Automatic screening method for molecular additives in perovskite light-emitting diodes and photovoltaic cells

Through the combination of deep generation model and QSAR prediction model, the automated screening of perovskite optoelectronic device additives is achieved, solving the problem of inefficiency in the existing technology, significantly improving device performance and reducing development costs.

CN119091998AActive Publication Date: 2024-12-06NORTHWESTERN POLYTECHNICAL UNIV

Patent Information

Application Number
CN202411185823.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-12-06
Estimated Expiration
2044-08-27

AI Technical Summary

Technical Problem

When developing perovskite optoelectronic device additives, the existing technology faces the problems of difficulty in exploring huge chemical space, low efficiency of experimental trial and error methods, and difficulty in effectively correlating the generated molecules and target properties in deep learning models, resulting in limited improvement in device performance.

Method used

Using deep generation model and automated screening method based on QSAR prediction model, high-performance automated screening of perovskite light-emitting diodes and photovoltaic cell additives is achieved through training generative models and multiple rounds of screening processes, combined with deep learning algorithms and experimental data.

Benefits of technology

Under the limited data set, the rapid identification of high-performance additive molecules significantly improves device performance, reduces time and economic costs, and achieves optimization of perovskite optoelectronic devices performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091998A_ABST
    Figure CN119091998A_ABST
Patent Text Reader

Abstract

The invention provides an automatic screening method for molecular additives in perovskite light-emitting diodes and photovoltaic cells, which combines a generative model and a prediction model in the deep learning field on the basis of fully utilizing an existing additive data set, and realizes automatic screening of the molecular additives in the perovskite light-emitting diodes and the photovoltaic cells through a development thought of'first generation and then screening '. New molecules similar to existing additive molecular structures are generated firstly, and then the generated molecules are subjected to multiple rounds of screening with ideal additives as targets, so that automatic rapid development of PeLED additive molecules is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of computer-aided molecular design, and in particular to an automated screening method for molecular additives in perovskite light-emitting diodes and photovoltaic cells. Background Art

[0002] Perovskite materials have shown great development prospects and are widely used in the fields of solar cells, light-emitting diodes, photodetectors, etc. due to their excellent optoelectronic properties such as low exciton binding energy, long carrier lifetime, and adjustable band gap. Among them, perovskite light-emitting diodes (PeLEDs) have become the focus of competition in the field of new light-emitting and display technologies around the world, while perovskite solar cells (PSCs) have performed well in the photovoltaic field. Despite this, due to inevitable internal and interface defects, the energy conversion efficiency of perovskite optoelectronic devices still has room for improvement. Adding suitable organic molecules to the precursor solution to reduce defects in the perovskite film and optimize the perovskite layer and its interface is considered an effective strategy to improve device performance. Therefore, the development of efficient additives that can significantly improve the performance of perovskite optoelectronic devices is of great significance for promoting the development of photovoltaic power generation and lighting display technology.

[0003] Molecular design is generally understood as the process of designing and synthesizing molecules with desirable properties that meet specific needs. The main challenge of this process is that the chemical space that needs to be explored is extremely large, containing about 10 60 There are 40 possible molecular structures in the space, and the space is also unevenly distributed and unstructured. Given that different molecules have very different effects on device performance, how to quickly and effectively identify molecular structures with ideal properties from this huge chemical space is a key challenge in the molecular development process.

[0004] Due to the complexity of the effects of additive molecules on perovskite crystallization, morphology, point defects and precursor chemistry, the current development of additives for perovskite optoelectronic devices mainly relies on experimental trial and error, which is time-consuming and inefficient. With the improvement of computing power, people have begun to try to use machine learning-based quantitative structure-activity relationship (QSAR) models to screen additives based on small molecule structures. However, this method is usually limited to existing chemical databases and it is difficult to effectively explore unknown chemical space, so the screened molecules often lack novelty. With the further development of artificial intelligence technology, deep generative models can explore chemical space more comprehensively by modeling the probability distribution behind the molecular set, thereby generating new molecular structures. However, how to effectively associate these generated molecular structures with target properties still needs to be adjusted according to specific circumstances. In the additive development process of this patent, the label data obtained by experimental testing and characterization is relatively limited, and the potential information contained therein inevitably has limitations. How to maximize the use of limited data sets for additive development is another problem that needs to be solved. Based on such a small sample data set, exploring the molecular chemical space as much as possible to find additive molecular structures that can significantly improve device performance is the main challenge facing the current development of perovskite optoelectronic device additives. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention proposes an automated screening method for molecular additives in perovskite light-emitting diodes and photovoltaic cells based on a deep generative model and a virtual screening idea based on a QSAR prediction model, thereby realizing the automated development of high-performance additives.

