High-infrared radiation characteristic and high-entropy rare earth tantalate material screening method based on data driving and application of high-infrared radiation characteristic and high-entropy rare earth tantalate material screening method

Abstract

The invention discloses a data-driven high-infrared-radiation-characteristic high-entropy rare earth tantalate material screening method and application, and belongs to the technical field of thermal barrier coating materials. The method aims at breaking through the infrared radiation performance bottleneck of a traditional rare earth tantalate ceramic material, a high-precision prediction model of the infrared emissivity of the rare earth tantalate is innovatively developed by constructing a multi-system common data set, and rapid evaluation of the infrared spectral characteristics of the A-site high-entropy rare earth tantalate RE3TaO7 material is achieved. On the basis of the model, a multi-element rare earth tantalate system which comprises (Y < 0.2 > Dy < 0.2 > Gd < 0.2 > Sm < 0.2 > Nd < 0.2 >) < 3 > TaO7 and (Nd < 0.2 > Eu < 0.2 > Dy < 0.2 > Er < 0.2 > Yb < 0.2 >) < 3 > TaO7 and has excellent infrared radiation characteristics is successfully screened out. According to the method, the limitation of traditional small sample material research is broken through through big data modeling, and an efficient theoretical design normal form is provided for developing a novel high-entropy infrared radiation material.

Description

Technical Field

[0001] The present invention belongs to the technical field of thermal barrier coating materials, and specifically relates to a data-driven screening method and application of high-infrared radiation characteristic high-entropy rare earth tantalate materials. Background Art

[0002] Radiative thermal management optimizes energy transfer by regulating both thermal conduction and thermal radiation, and has important applications in strategic fields such as aerospace and nuclear power engineering. Research has shown that at high temperatures (>800°C), the efficiency of radiative heat transfer is significantly improved, with its contribution approaching a balance with that of thermal conduction. This has driven the development of novel thermal protection materials that combine high infrared radiation properties with low thermal conductivity. Rare earth tantalates (RE3TaO7) are promising candidates due to their low thermal conductivity (<2 W / m·K), excellent phase stability, and matching thermal expansion coefficient. However, their wide bandgap and single lattice vibrational mode result in insufficient infrared radiation performance (infrared emissivity <0.5). High-entropy design offers a new approach to controlling their infrared radiation properties through the synergistic effect of multiple principal components. However, the high-dimensional combinatorial space of multicomponents limits the efficiency of traditional trial-and-error methods. Recent advances in materials informatics offer new avenues for overcoming this bottleneck. Although machine learning has been successfully applied to similar property predictions, such as thermal conductivity prediction, rare earth tantalate systems still face key challenges: (1) the lack of infrared radiation performance databases; and (2) existing literature data are concentrated on a limited number of element combinations, forming a high-dimensional sparse dataset. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention provides a data-driven screening method and application for high-infrared radiation characteristics and high-entropy rare earth tantalate materials. The present invention constructs a random forest model and uses the model to quickly predict in the component space to achieve rapid screening of the infrared emissivity of RE3TaO7 materials.

