Large-scale direct band gap double perovskite design method and system based on ensemble learning, and medium
Through integrated learning methods and SHAP analysis, the direct bandgap materials of dual perovskites are quickly screened out, solving the problem of time-consuming and resource-consuming of traditional methods, and achieving the efficient design and commercialization of dual perovskite solar cells.
Patent Information
- Application Number
- CN202411207278.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-12-03
AI Technical Summary
In the field of perovskite solar cells, traditional methods for finding suitable direct bandgap dual perovskite materials consume time and resources, hindering the commercialization process of dual perovskite solar cells.
Using an integrated learning method, the data of the double perovskite material is obtained, cleaning, preprocessing and feature screening is performed, the model is constructed using machine learning algorithms, and interpretability analysis is performed in combination with SHAP method to screen out the double perovskite with direct bandgap. Material.
The rapid and accurate design of direct bandgap double perovskite materials is achieved, reducing the cost of trial and error in experiments and accelerating the production process of double perovskite solar cells.
Smart Images

Figure CN119069047A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of perovskite batteries, and in particular to a large-scale direct bandgap double perovskite design method, system and medium based on integrated learning. Background Art
[0002] As a new generation of photovoltaic materials, halide perovskites with the general formula ABX3 have been widely studied. However, problems such as poor stability, weak adjustability and lead pollution limit their commercial application. Double perovskites have excellent properties such as environmental friendliness, good stability and adjustable photoelectric properties. Therefore, they have broad application prospects as alternative materials for perovskite solar cells. Lead-free inorganic double perovskites are considered to be potential alternatives to monoperovskites.
[0003] To make high-efficiency double perovskite solar cells, double perovskite materials with suitable band gap types must be used. However, due to the huge potential element space, finding suitable direct band gap double perovskites through traditional trial and error or DFT calculations will consume a lot of time and resources, which may hinder the commercialization of double perovskite solar cells.
[0004] The rapid development of machine learning technology has provided new opportunities for the optimization of perovskite devices. Machine learning focuses on extracting hidden, effective and understandable knowledge from large amounts of data, and provides us with good inspiration by using various algorithms to learn and predict certain material properties. Ensemble learning is an important technique and method in machine learning. Its core idea is to improve the performance and generalization ability of the model by constructing and combining multiple learners (usually called base learners or weak learners). There is currently no method that attempts to classify inorganic double perovskites by direct or indirect band gap from the perspective of ensemble learning. Summary of the invention
[0005] To meet this demand, the present invention uses an integrated learning method to provide an intelligent, fast and accurate method for searching and designing direct bandgap double perovskites, so as to reduce the trial and error cost of experiments and accelerate the commercialization process of double perovskite batteries.
[0006] The purpose of the present invention can be achieved by the following technical solutions:
[0007] The first aspect of the present invention provides a direct bandgap double perovskite design method based on ensemble learning, comprising the following steps:
[0008] S1, obtaining information of a double perovskite material system of interest, obtaining known double perovskite material data, wherein the double perovskite material data includes component information of each site of the double perovskite and information on its band gap type, and establishing an initial database;
[0009] S2, based on ensemble learning, cleaning and preprocessing the data in the initial database to obtain a modeling data set for ensemble learning, wherein the modeling data set includes component information and band gap type information of each site of the double perovskite, taking the component information of each site as a variable, taking the band gap type information as a target variable, and classifying the problem as a classification problem;
[0010] S3, screen the features of the modeling data set to obtain the best feature subset, perform model training and evaluation on the best feature subset, use an integrated learning algorithm to improve the model prediction ability, obtain a machine learning prediction model, and use the SHAP method to perform interpretability analysis on the machine learning prediction model;
[0011] S4, by limiting the elements of each site, a theoretical double perovskite data set is created. After using the electroneutrality screening and tolerance factor screening methods to filter out unreasonable data, the filtered theoretical double perovskite data set is used as the input of the machine learning prediction model obtained in S3, and the model predicts that the double perovskite is a double perovskite with a direct band gap.