[0006] Since the perovskite films of PeLEDs and PSCs have the same ABX3-type crystal structure and similar defect types, they face similar challenges in the development of additives. The automated additive screening scheme proposed in the present invention has differences in technical implementation details for PeLEDs and PSCs, but tends to be unified in logical thinking and processes. Therefore, the present invention chooses to specifically demonstrate the development of high-efficiency additives in PeLEDs.

[0007] The automated screening method for molecular additives in perovskite light-emitting diodes and photovoltaic cells proposed in the present invention first uses a generative model that is fully trained in a large molecular data set to obtain a generative molecule set, and then uses multiple deep learning-based QSAR prediction models to perform multiple rounds of screening on the generated results based on the relevant properties of ideal additives, and finally obtains candidate molecules for device preparation and experimental testing.

[0008] The technical solution of the present invention is:

[0009] The automated screening method for molecular additives in perovskite light-emitting diodes and photovoltaic cells comprises the following steps:

[0010] Step 1: Establish a database of PeLED additive molecules, molecular donor quantity DN, and hydrogen bond acceptor strength β;

[0011] Step 2: Use the PeLED additive molecular database and the molecular donor quantity DN and hydrogen bond acceptor strength β database to train the effectiveness prediction model, donor quantity DN value prediction model and hydrogen bond acceptor strength β prediction model respectively;

[0012] Step 3: Use the pre-trained molecular generation model to generate similar new molecules for the molecules in the PeLED additive molecular database to obtain a generated molecule set;

[0013] Step 4: Based on the effectiveness prediction model trained in step 2, the prediction model of the donor quantity DN value and the hydrogen bond acceptor strength β, the generated molecule set obtained in step 3 is subjected to multiple rounds of screening with the ideal additive as the target.

[0014] Furthermore, in step 1, the PeLED additive molecule database is obtained by:

[0015] The reported additive molecular structures and experimental data are collected and established, including the names of the additive molecular structures and the device EQE efficiency values. Based on the molecular structures, the computational chemistry library Pybel and the RDKit information chemistry package are used to calculate the corresponding molecular descriptors and molecular fingerprint data, which together constitute the original PeLED additive molecular database.

[0016] Furthermore, in step 2, the model structure of the effectiveness prediction model is a multi-layer network structure with multiple inputs and a single output; the model input is the molecular descriptor in the original PeLED additive molecule database obtained in step 1, and the molecular fingerprint and the latent code of the molecule in the VAE model together constitute the representation characteristics of the additive molecule; and 0 or 1 labels are given as label inputs according to the effectiveness results of the molecules; the model output is the predicted probability of the effectiveness of the molecules.

[0017] Furthermore, the model input of the effectiveness prediction model is obtained through the following process:

[0018] The CalcDescriptors() method in the RDKit library is used to calculate the values ​​of all descriptors in the Descriptors._descList list for all molecules in the PeLED additive database obtained in step 1, and the descriptors whose value is 0 in all additive molecules and the descriptors whose standard deviation is less than the set value are removed; then the Pearson correlation coefficient is calculated between the remaining descriptors and the descriptors whose correlation is higher than the set value are removed; the remaining descriptors are combined with the molecular fingerprint and the latent code of the molecule in the VAE model as the representation features of the additive molecules and the input of the effectiveness prediction model.

[0019] Furthermore, in step 2, the MACCS molecular fingerprint data in the molecular donor quantity DN data set obtained in step 1 and the molecular descriptors including the most negative atomic charge of the molecule, the molecular HOMO energy level, and the molecular HUMO-LUMO energy gap are used as model inputs, the molecular predicted DN value is used as output, and the DN value measured by the standard experimental method is used as the target variable to establish a donor quantity DN value prediction model.