[0004] A data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials comprises the following steps: S1: Create a rare earth tantalate RE3TaO7 infrared emissivity dataset: The infrared emissivity of a material with the chemical formula RE3TaO7 in the 2.5μm to 14μm band was measured and collected to establish a rare earth tantalate RE3TaO7 infrared emissivity dataset. In RE3TaO7, RE is a combination of one or more rare earth elements selected from the group consisting of La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y. Ta represents tantalum, and O represents oxygen. The stoichiometric ratio of RE:Ta:O is 3:1:7. S2: Establish an extended multi-source material infrared emissivity dataset: Measure and collect chemical formulas of RE2O3, RE3NbO7, RETa3O9, The infrared emissivity of the type materials in the 2.5μm~14μm band is used to establish an extended multi-source material infrared emissivity data set; RE is a rare earth element, Nb is niobium, Ta is tantalum, and O represents oxygen; the element stoichiometric ratio of RE:O in RE2O3 is 2:3; the element stoichiometric ratio of RE:Nb:O in RE3NbO7 is 3:1:7; the element stoichiometric ratio of RE:Ta:O in RETa3O9 is 1:3:9; in It is composed of a unit or a multi-component combination of rare earth elements La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y; It is composed of a unit or a multi-component combination of rare earth elements Ta, Nb, and Zr; O represents oxygen; middle : : The elemental stoichiometric ratio of O is X:Y:Z; S3: Implement feature engineering optimization: The element composition is used as the coding method for different materials, RE3TaO7, RE2O3, RE3NbO7, RETa3O9, The element properties corresponding to the chemical formula are processed and feature transformation is carried out; the specific steps of feature transformation are as follows: suppose there are n elements, and the compounds composed of them can be expressed as component vectors ,in is the stoichiometric coefficient of the i-th element; for p compounds, their composition can be constructed as a p×n-dimensional composition matrix A; at the same time, the physical properties of the elements are taken as the characteristics of the elements, recorded as initial characteristics, assuming that the elements have m initial characteristics, and the physical properties of each element are represented as characteristic vectors , then the physical property parameters of n elements form an n×m-dimensional characteristic matrix B; through the matrix multiplication operation C=AB, the p×m-dimensional compound physical property matrix C can be obtained; among them, the j-th row vector of the matrix C Represents the characteristics of the physical properties of the j-th compound; the composition matrix A is normalized and recorded as ,but The calculation results represent the weighted average values ​​of the physical property parameters of each compound; this operation transforms the compound composition matrix into the initial characteristic matrix; S4: Execute feature dimensionality reduction optimization process: First, the recursive feature elimination method is used to screen the optimal number of sub-features α from the initial feature matrix. Based on the Pearson correlation coefficient matrix between the initial features and combined with SHAP feature importance analysis, the initial features are reduced in dimension to obtain α features. The reduced-dimensionality feature vector is constructed based on the reduced-dimensional features. S5: Use the extended multi-source material infrared emissivity dataset established in S2 as the source domain data, use the dimension-reduced feature vector of each material in the source domain data as input, and the corresponding emissivity vector as output to build a random forest prediction model, which is recorded as the pre-training model; use the rare earth tantalate RE3TaO7 infrared emissivity dataset established in S1 as the target domain data; the materials of the source domain data and the target domain data use the same features; S6: First, the features of the source domain data and the target domain data are jointly normalized to align the feature distributions of the source and target domains. Then, a fine-tuning training dataset is determined based on the source and target domain samples. The fine-tuning dataset is used to fine-tune the pre-trained model to obtain the final prediction model. S7: Construct a composition data set based on the full permutation of the five-element combinations of RE3TaO7 of 12 rare earth elements, generate a feature matrix based on the reduced-dimensional feature vector corresponding to the optimal sub-feature number α screened out in S4, and predict the infrared emissivity using the final prediction model saved in S6. Output a sequence of candidate materials in ascending order of predicted values; S8: The composition of the high entropy and high emissivity rare earth tantalate RE3TaO7 selected by S7 is (Y 0.2 Dy 0.2 Gd 0.2 Sm 0.2 Nd 0.2 )3TaO7 and (Nd 0.2 Eu 0.2 Dy 0.2 Er 0.2 Yb 0.2 )3TaO7, experimentally prepare the material and verify its performance, thus completing a data-driven method for screening high-infrared radiation characteristic high-entropy rare earth tantalate materials.

[0005] Advantages of the present invention: This invention achieves the following innovative breakthroughs in the research and development of infrared radiation materials: First, by incorporating transfer learning technology to construct a high-precision prediction model, a novel quantitative mapping relationship between the compositional space of rare earth tantalate materials and their infrared radiation properties is established, enabling efficient screening of high-entropy RE3TaO7 materials. Compared to the traditional trial-and-error method, which suffers from limitations and long processing times, this invention employs a data-driven strategy to construct a parameter prediction model (infrared emissivity), providing precise theoretical guidance for experimental design, significantly shortening the R&D cycle and reducing experimental costs. To address the challenge of scarce infrared radiation property data for rare earth tantalate systems, this paper employs a transfer learning framework. This approach pre-trains a source-domain model on a large dataset of multiple material systems, including RE2O3, RE3NbO7, RETa3O9, and RETa3O9. It then adapts the model to the property predictions of the RE3TaO7 system through parameter transfer and fine-tuning. This technical approach significantly improves the prediction accuracy of the target system while maintaining cross-system generalization. In terms of implementation, the present invention constructs a phased optimization mechanism: Step S6 integrates multi-source material data to construct an extended training set, leveraging prior knowledge of the source-domain model to enhance basic feature extraction capabilities. Step S7 employs an adaptive parameter fusion algorithm to optimize the matching of the source-domain model parameters with the training parameters of the target-domain RE3TaO7 dataset, and dynamically adjusts feature weights to achieve precise adaptation of the model to the target material system. This hierarchical optimization approach significantly improves the prediction accuracy and stability of the final model compared to traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 It is a schematic diagram of the overall process of the present invention; Figure 2 Descriptor correlation scatter matrix diagram - histogram - Pearson coefficient diagram; Figure 3 The importance of elemental characteristics of each rare earth element and infrared emissivity; Figure 4 is the correlation between the element characteristics of each rare earth element and infrared emissivity; Figure 5 is (Y 0.2 Dy 0.2 Gd 0.2 Sm 0.2 Nd 0.2 )3TaO7 and (Nd 0.2 Eu 0.2 Dy 0.2 Er 0.2 Yb 0.2 )Schematic diagram comparing the actual and predicted values ​​of infrared emissivity of 3TaO7. DETAILED DESCRIPTION