[0012] Further: In S1, the process of establishing the initial database includes the following sub-steps:
[0013] S1.1, obtain the information of the double perovskite material system of interest, and collect the information of the components of each site and the band gap type of the known double perovskite materials from the literature and public material databases;
[0014] S1.2, component information includes the atomic radius, atomic number, atomic radius, van der Waals radius, Pauling electronegativity, number of valence electrons, ionization energy, dipole polarizability, oxidation state and ionic radius of each site of the double perovskite material (chemical formula A2BB′X6). There are a total of 40 characteristics at the four sites. One element may have multiple oxidation states, and the most common one is taken here.
[0015] S1.3, integrate the component information and band gap type of each site to establish an initial double perovskite material data set.
[0016] In S2, according to the focus content, the target variable of ensemble learning is the category label for the classification task, with 1 as the direct band gap and 0 as the indirect band gap.
[0017] Further: S2 specifically includes the following sub-steps:
[0018] S2.1, based on machine learning, outlier processing is performed on the features of the initial data set, and then data normalization is performed. The features used in the present invention are component features, which have almost no missing values and do not require additional missing value processing, so as to obtain a data set for integrated learning modeling. The modeling data set includes component features and band gap categories of each site of the double perovskite;
[0019] S2.2, based on machine learning, for the problem of class imbalance, sampling is required to increase data to make the categories balanced. For a specific data set, if the category label ratio exceeds 1:10, data sampling processing is required.
[0020] Further: In S2.1, the upper limit of the outlier is Q3+1.5×(Q3-Q1), and the lower limit of the outlier is Q1-1.5×(Q3-Q1), where Q1 is the first quartile and Q3 is the third quartile; data normalization uses the standardization operation, x^'=(x-μ) / σ, where x is the original data, μ is the data mean, σ is the standard deviation of the data, and x' is the normalized data.
[0021] Furthermore, S3 specifically includes the following sub-steps:
[0022] S3.1, perform Pierce correlation calculation on the features of the data set obtained in S2, filter out the features with high correlation, and obtain a feature subset with low collinearity;
[0023] S3.2, using multiple ensemble learning model algorithms to perform model training on low collinearity feature subsets, wherein the ensemble learning model algorithm is a classification algorithm and / or a regression algorithm, using the F1-score as an indicator, using ten-fold cross validation for each model, and taking the average score of each fold cross validation as the model performance indicator to evaluate the model performance and use it as a baseline;
[0024] S3.3, based on machine learning, uses the RFE method to screen the low collinearity feature set, improve the model interpretability and optimize the model;
[0025] S3.4, using grid search to optimize model hyperparameters of the optimized multiple ensemble learning models, wherein the goal is to improve the ten-fold crossover score, search the hyperparameter space of each model, and obtain the optimal model;
[0026] S3.5, cite the SHAP method to perform interpretability analysis on the features of the ensemble learning prediction model.
[0027] Furthermore, in S3.2, the low collinearity feature subset is trained using four classification algorithms, namely, Gradientboosting classification, Random Forest, LGBM, and Catboost, respectively.
[0028] Furthermore, S4 includes the following sub-steps:
[0029] S4.1, combining possible elements at each site of double perovskite to create a theoretical double perovskite database;
[0030] S4.2, using electroneutrality screening, remove the non-electroneutral double perovskites from the theoretical double perovskite database;
[0031] S4.3, using the tolerance factor screening method, calculate the tolerance factor of double perovskites in the theoretical double perovskite database, and delete those with a tolerance factor less than 4.18;
[0032] S4.4, using the double perovskite data screened by electrical neutrality and tolerance factor as the input of the optimal model obtained in S3, obtaining the band gap category predicted by the model, screening out the part predicted to be a direct band gap, and completing the design process of the direct band gap double perovskite.
[0033] Furthermore, in S4.3, the tolerance factor screening method formula is specifically: Where r represents the ionic radius, the subscript represents the site to which the ionic radius belongs, n represents the oxidation state of the element, and r B The value of r in double perovskite A2BB′X6 is used B With r B ' as an alternative.
