Screening method of A3BX6 type halide perovskite material

Abstract

The invention relates to a screening method of an A3BX6 type halide perovskite material, and belongs to the technical field of artificial intelligence application and novel semiconductor photoelectric material development, original parameters related to composition elements of A3BX6 type halide perovskite are extracted to form an initial data space, and the formation energy and band gap of the A3BX6 type halide perovskite are obtained through screening by a machine learning method; therefore, the perovskite material which can be applied to the photoelectric fields of photoelectric detection, photocatalysis and the like is screened out. According to the method, the workload of researchers in the experiment process can be remarkably reduced, time and resources are saved, the research efficiency is improved, and a novel halide perovskite photoelectric material with high performance can be found more quickly.

Description

Technical Field

[0001] The invention belongs to the technical field of artificial intelligence application and development of new semiconductor optoelectronic materials, and relates to a method for screening A3BX6 type halide perovskite materials. Background Art

[0002] Halide perovskite materials have become potential candidate materials for the next generation of optoelectronic devices due to their excellent optoelectronic properties. Compared with traditional semiconductor materials, halide perovskites exhibit a series of unique optoelectronic properties and excellent performance. These properties include adjustable band structure, strong light absorption ability, long carrier transmission distance and large carrier migration rate. This makes halide perovskites widely used in solar cells, photodetectors, light-emitting diodes and photocatalysis. Among them, zero-dimensional metal halides with efficient photoluminescence have broad application prospects in the field of optoelectronics due to their unique crystal structure and physical and chemical properties. However, A3BX6-type halide perovskite materials have a wide range of element combinations. It will take a lot of manpower and material resources to search for suitable halide perovskites by traditional experimental methods alone. How to efficiently search for new halide perovskite optoelectronic materials has become a difficulty and hot spot in current research. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a method for screening A3BX6 type halide perovskite materials.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] A method for screening A3BX6 type halide perovskite materials, comprising the following steps:

[0006] S1: The formation energy and band gap data of A3BX6 type halide perovskite were screened in The Open Quantum Materials Database (OQMD), and then the relevant atomic element parameters were extracted as the original data set through the Pymatgen library;

[0007] S2: preprocess the original data set and divide it into a regression training set and a regression test set;

[0008] S3: inputting the formation energy regression training set into different algorithms, and using regression evaluation indicators to compare and select the optimal formation energy prediction algorithm;

[0009] S4: using the formation energy regression training set to train the optimal formation energy prediction algorithm, and using the naive Bayes algorithm to optimize and adjust the model parameters to obtain the optimal formation energy regression prediction model;

[0010] S5: Testing the optimal formation energy regression prediction model using the formation energy regression test set to predict the formation energy of the formation energy regression test set;

[0011] S6: Use the regression evaluation index to quantitatively evaluate the difference between the formation energy of the predicted regression test set and the actual sample value to obtain the formation energy regression evaluation result;

[0012] S7: inputting the virtual sample into the formation energy regression prediction model to predict the formation energy of the virtual sample;

[0013] S8: Divide the data set preprocessed in step S2 into a bandgap regression training set and a bandgap regression test set;

[0014] S9: replacing the formation energy regression training set and the formation energy regression test set in steps S3 to S7 with the band gap regression training set and the band gap regression test set, repeating steps S3 to S7, and obtaining an optimal band gap regression prediction model, a band gap regression evaluation result, and a predicted band gap of a virtual sample;

[0015] S10: According to the predicted formation energy and band gap of the virtual samples, a candidate set of virtual samples that meet the range is screened out, and virtual samples with a band gap of 2-4 eV are selected for DFT calculation to verify the accuracy of the prediction model;

[0016] S11: Screening out new A3BX6-type halide perovskite high-performance light-emitting materials from the virtual sample candidate set.

[0017] Furthermore, the preprocessing of the original data set described in step S2 includes re-extracting or deleting sample data with abnormal values, supplementing or deleting missing data; and screening the feature space using the tail elimination method and the cross-correlation method to eliminate redundant features.

