An XGBoost-based halide perovskite band gap prediction method and system

Abstract

The application discloses a halide perovskite band gap prediction method and system based on XGBoost, which comprises the following steps: for ABX3 type and derivative structure halide perovskite, collecting original feature data and carrying out multi-stage preprocessing to obtain an optimized target feature set; based on the target feature set, a machine learning training database containing intrinsic structure parameters of the halide perovskite is constructed; a multi-dimensional feature analysis method combining correlation analysis, explainability analysis and visual comparative analysis is used to analyze the features of the machine learning training database to obtain a final training data set; a band gap prediction model is constructed based on the XGBoost algorithm, the band gap prediction model is optimized through GridSearchCV combined with 10-fold cross-validation, and the optimized band gap prediction model is used to realize the prediction of the halide perovskite band gap. The application can complete the prediction of the band gap, has the advantages of fast calculation speed, high accuracy and low cost.

Description

Technical Field

[0001] This invention relates to a method and system for predicting the bandgap of halide perovskites based on XGBoost, belonging to the field of perovskite solar cell material technology. Background Technology

[0002] In recent years, halide perovskites (ABX3 type and derived structures) have become popular research materials in fields such as photoelectric conversion, catalysis, and energy storage due to their excellent photoelectric properties, and are widely used in the research and development of devices such as solar cells, photodetectors, and light-emitting diodes. Compared with traditional oxide perovskites, halide perovskites have outstanding advantages such as high light absorption coefficient, excellent carrier mobility, and wide tunable bandgap. Among them, the bandgap is the core characterization parameter of the electronic structure of halide perovskites. By precisely controlling the bandgap, the same series of halide perovskites can be efficiently adapted to the needs of different application scenarios such as photovoltaic conversion, photoluminescence, and photoelectric detection. Therefore, accurate prediction of the bandgap is of great significance for the research and development and industrialization of halide perovskite materials.

[0003] Currently, the traditional method for calculating the band gap of halide perovskites is mainly high-throughput density functional theory (DFT). However, this method has significant drawbacks in practical applications and cannot meet the high-efficiency requirements of materials research and development.

[0004] Firstly, the computational efficiency is extremely low. The bandgap calculation of single-component halide perovskites takes several hours to several days. For high-throughput material screening scenarios involving multi-component doping and structural derivation (which typically need to cover thousands to tens of thousands of systems), the time cost of the DFT method has exceeded the affordability of actual research and development.

[0005] Secondly, it is difficult to balance accuracy and computational cost. When using low-precision functionals for calculation, the bandgap prediction error is large, which cannot provide a reliable basis for material design. While using high-precision functionals can improve the prediction accuracy, the computational cost increases exponentially and the hardware computing power requirements are high, which limits its promotion and application in large-scale research and development. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method and system for predicting the bandgap of halide perovskites based on XGBoost. It directly utilizes the intrinsic structural parameters of halide perovskites to construct a machine learning prediction model, trains the machine learning model, and completes the bandgap prediction. It features fast computation speed, high accuracy, and low cost.

[0007] To achieve the above objectives, this invention employs an XGBoost-based halide perovskite bandgap prediction method, comprising the following steps:

[0008] S1. For ABX3 type and derived halide perovskites, collect raw feature data and perform multi-stage preprocessing to obtain the optimized target feature set.

[0009] S2. Based on the target feature set, construct a machine learning training database containing intrinsic structural parameters of halide perovskites;

[0010] S3. Using a multi-dimensional feature analysis method that combines correlation analysis, interpretability analysis, and visual comparative analysis, the features of the machine learning training database are analyzed to obtain the final training dataset.

[0011] S4. A bandgap prediction model is constructed based on the XGBoost algorithm. The bandgap prediction model is optimized by combining GridSearchCV with 10-fold cross-validation. The optimized bandgap prediction model is used to predict the bandgap of halide perovskites.

[0012] As an improvement, in step S1, the original feature data is collected and subjected to multi-stage preprocessing to obtain an optimized target feature set, specifically including:

[0013] S101. Use the MP-API interface provided by Pymatgen to call the Materials Project database, collect the original feature data of halide perovskite, and delete invalid features, constant features and fixed column data.

[0014] S102. Use the ColumnTransformer tool to process numerical features and categorical features respectively. For numerical features, first fill missing values ​​with the median and then perform standardization; for categorical features, fill missing values ​​with the mode.