[0007] Specific embodiment 1: This embodiment is a data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials, comprising the following steps: S1: Create a rare earth tantalate RE3TaO7 infrared emissivity dataset: The infrared emissivity of a material with the chemical formula RE3TaO7 in the 2.5μm to 14μm band was measured and collected to establish a rare earth tantalate RE3TaO7 infrared emissivity dataset. In RE3TaO7, RE is a combination of one or more rare earth elements selected from the group consisting of La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y. Ta represents tantalum, and O represents oxygen. The stoichiometric ratio of RE:Ta:O is 3:1:7. S2: Establish an extended multi-source material infrared emissivity dataset: Measure and collect chemical formulas of RE2O3, RE3NbO7, RETa3O9, A X B Y O Z The infrared emissivity of the type materials in the 2.5μm~14μm band is used to establish an extended multi-source material infrared emissivity data set; RE is a rare earth element, Nb is niobium, Ta is tantalum, and O represents oxygen; the element stoichiometric ratio of RE:O in RE2O3 is 2:3; the element stoichiometric ratio of RE:Nb:O in RE3NbO7 is 3:1:7; the element stoichiometric ratio of RE:Ta:O in RETa3O9 is 1:3:9; in It is composed of a unit or a multi-component combination of rare earth elements La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y; It is composed of a unit or a multi-component combination of rare earth elements Ta, Nb, and Zr; O represents oxygen; middle The element stoichiometric ratio is X:Y:Z; S3: Implement feature engineering optimization: The element composition is used as the coding method for different materials, RE3TaO7, RE2O3, RE3NbO7, RETa3O9, The element properties corresponding to the chemical formula are processed and feature transformation is carried out; the specific steps of feature transformation are as follows: suppose there are n elements, and the compounds composed of them can be expressed as component vectors ,in is the stoichiometric coefficient of the i-th element; for p compounds, their composition can be constructed as a p×n-dimensional composition matrix A; at the same time, the physical properties of the elements are taken as the characteristics of the elements, recorded as initial characteristics, assuming that the elements have m initial characteristics, and the physical properties of each element are represented as characteristic vectors , then the physical property parameters of n elements form an n×m-dimensional characteristic matrix B; through the matrix multiplication operation C=AB, the p×m-dimensional compound physical property matrix C can be obtained; among them, the j-th row vector of the matrix C Represents the characteristics of the physical properties of the j-th compound; the composition matrix A is normalized and recorded as ,but The calculation results represent the weighted average values ​​of the physical property parameters of each compound; this operation transforms the compound composition matrix into the initial characteristic matrix; S4: Execute feature dimensionality reduction optimization process: First, the recursive feature elimination method is used to screen the optimal number of sub-features α from the initial feature matrix. Based on the Pearson correlation coefficient matrix between the initial features and combined with SHAP feature importance analysis, the initial features are reduced in dimension to obtain α features. The reduced-dimensionality feature vector is constructed based on the reduced-dimensional features. S5: Use the extended multi-source material infrared emissivity dataset established in S2 as the source domain data, use the dimension-reduced feature vector of each material in the source domain data as input, and the corresponding emissivity vector as output to build a random forest prediction model, which is recorded as the pre-training model; use the rare earth tantalate RE3TaO7 infrared emissivity dataset established in S1 as the target domain data; the materials of the source domain data and the target domain data use the same features; S6: First, the features of the source domain data and the target domain data are jointly normalized to align the feature distributions of the source and target domains. Then, a fine-tuning training dataset is determined based on the source and target domain samples. The fine-tuning dataset is used to fine-tune the pre-trained model to obtain the final prediction model. S7: Construct a composition data set based on the full permutation of the five-element combinations of RE3TaO7 of 12 rare earth elements, generate a feature matrix based on the reduced-dimensional feature vector corresponding to the optimal sub-feature number α screened out in S4, and predict the infrared emissivity using the final prediction model saved in S6. Output a sequence of candidate materials in ascending order of predicted values; S8: The composition of the high entropy and high emissivity rare earth tantalate RE3TaO7 selected by S7 is (Y 0.2 Dy 0.2 Gd 0.2 Sm 0.2 Nd 0.2 )3TaO7 and (Nd 0.2 Eu 0.2 Dy 0.2 Er 0.2 Yb 0.2 )3TaO7, experimentally prepare the material and verify its performance, thus completing a data-driven method for screening high-infrared radiation characteristic high-entropy rare earth tantalate materials.

[0008] This embodiment is a data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials. It aims to break through the infrared radiation performance bottleneck of traditional rare earth tantalate ceramic materials. By constructing a multi-system common data set, a high-precision prediction model for the infrared emissivity of rare earth tantalate is innovatively developed, which realizes the rapid evaluation of the infrared spectral characteristics of A-site high-entropy rare earth tantalate RE3TaO7 materials. Based on this model, the following materials were successfully screened: (Y 0.2 Dy 0.2 Gd 0.2 Sm 0.2 Nd 0.2 )3TaO7 and (Nd 0.2 Eu 0.2 Dy 0.2 Er 0.2 Yb 0.2 )3TaO7, a multi-component rare earth tantalate system with excellent infrared radiation properties, has been experimentally verified. This method, through big data modeling, overcomes the limitations of traditional small-sample material research and provides an efficient theoretical design paradigm for the development of new high-entropy infrared radiation materials.

[0009] Specific Embodiment 2: This embodiment differs from Specific Embodiment 1 in that RE in RE3TaO7 described in S1 is a single, binary, ternary, quaternary, or quinary rare earth element selected from La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y. The infrared emissivity dataset for the rare earth tantalate RE3TaO7 established in S1 covers emissivity data for rare earth tantalates with combinations of single to quinary rare earth elements. Other steps are the same as in Specific Embodiment 1.