[0020] Furthermore, in step 2, the donor quantity DN value prediction model is trained using five-fold cross validation, and the average value of the model predictions in the five cross validations is used as the final model performance.

[0021] Furthermore, in step 3, the pre-trained molecular generation model adopts a variational autoencoder and is pre-trained on the ZINC dataset.

[0022] Furthermore, the input of the variational autoencoder is the SMILES string corresponding to each molecular structure in the PeLED additive molecule database; after the SMILES string is converted into a one-hot vector, it is further input into the fully connected layer through a multi-layer 1D convolutional layer, and the latent code is obtained after standardization. After applying random perturbations to the latent code, a series of new perturbation codes are obtained. After the new perturbation code is decoded by the decoder, a new SMILES string similar to the molecular structure of the PeLED additive molecule database can be obtained.

[0023] Furthermore, the screening in step 4 includes the following steps:

[0024] Step 4.1: Structural rationality screening:

[0025] Use the MolFromSmiles() method in the RDKit library to process the decoded new SMIELS string, convert the SMILES string into a Mol object in RDKit, check the chemical rationality of the SMILES during the conversion process, and remove the unreasonable parts;

[0026] Step 4.2: Effectiveness screening:

[0027] Input the SMIELS character string that can be successfully converted in step 4.1 into the effectiveness prediction model obtained in step 2 to obtain an effectiveness prediction probability value ranging from 0 to 1, and select the molecular structure whose effectiveness prediction probability is greater than the set threshold to form the first candidate molecule set;

[0028] Step 4.3: Screening of donor number DN and hydrogen bond acceptor strength β:

[0029] Input the molecules in the first candidate molecule set obtained in step 4.2 into the donor quantity DN value prediction model and hydrogen bond acceptor strength β prediction model obtained in step 2, output the DN prediction value and β prediction value of each molecule, and select the molecular structures with DN value and β value within a certain range to form the second candidate molecule set;

[0030] Step 4.4: Availability and affordability screening:

[0031] The PubChem API was used to obtain the supplier information of the molecules in the second candidate molecule set obtained in step 4.3, and the molecules with complete supplier information and quotations within the set range were selected to form the third candidate molecule set.

[0032] Furthermore, for the molecules in the third candidate molecule set, device preparation and experimental testing are carried out to obtain test results of the corresponding devices; and the test results are added to the PeLED additive molecule database to update the prediction model training, similar molecule generation, and multiple rounds of screening processes with the ideal additive properties as the goal.

[0033] Beneficial Effects

[0034] (1) Figure 2 The present invention combines a deep generative model with a discriminative prediction model, by first generating similar structures based on an existing data set and then performing multi-layer judgment and screening on the generated results, thereby making full use of the existing PeLEDs additive molecule database and achieving effective development of PeLED additive molecules with limited experimental data.

[0035] (2) The development framework based on deep learning algorithms, which includes a molecular property prediction model and a variational autoencoder generation model, not only shows good performance in prediction and generation tasks, but also significantly reduces time and economic costs compared to traditional trial and error methods.

[0036] (3) In the screening part for high-performance additives, a method for jointly screening high-performance additives by combining the Gutmann donor number DN and the Kamlet and Taft hydrogen bond acceptor strengths β of the molecule is proposed, which can be used to quickly identify and screen high-performance additive molecules.

[0037] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0039] Figure 1 This is a flowchart of the automated screening of molecular additives in perovskite light-emitting diodes and photovoltaic cells based on artificial intelligence proposed in the present invention.

[0040] Figure 2 This is a detailed full-process framework diagram of the automated screening of molecular additives in perovskite light-emitting diodes and photovoltaic cells based on artificial intelligence proposed in the present invention.

[0041] Figure 3 This is the input structure diagram of the DN value prediction model.

[0042] Figure 4 It is a schematic diagram of the process of generating similar molecules using the VAE model proposed in the present invention.

[0043] Figure 5 Schematic diagram of the multi-process screening for ideal additives.

[0044] Figure 6 is a flow chart of the multi-process screening for ideal additives.

[0045] Figure 7 These are the performance results of different models in the process of modeling the number of molecular donors DN and the strength of hydrogen bond acceptors β.