[0015] S103. Perform feature filtering sequentially to remove near-constant features with variances approaching 0; calculate the correlation coefficient matrix of numerical features and remove redundant variables from features with correlation coefficients > 0.9; use the SelectKBest algorithm to filter out the features most strongly correlated with the bandgap target variable to form the target feature set.

[0016] As an improvement, in step S2, the intrinsic structural parameters include chemical formula, material type, whether there is a direct band gap, A / B / X atoms, radius R_A / R_B / R_X, electronegativity Chi_A / Chi_B / Chi_X, tolerance factor, octahedral factor, electronegativity difference, crystal system, bond length, density, convex hull energy, formation energy, and total magnetization.

[0017] As an improvement, in step S3, the correlation analysis is performed by using a correlation heatmap to calculate the Pearson correlation coefficient of each feature, quantify the strength of the linear association between features, and help eliminate redundant features.

[0018] The interpretability analysis uses SHAP plots to calculate the SHAP value of each feature on the bandgap prediction result, clarifies the positive or negative contribution of each feature to the output of the bandgap prediction model, and sorts them to obtain the core features that affect the bandgap.

[0019] The visualization and comparative analysis uses radar charts to visually display the differences and patterns in the characteristic distribution of different halide perovskite samples across the intrinsic structural parameters.

[0020] As an improvement, step S4 specifically includes:

[0021] S401. Split the final training dataset obtained in step S3 into a training set and a test set in an 80%:20% ratio. Construct a bandgap prediction model based on the XGBoost algorithm. Optimize the model using GridSearchCV with 10-fold cross-validation to achieve the desired R-value for the optimized bandgap prediction model. 2 =0.863, mean square error MSE=0.611;

[0022] S402. Obtain the original characteristic data of the halide perovskite to be predicted, and input it into the optimized bandgap prediction model after multi-stage preprocessing in step S1. The model outputs the bandgap prediction value to complete the bandgap prediction.

[0023] As an improvement, the formula for the XGBoost algorithm is:

[0024] ;

[0025] in, Let K be the predicted value for the i-th sample, and K be the total number of decision trees. The k-th tree pairs samples The output score, Let be the function space for all regression trees.

[0026] As an improvement, in the ABX3 type halide perovskite, the A-site is a monovalent cation, such as an alkali metal ion Cs. + Organic small molecule ammonium ion MA + (Methylammonium), FA + (Formamidinium); the B-site is a divalent metal cation, such as Ge. 2+ Sn 2+ Pb 2+ The X position is a monovalent halide anion, such as Cl. - ,Br - I - .

[0027] A second aspect of the present invention also provides an XGBoost-based halide perovskite bandgap prediction system for implementing the aforementioned XGBoost-based halide perovskite bandgap prediction method, comprising:

[0028] The data preprocessing module is used to collect raw feature data and perform multi-stage preprocessing on ABX3 type and derived halide perovskites to obtain an optimized target feature set.

[0029] A database construction module is used to construct a machine learning training database containing intrinsic structural parameters of halide perovskites based on the target feature set.

[0030] The feature analysis module is used to analyze the features of the machine learning training database by employing a multi-dimensional feature analysis method that combines correlation analysis, interpretability analysis, and visual comparative analysis to obtain the final training dataset.

[0031] The model prediction module is used to build a bandgap prediction model based on the XGBoost algorithm. The bandgap prediction model is optimized by combining GridSearchCV with 10-fold cross-validation. The optimized bandgap prediction model is used to predict the bandgap of halide perovskites.

[0032] As an improvement, a data storage module is also included, which stores the original feature data, target feature set, machine learning training database, optimized bandgap prediction model parameters and bandgap prediction results, and supports data traceability and secondary retrieval.

[0033] Mechanism of the invention:

[0034] Based on the inherent physical law that the band gap value of ABX3 type and derived halide perovskites is jointly regulated by the intrinsic crystal structure, elemental composition and physicochemical parameters of the material, we construct an intelligent band gap prediction mechanism by utilizing the complex nonlinear correlation between various intrinsic structural parameters of halide perovskites and the band gap.