[0034] A second aspect of the present invention provides a system for integrated learning of large-scale direct bandgap double perovskite cell design, comprising:
[0035] An initial database establishment module is used to obtain information about the double perovskite material system of interest, obtain known double perovskite material data, wherein the double perovskite material data includes component information of each site of the double perovskite and information on its band gap type, and establish an initial database;
[0036] A data preprocessing module, based on ensemble learning, cleans and preprocesses the data in the initial database to obtain a modeling data set for ensemble learning, wherein the modeling data set includes component information and band gap type information of each site of the double perovskite, takes the component information of each site as a variable, takes the band gap type information as a target variable, and classifies the problem as a classification problem;
[0037] The feature engineering and model acquisition module screens the features of the modeling data set to obtain the best feature subset, performs model training and evaluation on the best feature subset, and uses an integrated learning algorithm to improve the model's prediction ability, obtain a machine learning prediction model, and use the SHAP method to perform interpretability analysis on the machine learning prediction model;
[0038] The direct band gap double perovskite design module creates a theoretical double perovskite data set by limiting the elements at each site. After filtering out unreasonable data using the electroneutrality screening and tolerance factor screening methods, the filtered theoretical double perovskite data set is used as the input of the machine learning prediction model obtained in S3, and the model predicts that the double perovskite is a direct band gap.
[0039] A third aspect of the present invention provides a storage medium containing computer executable instructions, which, when executed by a computer processor, is used to execute the direct bandgap double perovskite design method based on ensemble learning as described in any one of claims 1 to 8.
[0040] Compared with the prior art, the present invention has the following technical advantages:
[0041] 1) The present invention establishes a double perovskite direct band gap classifier based on ensemble learning, which provides an effective exploration direction for identifying whether the double perovskite is a direct band gap.
[0042] 2) The present invention can not only provide the influence trend of each component characteristic on the direct band gap of the double perovskite material, but also quickly design a double perovskite material with a direct band gap to reduce the trial and error cost of the experiment.
[0043] 3) By learning the known double perovskite bandgap category data, the present invention can not only provide the influence trend of each characteristic on the bandgap category, but also quickly give the desired direct bandgap double perovskite, so as to reduce the number of experiments in searching for direct bandgap double perovskite and accelerate the production process of double perovskite solar cells. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 is a flow chart of the direct bandgap double perovskite material design method in Example 1 of the present invention;
[0045] Figure 2 is a flow chart of step 2 of designing a direct bandgap double perovskite material in Example 1 of the present invention;
[0046] Figure 3 is a flow chart of step 3 of direct bandgap double perovskite material design in Example 1 of the present invention;
[0047] Figure 4 is a flow chart of step 4 of designing a direct bandgap double perovskite material in Example 1 of the present invention;
[0048] Figure 5 is a data distribution diagram of the X-site element in the modeling data set in Example 1 of the present invention;
[0049] Figure 6 is a distribution diagram of direct band gap and indirect band gap data in the modeling data set in Example 1 of the present invention;
[0050] Figure 7 It is a line graph of the number of features and the ten-fold cross validation scores calculated by the RFE method in step 3.3 for the four ensemble learning models in Example 1 of the present invention;
[0051] Figure 8 is a ten-fold cross validation score histogram of the optimal number of feature subsets for the four ensemble learning models in Example 1 of the present invention;
[0052] Fig. 9 The contribution of the Pauling electronegativity characteristics of the A site to the target variable is analyzed using the SHAP method in Example 1 of the present invention;
[0053] Fig.10 The contribution of the Pauling electronegativity characteristics of the B site to the target variable is analyzed using the SHAP method in Example 1 of the present invention;
[0054] Fig.11 The SHAP method is used in Example 1 of the present invention to analyze the contribution of the ionic radius characteristics of the B' site to the target variable;
[0055] Fig.12 These are the individual site elements used to construct the theoretical double perovskite database in step 4.1 in Example 1 of the present invention. DETAILED DESCRIPTION
[0056] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms and other features not clearly described in this technical solution are all considered to be common technical features disclosed in the prior art.
[0057] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments and the accompanying drawings specifically illustrate the method of large-scale direct bandgap double perovskite design based on ensemble learning of the present invention.
[0058] Example 1
[0059] Figure 1 is a flow chart of the direct bandgap double perovskite material design method in Example 1 of the present invention;
[0060] like Figure 1 As shown, this embodiment provides a method for designing large-scale direct bandgap double perovskite cells based on integrated learning, including the following steps:
[0061] Step 1: According to the inorganic double perovskite system studied in this embodiment, data is collected from literature laboratories or public databases, including double perovskite component information and band gap information, to establish an initial data set.