[0046] Figure 8 This is a scatter plot of the DN and β distribution of molecules with different efficiencies after predicting the molecules in the original PeLED additive molecule database using the donor number DN and hydrogen bond acceptor strength β prediction model. DETAILED DESCRIPTION

[0047] Embodiments of the present invention are described in detail below. The embodiments are exemplary and intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0048] like Figure 1 As shown, the artificial intelligence-based automated screening method for molecular additives in perovskite light-emitting diodes and photovoltaic cells proposed in the present invention comprises the following steps:

[0049] Step 1: Build the database, including the following steps:

[0050] S100, establish the PeLED additive molecular database, molecular donor number DN and hydrogen bond acceptor strength β database:

[0051] By consulting literature and other means, we collected and established the reported additive molecular structures and experimental data, including the names of the additive molecular structures and the device EQE efficiency values. Based on the molecular structures, we used the computational chemistry library Pybel and the RDKit information chemistry package to calculate the corresponding molecular descriptors, molecular fingerprints and other data, including 208 molecular descriptors and MACCS, ECFP, and FP2 fingerprints, which together constitute the original PeLED additive molecular database.

[0052] Gutmann donor number DN data is given by literature reports; due to historical reasons, the donor number DN value is an empirical parameter, which was originally defined as the negative enthalpy of the formation of a 1:1 adduct between a Lewis base and a standard Lewis acid SbCl5 (antimony pentachloride) in a dilute solution of a non-coordinating solvent 1,2-dichloroethane with a DN of zero. Later, scholars found that it could be obtained by linear fitting of the chemical shift of nuclear magnetic resonance (NMR). Therefore, the database contains DN values ​​obtained by these two methods, as well as descriptors and MACCS fingerprint data calculated using the RDKit information chemical package and the PSI4 calculation program to form a solvent molecule donor number (DN) database. Specifically, Psi4 and RDKit chemical toolkits are used to calculate and select 59 molecular descriptors including the most negative atomic charge, HOMO energy level, and HOMO-LUMO energy gap, as well as 167 MACCS molecular fingerprint data, which are used to construct the solvent molecule donor number DN data set. Kamlet and Taft hydrogen bond acceptor strength β is intended to measure the strength of the hydrogen bond basicity or hydrogen bond acceptor strength of the solvent. The solvent hydrogen bond acceptor strength β data set is obtained in a similar form.

[0053] Step 2: Prediction model training:

[0054] S200, using the original PeLED additive molecular database and the molecular donor number DN and hydrogen bond acceptor strength β database to train the effectiveness prediction model, the donor number DN value and the hydrogen bond acceptor strength β prediction model respectively.

[0055] (1) Training effectiveness prediction model:

[0056] Experience shows that the efficiency of devices without any additives can reach about 10%. Therefore, molecules that increase the device efficiency by more than 10% after being added are considered effective additives. First, the fingerprints and descriptors of several different molecules are used as input features to establish an effectiveness prediction model to determine whether the additive is effective.

[0057] The model structure of the effectiveness prediction model is a multi-layer network structure with multiple inputs and a single output.

[0058] The model input is the molecular descriptor in the original PeLED additive molecule database obtained in step 1, as well as the molecular fingerprint and the hidden code of the molecule in the VAE model, which together constitute the representation characteristics of the additive molecule; and 0 or 1 labels are given as label input according to the effectiveness results of the molecule. The model output is the predicted probability of the effectiveness of the molecule.

[0059] When training the model, the loss function used is Binary Crossentropy, and the optimization algorithm is the standard back propagation algorithm, which aims to minimize the error between the predicted value and the label.

[0060] Specifically in this embodiment, the CalcDescriptors() method in the RDKit library is used to calculate all descriptor values ​​of all molecules in the Descriptors._descList list in the original PeLED additive database obtained in step 1, and the descriptors that are all 0 in all additive molecules are removed; further, the descriptors with a standard deviation that is too small (less than 0.2) are removed, and then the Pearson correlation coefficients are calculated between the remaining descriptors and the descriptors with correlations higher than the set value are removed. The remaining descriptors are finally combined with the molecular MACCS, ECFP and other fingerprints and the implicit codes of the molecules in the VAE model as the representation features of the additive molecules as the input of the predictive neural network model.