[0010] Specific embodiment 3: This embodiment differs from either specific embodiment 1 or 2 in that the RE in RE2O3, RE3NbO7, and RETa3O9 described in S2 is a single or multiple rare earth element selected from La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y. The extended multi-source material infrared emissivity dataset established in S2 covers rare earth oxides, rare earth niobates, and rare earth tantalates with single to five rare earth element combinations. The X:Y:Z ratio in S2 is 2:0:3, 2:2:7, 3:1:7, 1:1:4, 1:3:9, 2:2:7, or 1:1:4. Other steps are the same as those in specific embodiments 1 or 2.

[0011] Specific embodiment 4: This embodiment differs from specific embodiments 1 to 3 in that the physical properties of the element described in S3 include atomic weight, energy level, electronegativity, melting point, boiling point, electron affinity, first ionization energy, radius, hardness, modulus, density, thermal conductivity, electrical conductivity, heat of fusion, heat of vaporization, specific heat capacity, period, number of s-orbital electrons, number of p-orbital electrons, number of d-orbital electrons, and number of f-orbital electrons. The other steps are the same as specific embodiments 1 to 3.

[0012] Specific embodiment 5: This embodiment differs from specific embodiments 1 to 4 in that the process of reducing the dimension of the initial features to obtain α features based on the Pearson correlation coefficient matrix between the initial features and the SHAP feature importance analysis in S4 includes: (1) Use recursive feature elimination method to select the optimal number of sub-features; (2) Use Pearson correlation coefficient to analyze the correlation between different features, use 0.75 as the threshold for high correlation to select multiple pairs of features, and combine feature importance analysis to eliminate the features with low importance in each pair of highly correlated features; gradually eliminate the features with the lowest feature importance until the number of features reaches the optimal prediction accuracy, where the prediction accuracy is determined by the determination coefficient R 2 Judgment, when the number of features reaches 12, R 2 =0.88, the maximum value; the final features eliminated were: energy level, boiling point, hardness, modulus, density, conductivity, specific heat, period, and number of p-orbital electrons; the α features retained were: relative molecular mass, electronegativity, melting point, electron affinity, first ionization energy, average atomic radius, element thermal conductivity, heat of vaporization, heat of fusion, number of s-orbital electrons, number of d-orbital electrons, and number of f-orbital electrons; the feature importance score was: electronegativity > heat of vaporization > first ionization energy > melting point > number of d-electrons > element thermal conductivity > number of f-electrons > atomic radius > electron affinity > heat of fusion > relative molecular mass > number of s-electrons. The other steps were the same as in Specific Embodiments 1 to 4.

[0013] Specific embodiment 6: This embodiment differs from specific embodiments 1 to 5 in that the random forest model in S5 includes 100 decision trees, uses mean square error as the splitting criterion, sets the maximum number of features to the square root of the total number of features, and sets the random seed to 42. The other steps are the same as specific embodiments 1 to 5.

[0014] Specific implementation method seven: The difference between this implementation method and specific implementation methods one to six is ​​that the specific steps of S6 are as follows: ① Feature alignment: After merging the source domain data and the target domain data, use the source domain normalizer to jointly normalize the two to align the feature distribution of the source domain and the target domain; ② Sample weight allocation: Fix the source domain sample weight to 0.6 and the target domain sample weight to 1.0, and construct a sample weight vector; ③ Data merging: Merge the standardized source domain data and target domain data into a training set, and merge the output targets of the source domain and the target domain at the same time; ④ Model training: Use the merged data and sample weights to train the random forest model to complete the construction of the fine-tuning model; ⑤ Model saving and evaluation: Save the fine-tuned model, and evaluate the model performance on the target domain data, output the mean square error, determination coefficient, mean absolute error and mean absolute percentage error indicators to verify the predictive ability of the transfer learning model; Among them, the termination condition: The optimization process of transfer learning can be based on the target domain The performance indicators on the data are the mean square error and MSE as the criteria; when the performance indicators of the model on the target domain data reach the optimal level, that is, the mean square error is minimized or the performance indicators no longer improve significantly in multiple iterations, the optimization process can be terminated; the qualification standard: the introduced source domain data should improve the model performance on the target domain data and not cause negative transfer; specifically, the mean square error of the performance indicator of the model on the target domain data should be better than the baseline model trained only with the target domain data; if the model performance does not improve or decreases after the introduction of the source domain data, it is considered unqualified; unqualified processing: weight adjustment: by adjusting the weight of the source domain data, re-optimize the model to ensure that the source domain data has an appropriate impact on the target domain data; eliminate some source domain data: analyze the feature similarity of the source domain data and the target domain data, and eliminate the source domain data with large differences from the target domain data; reselect source domain data: if the above method is invalid, it is necessary to re-select source domain data that is more similar to the feature distribution of the target domain data and re-perform transfer learning training. The other steps are the same as the specific implementation methods one to six.