[0018] Furthermore, the evaluation index in step S6 includes the root mean square error (RMSE) and the coefficient of determination (R 2 );

[0019] The root mean square error is defined as follows:

[0020]

[0021] The coefficient of determination is defined as follows:

[0022]

[0023] where y i is the true value, is the predicted value, Represents the average value of the true value. When the error between the true value and the predicted value is larger, the RMSE value is larger. Conversely, the smaller the value is, the better the prediction effect of the model is. 2 The larger the value, the better the model effect.

[0024] The beneficial effects of the present invention are as follows: the present invention provides an efficient screening method for A3BX6 type halide perovskite material optoelectronic device materials based on machine learning, which is an effective solution for obtaining material band gap and formation by traditional experimental methods and complex first-principles calculations, saving experimental resources and time, and screening required materials more quickly and efficiently. The formation energy regression prediction model and band gap regression prediction model of halide perovskite established based on machine learning take atomic related parameters as features, establish nonlinear mapping between them and band gap and formation energy, and can not only predict material band gap and formation energy, but also effectively discover the relationship between material band gap, formation energy and constituent atomic parameters, providing ideas for efficient discovery of A3BX6 type halide perovskite optoelectronic materials and improving experimental efficiency.

[0025] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0027] Figure 1 A flow chart of an efficient screening method for novel A3BX6 type halide perovskite material optoelectronic device materials based on machine learning provided in an embodiment of the present invention;

[0028] Figure 2 A comparison of different machine learning methods used in modeling the formation energy regression prediction model and the band gap regression prediction model of the A3BX6 type halide perovskite in the embodiment of the present invention;

[0029] Figure 3 The modeling result of the formation energy regression prediction model of A3BX6 type halide perovskite in the embodiment of the present invention;

[0030] Figure 4 This is the modeling result of the band gap regression prediction model of A3BX6 type halide perovskite in the embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0032] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0033] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0034] Machine learning is used to efficiently screen new A3BX6 halide perovskite materials for optoelectronic devices. The specific method is as follows:

[0035] See also Figure 1 , a novel A3BX6 perovskite material photoelectric device material efficient screening method, comprising the following steps:

[0036] (1) The formation energy and band gap data of A3BX6 halide perovskite were obtained by screening in the OQMD database, and then the relevant atomic parameters were extracted through the Pymatgen library as the original data set samples for machine learning;

[0037] (2) Preprocessing the original data set: preprocessing the data set in step (1) and re-extracting or deleting sample data with abnormal values;

[0038] (3) The data set samples obtained after preprocessing in step (2) are used to eliminate redundant features using the mutual information method to obtain the final data set for modeling, the features of which include the ionization energy, electronegativity, atomic orbital energy, Mendeleev number and molar volume of the elements, as well as the tolerance factor and tolerance factor. The final data set is randomly divided into a formation energy regression training set and a formation energy regression test set in a ratio of 4:1;

[0039] (4) Constructing a regression prediction model for formation energy:

[0040] The original data obtained from the database are input into 12 algorithms. By comparing the 12 machine learning algorithms, namely, decision tree (DT), gradient boosting regression (GBR), integrated regression Bootstrap aggregating (Bagging) and random forest (RF), support vector machine (SVM), linear regression (LR), ridge regression (RG), Bayesian ridge regression (BR), extra tree (ET), adaptive boosting (Adaboost Regression, ABR), integrated regression Bagging, K neighbor regression (KNR), neural network (Multilayer Perceptron, MLP), a most stable and effective machine learning method is found, namely, gradient boosting regression GBR. This method and the formation energy regression training set obtained in step (3) are used to build a model. At the same time, the naive Bayes algorithm is used to optimize the model parameters to obtain the optimal formation energy regression prediction model. Figure 2 Compare the training accuracy results of multiple models.

[0041] (5) Predict the formation energy Regress the formation energy of the test set samples:

[0042] Predicting the formation energy of the regression test set obtained in step (3) according to the formation energy regression prediction model established in step (4);

[0043] (6) Using the regression evaluation index to quantitatively evaluate the difference between the predicted formation energy of the regression test set and the actual sample value, and obtain the formation energy regression evaluation result;

[0044] The evaluation indicators include Root Mean Square Error (RMSE) and Coefficient of Determination (R 2 ).