[0061] The label input is to label the collected additives as "effective" and "ineffective" according to the high and low device efficiency, and use "1" and "0" as labels. After the model is trained on the existing additive database, its accuracy on the validation set is observed to be over 90%. The trained effectiveness prediction model is used to predict the generated molecules and retain the more likely effective molecules based on the prediction results.

[0062] (2) Training the prediction model of donor number DN and hydrogen bond acceptor strength β:

[0063] The non-radiative effect caused by defects such as halogen vacancies that are common in perovskite films is the main reason for the limitation of EQE of PeLEDs. Small organic molecules containing lone pairs of electrons can control the crystallization kinetics and reduce the density of halogen vacancies by forming strong coordination with lead. Studies have shown that the coordination strength between additive molecules and lead ions is related to their own electron-donating ability. Molecules with strong electron-donating ability have stronger coordination ability with lead ions, thereby forming high-quality films. In addition, the strong hydrogen bonds formed between additives and organic cations will affect the crystallization and passivation processes, resulting in deterioration of device efficiency. In view of the regular influence of the electron-donating ability and hydrogen-bonding ability of additives on device efficiency, we consider whether we can further screen the generated molecules based on the quantification of these two properties. We choose to use the Gutmann donor number DN parameter to represent the electron-donating ability of additive molecules, and use Kamlet and Taft hydrogen bond acceptor strengths β to measure the hydrogen-bonding ability of molecules, and establish prediction models for these two molecular properties, respectively, hoping to play a role in the screening process of ideal additives. The model training details are as follows:

[0064] The MACCS molecular fingerprint data in the solvent molecule donor number DN data set obtained in step 1 and the molecular descriptors including the most negative atomic charge of the molecule, the molecular HOMO energy level, and the molecular HUMO-LUMO energy gap are used as model input, the molecular predicted DN (donor number) value is used as output, and the DN value measured by the standard experimental method is used as the target variable to establish a donor number DN value prediction model.

[0065] Specifically for this embodiment, Figure 3 As shown, for organic molecular solvents, the CalcDescriptors() method in the RDKit library is used to calculate the descriptor values, remove descriptors that are 0 in all molecules, and remove descriptors with a standard deviation that is too small (no more than 0.15), and then perform correlation deduplication. Finally, 56 descriptors remain, including the maximum EState index (MaxEStateIndex) in the molecule, the minimum EState index (MinEStateIndex) in the molecule, and the molecular weight (Molecular Weight). Then, the PSI4 chemical information calculation program is used to calculate the three molecular property indicators of the most negative atomic charge of the molecule, the donor orbital energy (HOMO energy level), and the HOMO-LUMO energy gap, and add them to the molecular descriptor data. Then, normalization and interpolation are performed to limit the value range of all descriptors to between -1 and 1 so that the influence of each feature on the model is at the same order of magnitude, and the normalized descriptor data is combined with the molecular MACCS fingerprint as the representation feature of the molecule.

[0066] For the established donor number DN prediction model, five-fold cross validation was used for model training, and the average of the model predictions in the five cross validations was used as the final model performance.

[0067] The specific steps include: evenly dividing the solvent molecule donor number DN dataset into five independent subsets; selecting one subset as the test set each time, and merging the other four subsets as the training set; training the model on the training set and evaluating the performance of the model on the corresponding test set, repeating this process five times, selecting a different subset as the test set each time; finally, averaging the output results of the five models to produce the final prediction output. By integrating the model predictions on different test sets, the impact of the accidental error of a single model can be effectively reduced, thereby improving the overall stability and accuracy of the prediction results.

[0068] In this embodiment, Figure 7 As shown in Figure 2, we train models such as Ridge Regression, Random Forest, Support Vector Machine, Gradient Boosting, and Neural Network and observe the training results. We find that the prediction model based on the neural network performs best on the validation set. Specifically, the regression determination coefficient R 2 The average absolute error (MAE) was 0.84, the mean square error (MSE) was 4.21, the mean square error (MSE) was 33.59, and the root mean square error (RMS) was 5.64. The trained DN value prediction model based on the neural network was used to regress and predict the DN values ​​of the new solvents in the collected test set. The performance of the trained gradient boosting prediction model and the neural network prediction model in predicting the DN values ​​of the molecules in the test set is shown in Table 1. The mean square error (MSE) was calculated to be 8.1, which is acceptable.