[0035] The band gap of halide perovskites is inherently determined by a combination of microscopic intrinsic parameters, including ion species, ionic radius, electronegativity, lattice constant, formation energy, bond length, and bond angle. First-principles calculations, bulk models, or cluster models rely on complex crystal configurations and iterative quantum chemical simulations, resulting in cumbersome modeling processes, high atomic computation requirements, long processing times, and high computational costs. Furthermore, these methods only achieve black-box numerical fitting and cannot analyze the contribution and direction of each structural parameter's influence on the band gap. This invention, however, does not rely on artificially constructed cluster models or require quantitative software for structural optimization and verification. It directly collects raw characteristic data of halide perovskites in batches from open-source materials databases. Through a multi-stage data preprocessing mechanism involving filling missing values ​​based on categorical and numerical features, standardization transformation, near-constant feature removal, high collinearity redundant feature removal, and correlation-based feature selection, it eliminates data noise, removes dimensional bias, suppresses feature multicollinearity, and selects core intrinsic features strongly correlated with the band gap. This provides a standardized input dataset with high signal-to-noise ratio and high correlation for model learning.

[0036] Meanwhile, this invention employs a multi-dimensional feature analysis mechanism integrating correlation heatmap analysis, SHAP interpretability analysis, and radar chart visualization comparison analysis: It uses Pearson correlation coefficient to quantify the strength of linear correlations between features, further eliminating redundant features; it quantitatively characterizes the correlation between each intrinsic structural parameter through SHAP values, overcoming the black-box defect of traditional machine learning models that cannot be interpreted; and it utilizes radar charts to represent the differences between different models in multiple dimensions. Finally, this invention uses the XGBoost gradient boosting decision tree ensemble learning algorithm as its core, relying on the hyperparameter iterative optimization mechanism of GridSearchCV combined with ten-fold cross-validation to automatically traverse and optimize model hyperparameters, effectively suppressing model overfitting and underfitting, and accurately learning and solidifying the nonlinear mapping rules between halide perovskite intrinsic structural parameters and band gap; for halide perovskites with unknown components and unknown structures to be tested, only the same standardized preprocessing procedure is needed to input feature parameters, and the band gap prediction value can be directly inferred and output through the trained optimal model.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] (1) It can replace the traditional high-throughput density functional theory (DFT) calculation method. Even for multi-doped and structure-derived halide perovskite systems, it can quickly complete the band gap prediction, greatly reduce the R&D cycle and time cost, and meet the high-throughput screening needs of industry and academia for halide perovskite materials.

[0039] (2) Based on the characteristics of ABX3 type and derived halide perovskites, through multi-stage preprocessing (invalid feature removal, missing value filling, and redundant feature screening) and multi-dimensional feature analysis (correlation analysis, interpretability analysis, and visualization comparison analysis), core features strongly correlated with the band gap are screened out. Combined with the strong fitting ability and parameter optimization mechanism of the XGBoost algorithm, the irreconcilable contradiction between the large prediction error of low-precision functionals and the high computational cost of high-precision functionals in the traditional DFT method is effectively solved, ensuring the accuracy and reliability of the band gap prediction results and providing accurate data support for material design.

[0040] (3) For ABX3 type and derived halide perovskites, the database construction and feature analysis are based on the intrinsic structural parameters of this type of material. Compared with general machine learning bandgap prediction methods, it avoids the defects of poor adaptability and weak generalization ability of general features. It can accurately match the structural characteristics and bandgap correlation of halide perovskites, and adapt to different composition (A / B / X site ion substitution) and different derived structure halide perovskite bandgap prediction scenarios, with a wider range of applications.

[0041] (4) Through multi-dimensional feature analysis methods (especially interpretability analysis and visualization comparison analysis), not only can the band gap be predicted quickly, but the correlation mechanism between each structural feature and the band gap can also be clarified, and the feature distribution differences of different samples can be displayed intuitively. This breaks through the limitations of traditional prediction models and can provide clear guidance for the structural optimization and composition control of halide perovskites, thus accelerating the research and development and industrialization of high-performance halide perovskite materials.