[0062] Specifically, the process of building a complete initial dataset includes the following sub-steps:
[0063] Step 1.1, according to the studied anode double perovskite (general formula A2BB′X6), 2938 data were collected from the MP database.
[0064] Step 1.2, the component information in the data selected is the easily accessible data, including the atomic radius, atomic number, atomic radius, van der Waals radius, Pauling electronegativity, number of valence electrons, ionization energy, dipole polarizability, oxidation state and ionic radius of each site of the double perovskite material. There are a total of 40 features at the four sites. One element may have multiple oxidation states. The most common one is taken here. The band gap information is the category label, whether it is a direct band gap.
[0065] Step 2: Based on ensemble learning, the initial data is cleaned and preprocessed to obtain a modeling data set for ensemble learning. The modeling data set includes the component information and bandgap type information of each site of the double perovskite. The component information of each site is used as a variable, and the bandgap type information is used as the target variable, and the problem is classified as a classification problem.
[0066] Figure 2 Flow chart of step 2 of direct bandgap double perovskite material design in Example 1. Figure 2 As shown, step 2 includes the following sub-steps:
[0067] Step 2.1, based on machine learning, outlier processing is performed on the features of the initial data set, and then data normalization is performed. The features used in Example 1 are component features, which have almost no missing values and do not require additional missing value processing. A data set for integrated learning modeling is obtained. The modeling data set includes component features and band gap categories of each site of double perovskite;
[0068] The upper bound of the outlier value used is Q3+1.5×(Q3-Q1), and the lower bound of the outlier value is Q1-1.5×(Q3-Q1), where Q1 is the first quartile and Q3 is the third quartile. Data normalization uses standardization operations. Where x is the original data, μ is the mean value of the data, σ is the standard deviation of the data, and x′ is the normalized data;
[0069] Figure 5 It is a data distribution diagram of the X-site element in the modeling data set in Example 1 of the present invention.
[0070] like Figure 5 As shown, the number of S-based and Se-based double perovskites in the X-site element is very small, and they are considered as outliers and deleted in this Example 1.
[0071] Step 2.2, based on machine learning, for the problem of class imbalance, sampling is required to increase data to make the categories balanced. For a specific data set, if the category label ratio exceeds 1:10, data sampling processing is required.
[0072] Figure 6 is a distribution diagram of direct band gap and indirect band gap data in the modeling data set in Example 1 of the present invention;
[0073] like Figure 6 As shown, in the data set used in this embodiment 1, the distribution of direct band gap and indirect band gap is not much different, and the ratio is about 3:1, so there is no need to perform class balanced sampling processing.
[0074] Step 3: Screen the features of the data set to obtain the best feature subset, perform model training and evaluation on the best feature subset, use an ensemble learning algorithm to improve the model prediction ability, obtain a machine learning model, and use the SHAP method to perform interpretability analysis on the machine learning prediction model;
[0075] Figure 3 It is a flow chart of step 3 of designing direct bandgap double perovskite material in Example 1 of the present invention.
[0076] like Figure 3 As shown, step 3 includes the following sub-steps:
[0077] Step 3.1, perform Pierce correlation calculation on the features of the data set obtained in step 2, filter out features with high correlation, and obtain a feature set with low collinearity;
[0078] Step 3.2: Use multiple ensemble learning algorithms to train models for low-collinearity feature subsets. The ensemble learning model algorithm is a classification algorithm or a regression algorithm. If the low-collinearity feature subset uses four classification algorithms (Gradientboosting classifier, Random Forest, LGBM, Catboost) to train models respectively, 4 models are obtained. Use the F1-score as an indicator, use ten-fold cross validation for each model, and use the average score of each fold cross validation as the model performance indicator to evaluate the model performance and use it as a baseline.
[0079] Step 3.3, based on machine learning, use the RFE method to screen the low collinearity feature set to improve the model interpretability and optimize the model.
[0080] Figure 7 It is a line graph of the number of features and the ten-fold cross validation scores calculated by the RFE method in step 3.3 for the four ensemble learning models in Example 1 of the present invention;
[0081] like Figure 7 As shown, in this embodiment 1, the use of the LGBM model can achieve a high ten-fold accuracy with relatively fewer features and determine the optimal feature set of other models.