[0015] Specific embodiment eight: The difference between this embodiment and specific embodiments one to seven is that: the 12 rare earth elements described in S7 include La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y; when constructing a composition data set based on the full permutation of the five-element combination of RE3TaO7 of the 12 rare earth elements in S7, the five-element combination is screened from the 12 rare earth elements, and the doping concentration range of each element is 0-1. The composition gradient is constructed with an incremental step size of 0.1, and the infrared emissivity is predicted by the final prediction model and arranged in ascending order; the SHAP method is used for interpretable machine learning analysis, and the correlation between element composition and emissivity is constructed through feature transformation, and elements with positive correlation with emissivity are screened, from large to small: Nd, Eu, Sm, Yb, and Tm; elements with negative correlation with emissivity are, from large to small: Y, Gd, Ho, and Ce; and elements with no obvious correlation are: Er, Dy, and La; further screening of suitable elements improves the emissivity performance of the material. The other steps are the same as those in Specific Embodiments 1 to 7.

[0016] Specific Embodiment 9: This embodiment differs from Specific Embodiments 1 to 8 in that the experimental preparation process described in S8 includes: weighing raw material powders according to stoichiometric weights, mixing in a planetary ball mill for 12 hours, drying, passing through a 500-mesh sieve, pressing into a Φ15×1 mm green billet, sintering at 1700°C for 10 hours to obtain a dense RE3TaO7 ceramic, and finally testing infrared emissivity. Other steps are the same as Specific Embodiments 1 to 8.

[0017] Specific embodiment ten: This embodiment is the application of the method in the rapid prediction and screening of high infrared radiation characteristics of rare earth tantalates.

[0018] The following examples are used to verify the beneficial effects of the present invention: Example 1: A data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials, comprising the following steps: S1: Create a rare earth tantalate RE3TaO7 infrared emissivity dataset: Measure and collect the infrared emissivity of materials with the chemical formula RE3TaO7 in the 2.5μm-14μm band to establish a rare earth tantalate RE3TaO7 infrared emissivity dataset (the data is derived from data disclosed in the prior art, if the prior art does not disclose independent experimental and test results). In the RE3TaO7, RE is a combination of one or more rare earth elements selected from La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y, Ta represents tantalum, O represents oxygen, and the stoichiometric ratio of RE:Ta:O is 3:1:7. S2: Establish an extended multi-source material infrared emissivity dataset: Measure and collect chemical formulas of RE2O3, RE3NbO7, RETa3O9, The infrared emissivity of the type materials in the 2.5μm~14μm band is used to establish an extended multi-source material infrared emissivity data set (the data comes from the data disclosed by the prior art, such as the independent experimental and test results that are not disclosed by the prior art); RE is a rare earth element, Nb is niobium, Ta is tantalum, and O represents oxygen; the elemental stoichiometric ratio of RE:O in RE2O3 is 2:3; the elemental stoichiometric ratio of RE:Nb:O in RE3NbO7 is 3:1:7; the elemental stoichiometric ratio of RE:Ta:O in RETa3O9 is 1:3:9; in It is composed of a unit or a multi-component combination of rare earth elements La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y; It is composed of a unit or a multi-component combination of rare earth elements Ta, Nb, and Zr; O represents oxygen; middle The element stoichiometric ratio is X:Y:Z; S3: Implement feature engineering optimization: The element composition is used as the coding method for different materials, RE3TaO7, RE2O3, RE3NbO7, RETa3O9, The element properties corresponding to the chemical formula are processed and feature transformation is carried out; the specific steps of feature transformation are as follows: suppose there are n elements, and the compounds composed of them can be expressed as component vectors ,in is the stoichiometric coefficient of the i-th element; for p compounds, their composition can be constructed as a p×n-dimensional composition matrix A; at the same time, the physical properties of the elements are taken as the characteristics of the elements, recorded as initial characteristics, assuming that the elements have m initial characteristics, and the physical properties of each element are represented as characteristic vectors , then the physical property parameters of n elements form an n×m-dimensional characteristic matrix B; through the matrix multiplication operation C=AB, the p×m-dimensional compound physical property matrix C can be obtained; among them, the j-th row vector of the matrix C Represents the characteristics of the physical properties of the j-th compound; the composition matrix A is normalized and recorded as ,but The calculation results represent the weighted average values ​​of the physical property parameters of each compound; this operation transforms the compound composition matrix into the initial characteristic matrix; S4: Execute feature dimensionality reduction optimization process: First, the recursive feature elimination method is used to screen the optimal sub-feature number α from the initial feature matrix. Based on the Pearson correlation coefficient matrix between the initial features and combined with the SHAP feature importance analysis, the initial features are reduced in dimension to obtain α features. The reduced-dimensional feature vector is constructed based on the reduced-dimensional features. The correlation scatter matrix diagram - histogram - Pearson coefficient diagram is shown in the figure below. Figure 2 As shown; S5: Use the extended multi-source material infrared emissivity dataset established in S2 as the source domain data, use the dimension-reduced feature vector of each material in the source domain data as input, and the corresponding emissivity vector as output to build a random forest prediction model, which is recorded as the pre-training model; use the rare earth tantalate RE3TaO7 infrared emissivity dataset established in S1 as the target domain data; the materials of the source domain data and the target domain data use the same features; S6: First, the features of the source domain data and the target domain data are jointly normalized to align the feature distributions of the source and target domains. Then, a fine-tuning training dataset is determined based on the source and target domain samples. The fine-tuning dataset is used to fine-tune the pre-trained model to obtain the final prediction model. S7: Construct a composition data set based on the full permutation of the five-element combinations of RE3TaO7 of 12 rare earth elements, generate a feature matrix based on the reduced-dimensional feature vector corresponding to the optimal sub-feature number α screened out in S4, and predict the infrared emissivity using the final prediction model saved in S6. Output a sequence of candidate materials in ascending order of predicted values; S8: The composition of the high entropy and high emissivity rare earth tantalate RE3TaO7 selected by S7 is (Y 0.2 Dy 0.2 Gd 0.2 Sm 0.2 Nd 0.2 )3TaO7 and (Nd 0.2 Eu 0.2 Dy 0.2 Er 0.2 Yb 0.2 )3TaO7, the material was experimentally prepared and the performance was verified. The infrared emissivity test results of the two ceramics were compared with the predicted results as shown in the figure Figure 5 As shown; that is, a data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials is completed; Specifically: RE in RE3TaO7 described in S1 is a single, binary, ternary, quaternary or quinary rare earth element among La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb and Y; the infrared emissivity data set of rare earth tantalate RE3TaO7 established by S1 covers the emissivity data of rare earth tantalates with combinations of one to five rare earth elements.