[0042] (5) Based on existing databases (Materials Project) and mature tools (XGBoost algorithm, feature analysis tools), it does not rely on expensive hardware such as high-end GPU clusters. The data acquisition and preprocessing process is standardized, and the model training and prediction steps are simple and efficient. Technical personnel in the relevant field can quickly reproduce it, which lowers the technical application threshold and facilitates its promotion and application in academia and industry. Attached Figure Description

[0043] Figure 1 This is a flowchart of the method of the present invention;

[0044] Figure 2 The coefficient of determination (Ri) between the bandgap prediction model of this invention and existing technologies is given. 2 Comparison chart;

[0045] Figure 3 A correlation heatmap of the intrinsic structural parameters of the halide perovskite of this invention;

[0046] Figure 4 Radar comparison diagram of intrinsic structure parameters of different halide perovskite samples of the present invention;

[0047] Figure 5 This is the SHAP interpretability analysis diagram of the present invention;

[0048] Figure 6 This is a training and fitting curve diagram of the bandgap prediction model of the present invention;

[0049] Figure 7 This is a comparison chart of the actual values ​​of six perovskites and the predicted values ​​of this invention. Detailed Implementation

[0050] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations thereof. In the absence of conflict, the embodiments and technical features in the embodiments can be combined with each other.

[0051] This invention provides an XGBoost-based method for predicting the band gap of halide perovskites, applicable to ABX3 type and derived halide perovskites. Through standardized data processing, multi-dimensional feature analysis, and machine learning model construction, this invention achieves rapid and accurate prediction of the band gap of halide perovskites, offering advantages such as low computational cost, fast prediction speed, strong model interpretability, and high accuracy.

[0052] like Figure 1 As shown, a method for predicting the bandgap of halide perovskites based on XGBoost is described, with the following specific steps:

[0053] S1. For ABX3 type and derived halide perovskites, raw feature data are collected and multi-stage preprocessing is performed to obtain an optimized target feature set; specifically including:

[0054] S101. Use the MP-API interface provided by Pymatgen to call the MaterialsProject database, collect raw feature data of halide perovskites in batches, perform preliminary cleaning of the raw data, delete invalid features, constant features and fixed column data to reduce data redundancy.

[0055] S102. Use the ColumnTransformer tool to differentiate between numerical and categorical features: for numerical features, use the median to fill missing values ​​and perform standardization; for categorical features, use the mode to fill missing values ​​to ensure data integrity and standardization.

[0056] S103. Perform multi-level feature screening: remove near-constant features with variance approaching 0; calculate the correlation coefficient matrix of numerical features and eliminate redundant variables in features with correlation coefficients greater than 0.9; use the SelectKBest algorithm with f_regression as the scoring function to select the top 80% of features that are correlated with the bandgap target variable, and finally form the optimized target feature set.

[0057] S2. Based on the target feature set, construct a machine learning training database containing intrinsic structural parameters of halide perovskites;

[0058] The intrinsic structural parameters include chemical formula, material type, whether there is a direct band gap, A / B / X atoms, radius R_A / R_B / R_X, electronegativity Chi_A / Chi_B / Chi_X, tolerance factor, octahedral factor, electronegativity difference, crystal system, bond length, density, convex hull energy, formation energy, and total magnetization. In this embodiment, a machine learning training database containing 4630 sets of samples was constructed.

[0059] S3. Employing a multi-dimensional feature analysis method combining correlation analysis, interpretability analysis, and visual comparative analysis, the features of the machine learning training database are quantitatively analyzed; specifically including:

[0060] S301, Correlation Analysis

[0061] Pearson correlation coefficients for each feature are calculated using a correlation heatmap. The strength of linear correlations between features is quantified using color gradients, which helps identify and remove highly redundant features, improving model training efficiency. Figure 3 As shown;

[0062] S302, Interpretability Analysis

[0063] The contribution of each feature to the bandgap prediction results was calculated using SHAP plots to clarify the positive or negative contribution of each feature to the model output. The importance of the features was ranked to determine the core intrinsic structural parameters affecting the bandgap of halide perovskites, such as... Figure 5 As shown;

[0064] S303, Visual Comparison Analysis

[0065] Radar charts are used to visualize key intrinsic structural parameters of different halide perovskite samples, intuitively reflecting the differences in the distribution of various materials in dimensions such as ionic radius, electronegativity, lattice constant, and formation energy. This provides data support for model optimization and material design. Figure 4 As shown;

[0066] S4. Construct a bandgap prediction model based on the XGBoost algorithm. The principle of XGBoost is as follows:

[0067] ;