[0045] The root mean square error is defined as follows:

[0046]

[0047] The coefficient of determination is defined as follows:

[0048]

[0049] where y i is the true value, is the predicted value, Represents the average value of the true value. When the error between the true value and the predicted value is larger, the RMSE value is larger. Conversely, the smaller the value is, the smaller the error between the actual value and the predicted value is, and the better the prediction effect of the model is. 2 The larger the value, the better the model effect.

[0050] According to the result obtained in step (5), the root mean square error and the determination coefficient between the predicted formation energy and the actual formation energy of the regression test set are calculated, and the obtained root mean square error and determination coefficient are used as the formation energy regression evaluation result.

[0051] (7) Prediction of the formation energy of virtual samples:

[0052] Input the virtual sample into the formation energy regression prediction model established in step (4) to obtain the predicted formation energy;

[0053] (8) Using the mutual information method to remove redundant features from the data set samples obtained after preprocessing in step (2), a final data set for modeling is obtained, wherein the features include the ionization energy, electronegativity, atomic orbital energy, Mendeleev number and molar volume of the elements, as well as the tolerance factor and tolerance factor. The final data set is randomly divided into a band gap regression training set and a band gap regression test set at a ratio of 4:1;

[0054] (9) Constructing a band gap regression prediction model:

[0055] By comparing 12 machine learning algorithms, namely decision tree DT, gradient boosting regression GBR, ensemble regression Bagging and random forest RF, support vector SVM, linear regression LR, ridge regression RG, Bayesian ridge regression BR, extra tree ET, adaptive boosting ABR, ensemble regression Bagging, K-nearest neighbor regression KNR, neural network MLP), we found the most stable and effective machine learning method, namely gradient boosting regression GBR. We used this method and the bandgap regression training set obtained in step (8) to build a model. At the same time, we used the naive Bayes algorithm to optimize the model parameters, and adjusted the learning rate (learning_rate), the number of weak classifiers (n_estimator), and the random sampling (subsample) to adjust the parameters. The coefficient of determination (R 2) and Root Mean Square Error (RMSE) as the optimization objectives to obtain the optimal band gap regression prediction model;

[0056] (10) Predict the band gap value of the band gap regression test set samples:

[0057] Predicting the band gap value of the regression test set obtained in step (8) according to the band gap regression prediction model established in step (9);

[0058] (11) using a regression evaluation index to quantitatively evaluate the difference between the predicted band gap value of the regression test set and the actual sample value, and obtaining a band gap regression evaluation result;

[0059] The evaluation indicators include Root Mean Square Error (RMSE) and Coefficient of Determination (R 2 ).

[0060] The root mean square error is defined as follows:

[0061]

[0062] The coefficient of determination is defined as follows:

[0063]

[0064] where y i is the true value, is the predicted value, Represents the average value of the true value, i is the sample, and n is the total number of samples. When the error between the true value and the predicted value is larger, the RMSE value is larger, and vice versa, the smaller the error between the actual value and the predicted value is, the better the prediction effect of the model is; R 2 The larger the value, the better the model effect.

[0065] According to the result obtained in step (10), the root mean square error and the determination coefficient between the predicted band gap value and the actual band gap value of the regression test set are calculated, and the obtained root mean square error and the determination coefficient are used as the band gap regression evaluation result.

[0066] (12) Predict the band gap value of the virtual sample:

[0067] Inputting the virtual sample into the band gap regression prediction model established in step (9) to obtain a predicted band gap value;

[0068] (13) DFT calculation verifies the accuracy of the predicted virtual samples:

[0069] According to the predictions of the formation energy and band gap values ​​obtained in step (7) and step (12), a candidate set of virtual samples that meet the range is screened out, and some virtual samples with a band gap of about 2-4 eV are selected for DFT calculation to verify the accuracy of the prediction model;

[0070] (14) Screening out new A3BX6 type halide perovskite high-performance luminescent materials:

[0071] A new type of A3BX6 halide perovskite material with high performance light-emitting properties is selected based on the virtual sample candidate set in step (13). Since the band gap has a great influence on the luminescent properties of the material, and the formation energy has a significant effect on the stability of the material, a suitable stable halide perovskite is selected from the virtual sample by combining the two;

[0072] This example uses machine learning methods to model the formation energy and band gap data of A3BX6 type halide perovskite materials, and establishes a formation energy regression model and a band gap regression model. The prediction accuracy of the formation energy and band gap reaches 0.989 and 0.986, and the mean square error is reduced to 0.216 and 0.075;

[0073] Figure 3 and Figure 4 This is a diagram showing the modeling results of the formation energy regression prediction model and the band gap regression prediction model of the A3BX6 type halide perovskite material in the embodiment of the present invention.