[0019] Specifically: RE in RE2O3, RE3NbO7, and RETa3O9 described in S2 is a single unit or a combination of multiple rare earth elements La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y; the extended multi-source material infrared emissivity data set established in S2 covers rare earth oxides, rare earth niobates, and rare earth tantalates of single to five rare earth element combinations; X:Y:Z described in S2 is 2:0:3, 2:2:7, 3:1:7, 1:1:4, 1:3:9, 2:2:7, or 1:1:4.

[0020] Specifically: The physical properties of the element described in S3 include atomic weight, energy level, electronegativity, melting point, boiling point, electron affinity, first ionization energy, radius, hardness, modulus, density, thermal conductivity, electrical conductivity, heat of fusion, heat of vaporization, specific heat capacity, period, number of s-orbital electrons, number of p-orbital electrons, number of d-orbital electrons, and number of f-orbital electrons.

[0021] Specifically: In S4, based on the Pearson correlation coefficient matrix between the initial features and combined with SHAP feature importance analysis, the process of reducing the dimension of the initial features to obtain α features includes: (1) Use recursive feature elimination method to select the optimal number of sub-features; (2) Use Pearson correlation coefficient to analyze the correlation between different features, use 0.75 as the threshold for high correlation to select multiple pairs of features, and combine feature importance analysis to eliminate the features with low importance in each pair of highly correlated features; gradually eliminate the features with the lowest feature importance until the number of features reaches the optimal prediction accuracy, where the prediction accuracy is determined by the determination coefficient R 2 Judgment, when the number of features reaches 12, R 2 =0.88, which is the maximum value; the final features eliminated are: energy level, boiling point, hardness, modulus, density, electrical conductivity, specific heat, period, and p-orbital electron number; the α features retained are: relative molecular mass, electronegativity, melting point, electron affinity, first ionization energy, average atomic radius, element thermal conductivity, heat of vaporization, heat of fusion, s-orbital electron number, d-orbital electron number, and f-orbital electron number; the feature importance score results are: electronegativity > heat of vaporization > first ionization energy > melting point > d-electron number > element thermal conductivity > f-electron number > atomic radius > electron affinity > fusion heat > relative molecular mass > s-electron number.

[0022] Specifically, the random forest model in S5 contains 100 decision trees, uses the mean square error as the splitting criterion, sets the maximum number of features to the square root of the total number of features, and sets the random seed to 42.

[0023] Specifically: The specific steps of S6 are as follows: ① Feature alignment: After merging the source domain data and the target domain data, use the source domain normalizer to jointly normalize the two to align the feature distribution of the source domain and the target domain; ② Sample weight allocation: Fix the source domain sample weight to 0.6 and the target domain sample weight to 1.0, and construct a sample weight vector; ③ Data merging: Merge the standardized source domain data and target domain data into a training set, and merge the output targets of the source domain and the target domain; ④ Model training: Use the merged data and sample weights to train the random forest model to complete the construction of the fine-tuning model; ⑤ Model saving and evaluation: Save the fine-tuned model, and evaluate the model performance on the target domain data, output the mean square error, determination coefficient, mean absolute error, and mean absolute percentage error indicators to verify the predictive ability of the transfer learning model.

[0024] Specifically: the 12 rare earth elements described in S7 include La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y; when constructing a composition data set based on the full permutation of the five-element combination of RE3TaO7 of the 12 rare earth elements in S7, the five-element combination is screened from the 12 rare earth elements, the doping concentration range of each element is 0-1, and the composition gradient is constructed with an incremental step of 0.1. The infrared emissivity is predicted by the final prediction model and arranged in ascending order; the SHAP method is used for interpretable machine learning analysis, and the correlation between element composition and emissivity is constructed through feature transformation, and the elements positively correlated with emissivity are screened, from large to small: Nd, Eu, Sm, Yb, and Tm; the elements negatively correlated with emissivity are, from large to small: Y, Gd, Ho, and Ce; and the elements with no obvious correlation are: Er, Dy, and La; the element feature importance and feature correlation of each rare earth element with infrared emissivity are as follows Figure 3 and 4 As shown; Specifically, the experimental preparation process described in S8 includes: weighing the raw material powder according to the stoichiometric amount, mixing it in a planetary ball mill for 12 hours and then drying it, passing it through a 500-mesh sieve and pressing it into a Φ15×1 mm green billet, sintering it at 1700°C for 10 hours to obtain a dense RE3TaO7 ceramic, and finally testing the infrared emissivity.