[0069] Table 1. Prediction results of DN value prediction model for test set molecules

[0070]

[0071] The prediction modeling of the strength of hydrogen bonds of solvent molecules is similar to the modeling process of donor number DN. Figure 3 A similar approach was used to obtain the molecular features and use them as model inputs, and cross-validation was also performed to evaluate and obtain the best prediction model, e.g. Figure 7As shown in Figure d, training comparisons were carried out on structures such as Ridge Regression, Random Forest, Support Vector Machine, Gradient Boosting, eXtreme Gradient Boosting, and Neural Network. The eXtreme Gradient Boosting model performed the best, with the lowest root mean square error RMSE of about 0.13.

[0072] Step 3: Generate similar new molecules:

[0073] S300, using the pre-trained molecular generation model to generate similar new molecules for the molecules in the original PeLED additive molecular database, thereby obtaining a generated molecule set.

[0074] The pre-trained molecular generative model uses a variational autoencoder (VAE). A variational autoencoder (VAE) is a generative model that generates data by encoding input data into an implicit latent space and decoding from the space to reconstruct the input data. The variational autoencoder VAE model in this embodiment is pre-trained on a large ZINC dataset.

[0075] Based on the pre-trained variational autoencoder VAE model, the SMILES string corresponding to each molecular structure in the original PeLED additive molecule database is input into the encoder of the autoencoder model. After the encoder code compression processing, the latent code of the original additive molecule in the 196-dimensional latent space is obtained. Random perturbations are applied to the latent code to obtain the perturbation code.

[0076] like Figure 4 As shown, the SMILES string is first converted into a one-hot vector and then further input into a fully connected layer containing 196 neurons through multiple layers of 1D convolutional layers. After standardization, a 196-dimensional latent code is obtained. After applying random small perturbations to the latent code, a series of new perturbation codes are obtained. After the new perturbation code is decoded by the decoder, a new SMILES string similar to the molecular structure of the original PeLED additive molecule database can be obtained.

[0077] The implementation is as follows: Repeat the input latent code multiple times to form a new data array Z, and the number of repetitions is determined by the number of decoding attempts. Add random noise, which is generated by drawing values ​​from a standard normal distribution (mean 0 and standard deviation 1), ensuring that each generated value is random and independent. Perform normalization on the generated noise, including calculating the norm of the noise array and dividing each noise element by this norm. This normalization step ensures that the total norm of the noise vector is standardized to 1, allowing its amplitude to be adjusted without changing the direction of the data. Apply a fixed amplitude noise to all data copies, and each data copy will have the same amplitude of noise added. After the above processing, the perturbed array Z is fed into the decoder, and the new molecule SMILES string is generated through the decoding process.

[0078] Step 4: Perform multiple rounds of screening on the generated results targeting the desired additive properties:

[0079] S400, performing multiple rounds of screening processes targeting ideal additives on the generated molecule set obtained in step 3. Figure 5 and Figure 6 The steps shown include:

[0080] S111, structural rationality screening:

[0081] The MolFromSmiles() method in the RDKit library is used to process the decoded new molecular SMIELS string. This method converts the SMILES string into a Mol object in RDKit. During the conversion process, the chemical rationality of the SMILES is preliminarily checked and the unreasonable parts are eliminated.

[0082] Specifically, during the conversion process, we first check whether the generated SMILES string follows the correct SMILES syntax and rules, and then check whether the connection of each atom conforms to the normal chemical valence state. For example, oxygen atoms usually form two bonds, carbon atoms can form four bonds, etc. If the number of connections of an atom exceeds its common valence state, the SMILES parsing process will fail.

[0083] S112, effectiveness screening:

[0084] The SMIELS character string that can be successfully converted in S111 is input into the effectiveness prediction model obtained in step 2 to obtain a effectiveness prediction probability value ranging from 0 to 1, and the molecular structures with effectiveness prediction probabilities greater than 0.7 are selected. It is believed that these molecules are more likely to be effective molecules that can improve device performance and constitute the first candidate molecule set.