[0068] in, Let K be the predicted value for the i-th sample, and K be the total number of decision trees. The k-th tree pairs samples The output score, Let be the function space of all regression trees;

[0069] Complete model training, parameter optimization, and bandgap prediction; specifically including:

[0070] S401. The final training dataset after multi-dimensional feature analysis is divided into training and test sets in an 80%:20% ratio; an initial bandgap prediction model is constructed based on the XGBoost algorithm, and hyperparameters are iteratively optimized using GridSearchCV combined with 10-fold cross-validation, such as... Figure 6 As shown; the optimized model's coefficient of determination R 2 The mean squared error (MSE) is 0.863, and the mean squared error (MSE) is 0.611, indicating excellent prediction accuracy. Figure 2 As shown;

[0071] S402. Obtain the characteristic data of the halide perovskite to be predicted. After multi-stage preprocessing in step S1, input the data into the optimized bandgap prediction model. The model directly outputs the bandgap prediction value, realizing the rapid and accurate prediction of the bandgap of halide perovskite.

[0072] This invention combines multi-stage data preprocessing, multi-dimensional feature analysis, and XGBoost model optimization to achieve efficient prediction of the band gap of ABX3 type and derived halide perovskite structures without complex high-throughput DFT calculations. Compared with traditional methods, this invention significantly reduces computational costs and time consumption, while possessing good interpretability and guidance, effectively accelerating the research and screening of novel halide perovskite materials.

[0073] Example 1

[0074] To verify the accuracy and reliability of the XGBoost-based halide perovskite bandgap prediction method of this invention, this embodiment selects six typical ABX3 type halide perovskite bulk phase materials as test samples, specifically including: formamidinium lead iodine (FAPbI3), methylamine lead iodine (MAPbI3), methylamine lead bromide (MAPbBr3), methylamine lead chloride (MAPbCl3), cesium lead iodine (CsPbI3), and cesium lead bromide (CsPbBr3).

[0075] For the above six test samples, following the multi-stage preprocessing process described in step S1 of this invention, the Materials Project database is called through the MP-API interface provided by Pymatgen to collect the original intrinsic structural feature data of each sample (including parameters such as elemental composition, ionic radius, electronegativity, lattice constant, and formation energy). Invalid feature deletion, missing value filling, standardization processing, and multi-level feature screening are completed in sequence to obtain standardized feature data that meets the model input requirements.

[0076] The standardized feature data is input into the XGBoost bandgap prediction model optimized in step S4 of this invention to obtain the predicted bandgap values ​​for each test sample. Simultaneously, the experimentally measured bandgap values ​​of the six halide perovskite phase materials disclosed in the prior art are retrieved, and the predicted values ​​are compared and verified with the experimental values. The comparison results are as follows: Figure 7 As shown.

[0077] Figure 7 In the graph, the horizontal axis represents six ABX3 halide perovskite samples with different compositions, and the vertical axis represents the bandgap value (unit: eV). The blue bars represent the bandgap prediction values ​​output by the XGBoost bandgap prediction model of this application, while the red bars represent the experimentally measured bandgap values ​​disclosed in the prior art. Comparative analysis shows that the deviations between the predicted and experimentally measured bandgap values ​​for the six test samples are strictly controlled within ±0.15 eV, exhibiting a small deviation range and good consistency. This fully demonstrates that the XGBoost bandgap prediction model constructed in this invention has high prediction accuracy and reliability, and can meet the practical application requirements for accurate bandgap prediction of halide perovskites.

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the bandgap of halide perovskites based on XGBoost, characterized in that, Includes the following steps: S1. For ABX3 type and derived halide perovskites, collect raw feature data and perform multi-stage preprocessing to obtain the optimized target feature set. S2. Based on the target feature set, construct a machine learning training database containing intrinsic structural parameters of halide perovskites; S3. Using a multi-dimensional feature analysis method that combines correlation analysis, interpretability analysis, and visual comparative analysis, the features of the machine learning training database are analyzed to obtain the final training dataset. S4. A bandgap prediction model is constructed based on the XGBoost algorithm. The bandgap prediction model is optimized by combining GridSearchCV with 10-fold cross-validation. The optimized bandgap prediction model is used to predict the bandgap of halide perovskites.