[0025] Figure 5 The experimental results show that these two materials have high emissivity in the infrared band, with average emissivity of 0.88 and 0.85 respectively, and the predicted results are 0.87 and 0.84.

[0026] In summary, the present invention constructs a high-precision rare earth tantalate infrared emissivity prediction model through transfer learning technology, successfully achieving rapid performance evaluation of multi-component RE3TaO7 materials. This technology has fundamentally resolved the constraints of small sample data sets on machine learning models. Through a knowledge transfer strategy across material systems, the common characteristic patterns acquired during the pre-training phase are effectively transferred to the target system, significantly improving prediction efficiency. Compared to traditional trial-and-error methods that rely on repeated experiments, the present invention makes the screening of high-infrared emissivity rare earth tantalate systems more convenient and cost-effective by constructing a quantitative prediction relationship between composition and performance.

Claims

1. A data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials, characterized by: The following steps are involved: S1: Create a rare earth tantalate RE3TaO7 infrared emissivity dataset: The infrared emissivity of a material with the chemical formula RE3TaO7 in the 2.5μm to 14μm band was measured and collected to establish a rare earth tantalate RE3TaO7 infrared emissivity dataset. In RE3TaO7, RE is a combination of one or more rare earth elements selected from the group consisting of La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y. Ta represents tantalum, and O represents oxygen. The stoichiometric ratio of RE:Ta:O is 3:1:

7. S2: Establish an extended multi-source material infrared emissivity dataset: Measure and collect chemical formulas of RE2O3, RE3NbO7, RETa3O9, The infrared emissivity of the type materials in the 2.5μm~14μm band is used to establish an extended multi-source material infrared emissivity data set; RE is a rare earth element, Nb is niobium, Ta is tantalum, and O represents oxygen; the element stoichiometric ratio of RE:O in RE2O3 is 2:3; the element stoichiometric ratio of RE:Nb:O in RE3NbO7 is 3:1:7; the element stoichiometric ratio of RE:Ta:O in RETa3O9 is 1:3:9; in It is composed of a unit or a multi-component combination of rare earth elements La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y; It is composed of a unit or a multi-component combination of rare earth elements Ta, Nb, and Zr; O represents oxygen; middle The element stoichiometric ratio is X:Y:Z; S3: Implement feature engineering optimization: The element composition is used as the coding method for different materials, RE3TaO7, RE2O3, RE3NbO7, RETa3O9, Data processing of element properties corresponding to chemical formulas and feature transformation; The specific steps of feature transformation are: suppose there are n elements, and the compounds composed of them can be expressed as component vectors ,in is the stoichiometric coefficient of the i-th element; for p compounds, their composition can be constructed as a p×n-dimensional composition matrix A; at the same time, the physical properties of the elements are taken as the characteristics of the elements, recorded as initial characteristics, assuming that the elements have m initial characteristics, and the physical properties of each element are represented as characteristic vectors , then the physical property parameters of n elements form an n×m-dimensional characteristic matrix B; through the matrix multiplication operation C=AB, the p×m-dimensional compound physical property matrix C can be obtained; among them, the j-th row vector of the matrix C Represents the characteristics of the physical properties of the j-th compound; the composition matrix A is normalized and recorded as ,but The calculation results represent the weighted average values ​​of the physical property parameters of each compound; this operation transforms the compound composition matrix into the initial characteristic matrix; S4: Execute feature dimensionality reduction optimization process: First, the recursive feature elimination method is used to screen the optimal number of sub-features α from the initial feature matrix. Based on the Pearson correlation coefficient matrix between the initial features and combined with SHAP feature importance analysis, the initial features are reduced in dimension to obtain α features. The reduced-dimensionality feature vector is constructed based on the reduced-dimensional features. S5: Use the extended multi-source material infrared emissivity dataset established in S2 as the source domain data, use the dimension-reduced feature vector of each material in the source domain data as input, and the corresponding emissivity vector as output to build a random forest prediction model, which is recorded as the pre-training model; use the rare earth tantalate RE3TaO7 infrared emissivity dataset established in S1 as the target domain data; the materials of the source domain data and the target domain data use the same features; S6: First, the features of the source domain data and the target domain data are jointly normalized to align the feature distributions of the source and target domains. Then, a fine-tuning training dataset is determined based on the source and target domain samples. The fine-tuning dataset is used to fine-tune the pre-trained model to obtain the final prediction model. S7: Construct a composition data set based on the full permutation of the five-element combinations of RE3TaO7 of 12 rare earth elements, generate a feature matrix based on the reduced-dimensional feature vector corresponding to the optimal sub-feature number α screened out in S4, and predict the infrared emissivity using the final prediction model saved in S6. Output a sequence of candidate materials in ascending order of predicted values; S8: The composition of the high entropy and high emissivity rare earth tantalate RE3TaO7 selected by S7 is (Y 0.2 Dy 0.2 Gd 0.2 Sm 0.2 Nd 0.2 )3TaO7 and (Nd 0.2 Eu 0.2 Dy 0.2 Er 0.2 Yb 0.2 )3TaO7, experimentally prepare the material and verify its performance, thus completing a data-driven method for screening high-infrared radiation characteristic high-entropy rare earth tantalate materials.