[0082] Step 3.4, use grid search to optimize the model hyperparameters of the four optimized ensemble learning models, wherein the goal is to improve the ten-fold cross validation score, search the hyperparameter space of each model, and obtain the optimal model. In this example 1, the optimization of the model hyperparameters did not achieve a good effect.
[0083] Figure 8 is a ten-fold cross validation score histogram of the optimal number of feature subsets for the four ensemble learning models in Example 1 of the present invention;
[0084] like Figure 8 As shown in FIG. 1 , the final effect of the four ensemble learning models used in Example 1 is shown. In this example, the F1-score of the maximized model is used as the target indicator, and the LGBM with better performance after optimization is selected as the final model.
[0085] In step 3.5, the SHAP method is used to perform interpretability analysis on the features of the ensemble learning prediction model.
[0086] Fig. 9 The contribution of the Pauling electronegativity characteristics of the A site to the target variable is analyzed using the SHAP method in Example 1 of the present invention;
[0087] Fig.10 The contribution of the Pauling electronegativity characteristics of the B site to the target variable is analyzed using the SHAP method in Example 1 of the present invention;
[0088] Fig.11 The SHAP method is used in Example 1 of the present invention to analyze the contribution of the ionic radius characteristics of the B' site to the target variable;
[0089] like Figure 9-11 As shown in Example 1, the SHAP method provides a perspective for model interpretation. For example, as the A-site Pauling electronegativity increases, the probability that the double perovskite is a direct band gap also increases. Fig.10 , 11 Same reason.
[0090] Figure 4 4 is a flow chart of step 4 of designing a direct bandgap double perovskite material in Example 1 of the present invention. Figure 4 As shown, step 4 includes the following sub-steps:
[0091] Step 4.1, combine the possible elements at each site of the double perovskite to create a theoretical double perovskite database;
[0092] Fig.12 Shown is the inorganic double perovskite building element table used in this Example 1.
[0093] Step 4.2, using electroneutrality screening, delete the non-electroneutral double perovskites in the theoretical double perovskite database;
[0094] Step 4.3, using the tolerance factor screening method, calculate the tolerance factor of double perovskites in the theoretical double perovskite database, and delete those with a tolerance factor less than 4.18;
[0095] Step 4.4, using the double perovskite data screened by electrical neutrality and tolerance factor as the input of the optimal model obtained in step 3, obtaining the band gap category predicted by the model, screening out the part predicted as direct band gap, and completing the design process of direct band gap double perovskite.
[0096] This embodiment also provides a system for large-scale direct band gap double perovskite battery design based on ensemble learning, using the system for large-scale direct band gap double perovskite battery design based on ensemble learning of the above-mentioned embodiment 1, including: an initial database establishment unit, a data preprocessing unit, a feature engineering and model acquisition unit, and a direct band gap double perovskite design unit.
[0097] The initial database establishment module collects known double perovskite material data based on the double perovskite material system under study, including the component information of each site of the double perovskite and its band gap type, to establish an initial database;
[0098] The data preprocessing module cleans and preprocesses the initial data based on ensemble learning to obtain a modeling data set for ensemble learning. The modeling data set includes the component information and bandgap type information of each site of the double perovskite. The component information of each site is used as a variable, and the bandgap type information is used as the target variable, and the problem is classified as a classification problem;
[0099] The feature engineering and model acquisition module screens the features of the data set to obtain the best feature subset, performs model training and evaluation on the best feature subset, and uses an integrated learning algorithm to improve the model's prediction ability, obtain a machine learning model, and use the SHAP method to perform interpretability analysis on the machine learning prediction model;
[0100] The direct bandgap double perovskite design module creates a theoretical double perovskite database by limiting the elements at each site. After filtering out unreasonable data using the electroneutrality screening and tolerance factor screening methods, the filtered data set is used as the input of the machine learning model obtained in step 3 to obtain a double perovskite predicted by the model to have a direct bandgap, thus completing the entire direct bandgap double perovskite design process.
