirelessly. The memory is a non-transitory computer-readable storage medium that can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as the program instructions / modules corresponding to the prediction method of high entropy oxide materials based on machine learning in the embodiment of the present application. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory, that is, implementing the prediction method of high entropy oxide materials based on machine learning in the above method embodiment. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include a memory remotely located relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the aforementioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. The one or more modules are stored in the memory and, when executed by the processor, perform the method for predicting high entropy oxide materials based on machine learning in the embodiment.

[0051] As another aspect, the present application also provides a computer-readable storage medium, which may be the computer-readable storage medium included in the device described in the above embodiment; or it may be a computer-readable storage medium that exists independently and is not assembled into the device. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the art. The computer-readable storage medium stores one or more programs, which are used by one or more processors to execute the prediction method for high entropy oxide materials based on machine learning described in the present application.

[0052] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A prediction method for high entropy oxide materials based on machine learning, characterized in that: The following steps are involved: Step S1: constructing an initial data set, obtaining the composition data of high-entropy rare earth titanate materials with perovskite single-phase structure and their corresponding thermal conductivity through literature mining, and extracting characteristic parameters, wherein the characteristic parameters include the average ionic radius of the A site in the high-entropy rare earth titanate ATiO3, the average atomic mass, the mass deviation, the radius deviation, the tolerance factor, and the temperature; Step S2: Clean and preprocess the data set based on machine learning, and then divide the data set into training set, test set, and prediction set; Step S3: Perform model training and evaluation on the dataset, screen out the optimal model and optimal feature subset, and perform five-fold cross validation; Step S4: Apply the trained prediction model to predict the thermal conductivity of the theoretical high entropy rare earth titanate database, and screen out target materials that meet the single-phase structure criterion and have a thermal conductivity lower than a threshold.

2. The method according to claim 1, wherein The literature data screening in step S1 adopts a triple criterion: the average ion radius is less than Radius deviation δr ≥ 0.118, mass deviation δm ≥ 0.357; xi is the atomic fraction of the element, ri is the ionic radius, and mi is the atomic mass.

3. The method according to claim 2, wherein ,The data preprocessing in step S2 specifically includes: using the interquartile range method to detect outliers: data outside the range of Q1-1.5IQR to Q3+1.5IQR are considered outliers, and the linear normalization formula x is applied. norm =(xx min ) / (x max -x min ), perform feature scaling, where x is the original data, x min is the minimum value of each feature, x max is the maximum value of each feature, x norm is the normalized data.

4. The method according to any one of claims 1 to 3, wherein ,The model training in step S3 specifically includes: step 3-1, calculating the Pearson correlation coefficient of the features of the data set obtained in S2, deleting features with a correlation coefficient greater than 0.9, and obtaining a low-correlation feature set; Step 3-2: Use gradient boosting decision tree model, Gaussian process regression model, support vector regression model, random forest model and extreme gradient boosting model to train the low correlation feature set. 2 Use RMSE as evaluation indicators to screen the model and determine the model to be used in the next step; Step 3-3: For the selected models, use the exhaustive method to filter the low-correlation feature set; first store the indexes of all features in a list, and then define best_R 2 The two variables best_feature_combination are used to record the currently found optimal R 2 and the corresponding feature combination, the initial values ​​of the two variables are set to 0 and None respectively; set up a for loop, starting from selecting 1 feature, gradually increasing to selecting all features in the low-correlation feature set, and calling the combinations function in the Python standard library itertools to loop to generate all non-repeated feature index combinations, select features by index, use the data under the current feature combination to train the model to obtain the prediction result, and then use the R of the current feature combination to calculate the prediction result. 2 Compared with the previously recorded optimal R 2 Compare, if the current R 2 Higher, update best_R 2 For the current R 2 , and update best_feature_combination to the current feature combination; finally output the optimal R 2 Combined with corresponding features; In steps 3-4, five-fold cross validation is used to evaluate the generalization ability of the model.

5. The method according to claim 4, wherein The step S4 includes: introducing an alkali metal element at the A position to perform charge balance, generating radius deviation, mass deviation and average ion radius parameters of the theoretical combination, and deleting data entries that do not meet the single-phase perovskite structure conditions.

6. The method according to claim 4, wherein The high entropy rare earth titanate is an ATiO3 type perovskite structure, wherein the A position contains a combination of at least three elements selected from lanthanide elements and alkali metal elements; the determination coefficient R of the prediction model on the validation set is 2 ≥0.97, and the predicted minimum thermal conductivity is 1.54 W / (m·K).

7. The method according to claim 6, wherein The single-phase perovskite structure conditions include: a tolerance factor t satisfying 0.8≤t≤1.0, an average electron concentration ≥2.5×10 22 cm -3 , the first ionization energy difference is ≤0.5eV.

8. An electronic device, characterized in that: The electronic device includes a processor and a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the high entropy oxide material prediction method based on machine learning as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, which is used to execute the prediction method of high entropy oxide materials based on machine learning as described in any one of claims 1 to 7.

Images