[0074] In summary, the present invention provides an efficient screening method for a novel A3BX6 type halide perovskite high-performance luminescent material. The formation energy and band gap data of the halide perovskite are obtained by screening in the OQMD database, and then the relevant atomic parameters are extracted through the Pymatgen library to generate the original data set. In order to screen out materials that can be applied to fields such as photodetectors and photocatalysis, a data set of formation energy regression and band gap regression is generated according to the prediction target. The formation energy regression model and band gap regression model of the novel A3BX6 type halide perovskite material are obtained by machine learning. The generated virtual A3BX6 type halide perovskite sample is input into the prediction model to obtain the predicted formation energy and band gap, and the virtual sample candidate set that meets the range is screened out, and some candidate set materials are selected for DFT calculation to verify the accuracy of the prediction model. The time and cost of high-throughput calculations are greatly reduced, and the relevant experimental synthesis can be quickly guided, reducing the experimental cost and the time of the experimenter.

Claims

1. A method for screening A3BX6 type halide perovskite materials, characterized in that: The following steps are involved: S1: The formation energy and band gap data of A3BX6 type halide perovskite were obtained by screening in the OQMD database, and then the relevant atomic element parameters were extracted as the original data set through the Pymatgen library; S2: preprocess the original data set and divide it into a regression training set and a regression test set; S3: inputting the formation energy regression training set into different algorithms, and using regression evaluation indicators to compare and select the optimal formation energy prediction algorithm; S4: using the formation energy regression training set to train the optimal formation energy prediction algorithm, and using the naive Bayes algorithm to optimize and adjust the model parameters to obtain the optimal formation energy regression prediction model; S5: Testing the optimal formation energy regression prediction model using the formation energy regression test set to predict the formation energy of the formation energy regression test set; S6: Use the regression evaluation index to quantitatively evaluate the difference between the formation energy of the predicted regression test set and the actual sample value to obtain the formation energy regression evaluation result; S7: inputting the virtual sample into the formation energy regression prediction model to predict the formation energy of the virtual sample; S8: Divide the data set preprocessed in step S2 into a bandgap regression training set and a bandgap regression test set; S9: replacing the formation energy regression training set and the formation energy regression test set in steps S3 to S7 with the band gap regression training set and the band gap regression test set, repeating steps S3 to S7, and obtaining an optimal band gap regression prediction model, a band gap regression evaluation result, and a predicted band gap of a virtual sample; S10: According to the predicted formation energy and band gap of the virtual samples, a candidate set of virtual samples that meet the range is screened out, and virtual samples with a band gap of 2-4 eV are selected for DFT calculation to verify the accuracy of the prediction model; S11: Screening out new A3BX6-type halide perovskite high-performance light-emitting materials from the virtual sample candidate set.

2. The method for screening A3BX6 type halide perovskite materials according to claim 1, characterized in that: The preprocessing of the original data set in step S2 includes re-extracting or deleting sample data with abnormal values, supplementing or deleting missing data; and screening the feature space using the tail elimination method and the cross-correlation method to eliminate redundant features.

3. The method for screening A3BX6 type halide perovskite materials according to claim 1, characterized in that: The evaluation indicators in step S6 include root mean square error (RMSE) and coefficient of determination (R 2 ); The root mean square error is defined as follows: The coefficient of determination is defined as follows: where y i is the true value, is the predicted value, Represents the average value of the true value. When the error between the true value and the predicted value is larger, the RMSE value is larger. Conversely, the smaller the value is, the better the prediction effect of the model is. 2 The larger the value, the better the model effect.

Images