[0101] It should be understood that the modules of the direct bandgap double perovskite design system described above can correspond to the steps of the double perovskite design method described in Example 1. Thus, the operations, features and advantages described above for the steps based on the method are also applicable to the modules of the system. For the sake of brevity, some operations, features and advantages are not described here.
[0102] This embodiment also proposes a computer-readable storage medium, wherein the storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the above-mentioned direct bandgap double perovskite design method based on integrated learning. The storage medium can be an electronic medium, a magnetic medium, an optical medium, an electromagnetic medium, an infrared medium, or a semiconductor system or a propagation medium. The storage medium can also include a semiconductor or solid-state memory, a magnetic tape, a removable computer disk, a random access memory (RAM), a read-only memory (ROM), a hard disk, and an optical disk. The optical disk can include a compact disk-read only memory (CD-ROM), a compact disk-read and write (CD-RW), and a DVD.
[0103] Based on the embodiments, it can be seen that the present invention can provide the influence trend of each component characteristic on the direct band gap of the double perovskite and can also speed up the provision of a double perovskite that can achieve the desired direct band gap, thereby reducing the trial and error cost of the experiment.
[0104] Based on the embodiments, it can be seen that the present invention establishes an integrated learning design process for the inorganic double perovskite system, provides a direction for establishing a direct band gap classifier for the inorganic double perovskite system, and accelerates the process of identifying the band gap type of inorganic double perovskite.
[0105] Based on the embodiments, it can be seen that the present invention also provides a direct bandgap material design process for an inorganic double perovskite system based on ensemble learning, which accelerates the cycle of discovering new direct bandgap inorganic double perovskites, reduces the resource consumption caused by experiments and large-scale calculations, and accelerates the optimization process of double perovskite solar cells.
[0106] The above description of the embodiments is to facilitate the understanding and use of the invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to these embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the scope of protection of the present invention.
Claims
1. A direct bandgap double perovskite design method based on ensemble learning, characterized in that: The steps include: S1, obtaining information of a double perovskite material system of interest, obtaining known double perovskite material data, wherein the double perovskite material data includes component information of each site of the double perovskite and information on its band gap type, and establishing an initial database; S2, based on ensemble learning, cleaning and preprocessing the data in the initial database to obtain a modeling data set for ensemble learning, wherein the modeling data set includes component information and band gap type information of each site of the double perovskite, taking the component information of each site as a variable, taking the band gap type information as a target variable, and classifying the problem as a classification problem; S3, screen the features of the modeling data set to obtain the best feature subset, perform model training and evaluation on the best feature subset, use an integrated learning algorithm to improve the model prediction ability, obtain a machine learning prediction model, and use the SHAP method to perform interpretability analysis on the machine learning prediction model; S4, by limiting the elements of each site, a theoretical double perovskite data set is created. After using the electroneutrality screening and tolerance factor screening methods to filter out unreasonable data, the filtered theoretical double perovskite data set is used as the input of the machine learning prediction model obtained in S3, and the model predicts that the double perovskite is a double perovskite with a direct band gap.
2. The direct bandgap double perovskite design method based on ensemble learning according to claim 1, characterized in that: In S1, the process of establishing the initial database includes the following sub-steps: S1.1, obtain the information of the double perovskite material system of interest, and collect the information of the components of each site and the band gap type of the known double perovskite materials from the literature and public material databases; S1.2, component information includes atomic radius, atomic number, atomic radius, van der Waals radius, Pauling electronegativity, number of valence electrons, ionization energy, dipole polarizability, oxidation state and ionic radius of each site of the double perovskite material A2BB′X6, with a total of 40 features for the four sites; S1.3, integrate the component information and band gap type of each site to establish an initial double perovskite material data set.
3. The direct bandgap double perovskite design method based on ensemble learning according to claim 1, characterized in that: S2 specifically includes the following sub-steps: S2.1, based on machine learning, outlier processing is performed on the features of the initial data set, and then data normalization is performed to obtain a data set for integrated learning modeling, where the modeling data set includes component characteristics and band gap categories of each site of the double perovskite; S2.2, based on machine learning, for the problem of class imbalance, it is necessary to sample and increase the data to make the categories balanced.