2. The method for predicting the bandgap of halide perovskites based on XGBoost according to claim 1, characterized in that, In step S1, the original feature data is collected and subjected to multi-stage preprocessing to obtain the optimized target feature set, specifically including: S101. Use the MP-API interface provided by Pymatgen to call the Materials Project database, collect the original feature data of halide perovskite, and delete invalid features, constant features and fixed column data. S102. Use the ColumnTransformer tool to process numerical features and categorical features respectively. For numerical features, first fill missing values ​​with the median and then perform standardization; for categorical features, fill missing values ​​with the mode. S103. Perform feature filtering sequentially to remove near-constant features with variances approaching 0; calculate the correlation coefficient matrix of numerical features and remove redundant variables from features with correlation coefficients > 0.9; use the SelectKBest algorithm to filter out the features most strongly correlated with the bandgap target variable to form the target feature set.

3. The method for predicting the bandgap of halide perovskites based on XGBoost according to claim 1, characterized in that, In step S2, the intrinsic structural parameters include chemical formula, material type, whether there is a direct band gap, A / B / X atoms, radius R_A / R_B / R_X, electronegativity Chi_A / Chi_B / Chi_X, tolerance factor, octahedral factor, electronegativity difference, crystal system, bond length, density, convex hull energy, formation energy, and total magnetization.

4. The method for predicting the bandgap of halide perovskites based on XGBoost according to claim 1, characterized in that, In step S3, the correlation analysis is to calculate the Pearson correlation coefficient of each feature using a correlation heatmap, quantify the strength of the linear association between features, and help to eliminate redundant features. The interpretability analysis uses SHAP plots to calculate the SHAP value of each feature on the bandgap prediction result, clarifies the positive or negative contribution of each feature to the output of the bandgap prediction model, and sorts them to obtain the core features that affect the bandgap. The visualization and comparative analysis uses radar charts to visually display the differences and patterns in the characteristic distribution of different halide perovskite samples across the intrinsic structural parameters.

5. The method for predicting the bandgap of halide perovskites based on XGBoost according to claim 1, characterized in that, Step S4 specifically includes: S401. Split the final training dataset obtained in step S3 into a training set and a test set in an 80%:20% ratio. Construct a bandgap prediction model based on the XGBoost algorithm. Optimize the model using GridSearchCV with 10-fold cross-validation to achieve the desired R-value for the optimized bandgap prediction model. 2 =0.863, mean square error MSE=0.611; S402. Obtain the original characteristic data of the halide perovskite to be predicted, and input it into the optimized bandgap prediction model after multi-stage preprocessing in step S1. The model outputs the bandgap prediction value to complete the bandgap prediction.

6. The method for predicting the bandgap of halide perovskites based on XGBoost according to claim 5, characterized in that, The formula for the XGBoost algorithm is as follows: ； in, Let K be the predicted value for the i-th sample, and K be the total number of decision trees. The k-th tree pairs samples The output score, Let be the function space for all regression trees.

7. The method for predicting the bandgap of halide perovskites based on XGBoost according to claim 1, characterized in that, In the ABX3 type halide perovskite, the A site is a monovalent cation, the B site is a divalent metal cation, and the X site is a monovalent halide anion.

8. A halide perovskite bandgap prediction system based on XGBoost, characterized in that, The method for predicting the bandgap of halide perovskites based on XGBoost as described in any one of claims 1-7 includes: The data preprocessing module is used to collect raw feature data and perform multi-stage preprocessing on ABX3 type and derived halide perovskites to obtain an optimized target feature set. A database construction module is used to construct a machine learning training database containing intrinsic structural parameters of halide perovskites based on the target feature set. The feature analysis module is used to analyze the features of the machine learning training database by employing a multi-dimensional feature analysis method that combines correlation analysis, interpretability analysis, and visual comparative analysis to obtain the final training dataset. The model prediction module is used to build a bandgap prediction model based on the XGBoost algorithm. The bandgap prediction model is optimized by combining GridSearchCV with 10-fold cross-validation. The optimized bandgap prediction model is used to predict the bandgap of halide perovskites.

9. The halide perovskite bandgap prediction system based on XGBoost according to claim 8, characterized in that, It also includes a data storage module for storing raw feature data, target feature sets, machine learning training databases, optimized bandgap prediction model parameters and bandgap prediction results, and supports data traceability and secondary retrieval.

Images