2. The data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials according to claim 1 is characterized in that In RE3TaO7 described in S1, RE is the single, binary, ternary, quaternary and quinary rare earth elements La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb and Y; the infrared emissivity data set of rare earth tantalate RE3TaO7 established by S1 covers the emissivity data of rare earth tantalates with combinations of single to quinary rare earth elements.

3. A data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials according to claim 1, characterized in that the RE in RE2O3, RE3NbO7, and RETa3O9 described in S2 is a single or multiple combination of rare earth elements La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y; the extended multi-source material infrared emissivity data set established in S2 covers rare earth oxides, rare earth niobates, and rare earth tantalates with single to five rare earth element combinations; and the X:Y:Z described in S2 is 2:0:3, 2:2:7, 3:1:7, 1:1:4, 1:3:9, 2:2:7, or 1:1:

4.

4. The data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials according to claim 1 is characterized in that The physical properties of the elements described in S3 include atomic weight, energy level, electronegativity, melting point, boiling point, electron affinity, first ionization energy, radius, hardness, modulus, density, thermal conductivity, electrical conductivity, heat of fusion, heat of vaporization, specific heat capacity, period, number of s-orbital electrons, number of p-orbital electrons, number of d-orbital electrons, and number of f-orbital electrons.

5. The data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials according to claim 4 is characterized in that In S4, based on the Pearson correlation coefficient matrix between the initial features and combined with SHAP feature importance analysis, the process of reducing the dimension of the initial features to obtain α features includes: (1) Use recursive feature elimination method to select the optimal number of sub-features; (2) Use Pearson correlation coefficient to analyze the correlation between different features, use 0.75 as the threshold for high correlation to select multiple pairs of features, and combine feature importance analysis to eliminate the features with low importance in each pair of highly correlated features; gradually eliminate the features with the lowest feature importance until the number of features reaches the optimal prediction accuracy, where the prediction accuracy is determined by the determination coefficient R 2 Judgment, when the number of features reaches 12, R 2 =0.88, which is the maximum value; the final features eliminated are: energy level, boiling point, hardness, modulus, density, electrical conductivity, specific heat, period, and p-orbital electron number; the α features retained are: relative molecular mass, electronegativity, melting point, electron affinity, first ionization energy, average atomic radius, element thermal conductivity, heat of vaporization, heat of fusion, s-orbital electron number, d-orbital electron number, and f-orbital electron number.

6. The data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials according to claim 1 is characterized in that The random forest model in S5 contains 100 decision trees, uses the mean square error as the splitting criterion, sets the maximum number of features to the square root of the total number of features, and sets the random seed to 42.

7. The data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials according to claim 1 is characterized in that The specific steps of S6 are as follows: ① Feature alignment: After merging the source domain data and the target domain data, use the source domain normalizer to jointly normalize the two to align the feature distributions of the source domain and the target domain; ② Sample weight allocation: Fix the source domain sample weight to 0.6 and the target domain sample weight to 1.0, and construct a sample weight vector; ③ Data merging: Merge the standardized source domain data and target domain data into a training set, that is, fine-tune the training data set, and merge the output targets of the source domain and the target domain; ④ Model training: Use the merged data and sample weights to train the random forest model to complete the construction of the fine-tuning model; ⑤ Model saving and evaluation: Save the fine-tuned model and evaluate the model performance on the target domain data, outputting the mean square error, determination coefficient, mean absolute error, and mean absolute percentage error indicators to verify the predictive ability of the transfer learning model.

8. The data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials according to claim 1 is characterized in that The 12 rare earth elements described in S7 include La, Ce, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, and Y. When constructing a composition data set based on the full permutation of five-element combinations of RE3TaO7 of the 12 rare earth elements in S7, five-element combinations are screened from the 12 rare earth elements, the doping concentration range of each element is 0-1, and a composition gradient is constructed with an incremental step size of 0.

1. The infrared emissivity is predicted by the final prediction model and arranged in ascending order.

9. The data-driven method for screening high-infrared radiation characteristics and high-entropy rare earth tantalate materials according to claim 1 is characterized in that The experimental preparation process described in S8 includes: weighing the raw material powder according to the stoichiometric amount, mixing it in a planetary ball mill for 12 hours and then drying it, passing it through a 500-mesh sieve and pressing it into a Φ15×1 mm green billet, sintering it at 1700℃ for 10 hours to obtain a dense RE3TaO7 ceramic, and finally testing its infrared emissivity.

10. Use of the method according to any one of claims 1 to 9 in the rapid prediction and screening of high infrared radiation characteristics of rare earth tantalates.

Images