4. The direct bandgap double perovskite design method based on ensemble learning according to claim 3, characterized in that: In S2.1, the upper bound of the outlier is Q3+1.5×(Q3-Q1), and the lower bound of the outlier is Q1-1.5×(Q3-Q1), where Q1 is the first quartile and Q3 is the third quartile; Data normalization uses the standardization operation, x^' = (x-μ) / σ, where x is the original data, μ is the data mean, σ is the standard deviation of the data, and x' is the normalized data.
5. The direct bandgap double perovskite design method based on ensemble learning according to claim 1, characterized in that: in, S3 includes the following sub-steps: S3.1, perform Pierce correlation calculation on the features of the data set obtained in S2, filter out the features with high correlation, and obtain a feature subset with low collinearity; S3.2, using multiple ensemble learning model algorithms to perform model training on low collinearity feature subsets, wherein the ensemble learning model algorithm is a classification algorithm and / or a regression algorithm, using the F1-score as an indicator, using ten-fold cross validation for each model, and taking the average score of each fold cross validation as the model performance indicator to evaluate the model performance and use it as a baseline; S3.3, based on machine learning, uses the RFE method to screen the low collinearity feature set, improve the model interpretability and optimize the model; S3.4, using grid search to optimize model hyperparameters of the optimized multiple ensemble learning models, wherein the goal is to improve the ten-fold crossover score, search the hyperparameter space of each model, and obtain the optimal model; S3.5, cite the SHAP method to perform interpretability analysis on the features of the ensemble learning prediction model.
6. The direct bandgap double perovskite design method based on ensemble learning according to claim 5, characterized in that: In S3.2, the low collinearity feature subset is used to train models using four classification algorithms, namely, Gradient boosting classification, Random Forest, LGBM, and Catboost.
7. The direct bandgap double perovskite design method based on ensemble learning according to claim 1, characterized in that: S4 includes the following sub-steps: S4.1, combining possible elements at each site of double perovskite to create a theoretical double perovskite database; S4.2, using electroneutrality screening, remove the non-electroneutral double perovskites from the theoretical double perovskite database; S4.3, using the tolerance factor screening method, calculate the tolerance factor of double perovskites in the theoretical double perovskite database, and delete those with a tolerance factor less than 4.18; S4.4, using the double perovskite data screened by electrical neutrality and tolerance factor as the input of the optimal model obtained in S3, obtaining the band gap category predicted by the model, screening out the part predicted to be a direct band gap, and completing the design process of the direct band gap double perovskite.
8. The direct bandgap double perovskite design method based on ensemble learning according to claim 7, characterized in that: In S4.3, the tolerance factor screening method formula is specifically: Where r represents the ionic radius, the subscript represents the site to which the ionic radius belongs, n represents the oxidation state of the element, and r B The value of r in double perovskite A2BB′X6 is used B With r B′ The average value is used as an alternative.
9. A system for integrated learning of large-scale direct bandgap double perovskite cell design, characterized in that include: An initial database establishment module is used to obtain information about the double perovskite material system of interest, obtain known double perovskite material data, wherein the double perovskite material data includes component information of each site of the double perovskite and information on its band gap type, and establish an initial database; A data preprocessing module, based on ensemble learning, cleans and preprocesses the data in the initial database to obtain a modeling data set for ensemble learning, wherein the modeling data set includes component information and band gap type information of each site of the double perovskite, takes the component information of each site as a variable, takes the band gap type information as a target variable, and classifies the problem as a classification problem; The feature engineering and model acquisition module screens the features of the modeling data set to obtain the best feature subset, performs model training and evaluation on the best feature subset, and uses an integrated learning algorithm to improve the model's prediction ability, obtain a machine learning prediction model, and use the SHAP method to perform interpretability analysis on the machine learning prediction model; The direct band gap double perovskite design module creates a theoretical double perovskite data set by limiting the elements at each site. After filtering out unreasonable data using the electroneutrality screening and tolerance factor screening methods, the filtered theoretical double perovskite data set is used as the input of the machine learning prediction model obtained in S3, and the model predicts that the double perovskite is a direct band gap.
10. A storage medium containing computer executable instructions, characterized in that: When the storage medium of the computer executable instructions is executed by a computer processor, it is used to perform the direct bandgap double perovskite design method based on ensemble learning as described in any one of claims 1 to 8.