[0085] S113, donor number DN and hydrogen bond acceptor strength β screening:

[0086] The molecules in the first candidate molecule set obtained in S112 are respectively input into the donor quantity DN value prediction model and the hydrogen bond acceptor strength β prediction model obtained in step 2, and the DN prediction value and β prediction value of each molecule are output, and the molecular structures with DN values ​​and β values ​​within a certain range are selected to form the second candidate molecule set.

[0087] Since the DN value describes the electron-donating ability of a molecule, the larger the DN value of the additive, the stronger its ability to bind to lead ions, which in turn affects the film morphology and crystallization process. In addition, the DN value of the solvent will affect the stability of the precursor solution, and if the DN value is too small, it will be difficult to form a stable precursor. At the same time, the strong hydrogen bonds formed between the additive and the organic cations will affect the crystallization and passivation processes, leading to deterioration of the device efficiency. Therefore, the hydrogen bonding ability of the additive molecules also needs to be within an appropriate range.

[0088] The corresponding properties of the molecules in the original PeLED additive molecule database were predicted using the donor number DN and hydrogen bond acceptor strength β prediction model and the results were obtained, such as Figure 8 As shown, a scatter plot is made with the predicted donor number DN and hydrogen bond acceptor strength β decibel of each molecule as the horizontal and vertical coordinates. The color of the point represents the efficiency of the corresponding device. The darker the color, the lower the device efficiency. It can be seen from the figure that high-performance additives with EQE efficiency greater than 20% are concentrated in a certain area in the upper right corner of the figure, while the EQE efficiency of additives in other areas is relatively low. In view of this phenomenon, considering the error of the prediction model, we propose to use the area representing DN values ​​30 to 50 and β values ​​0.8 to 1.2 as the possible distribution range of high-performance additives, and eliminate the generated molecules whose prediction results are not in this range. In this way, rapid screening of high-performance additives can be achieved.

[0089] Therefore, in order to screen out possible high-performance additives, the molecular structures whose DN and β values ​​are not within the specified range are selected and eliminated, and the remaining molecules constitute the second candidate molecular set.

[0090] S114, Availability and Purchasability Screening:

[0091] PubChem API was used to obtain the supplier information of the molecules in the second candidate molecule set obtained in step S113, and the molecules with complete supplier information and quotations within the set range were selected to form the third candidate molecule set, and the final number was about 10.

[0092] The molecules in the third candidate molecule set and the experimental candidate additive molecules are used for device preparation and experimental testing to obtain the JV curve and other test results of the corresponding device. The test results are added to the original PeLED additive molecule database to update the prediction model training, similar molecule generation, and multiple rounds of screening processes with ideal additive properties as the goal.

[0093] In summary, the present invention can be used for the development and design of PeLED additive molecules, and the known implementation methods and common sense of characteristics in the above-mentioned schemes are not described in detail here. It should be pointed out that for those skilled in the art, several improvements can be made without departing from the present invention, and these should also be regarded as the scope of protection of the present invention, which will not affect the implementation effect of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of the claims, and the specific implementation methods and other records in the specification are used to interpret the content of the claims.

Claims

1. An automated screening method for molecular additives in perovskite light-emitting diodes and photovoltaic cells, characterized in that: The following steps are involved: Step 1: Establish a database of PeLED additive molecules, molecular donor quantity DN, and hydrogen bond acceptor strength β; Step 2: Use the PeLED additive molecular database and the molecular donor quantity DN and hydrogen bond acceptor strength β database to train the effectiveness prediction model, donor quantity DN value prediction model and hydrogen bond acceptor strength β prediction model respectively; Step 3: Use the pre-trained molecular generation model to generate similar new molecules for the molecules in the PeLED additive molecular database to obtain a generated molecule set; Step 4: Based on the effectiveness prediction model trained in step 2, the prediction model of the donor quantity DN value and the hydrogen bond acceptor strength β, the generated molecule set obtained in step 3 is subjected to multiple rounds of screening with the ideal additive as the target.

2. The automated screening method for molecular additives in perovskite light-emitting diodes and photovoltaic cells according to claim 1, characterized in that: In step 1, the PeLED additive molecule database is obtained by: The reported additive molecular structures and experimental data are collected and established, including the names of the additive molecular structures and the device EQE efficiency values. Based on the molecular structures, the computational chemistry library Pybel and the RDKit information chemistry package are used to calculate the corresponding molecular descriptors and molecular fingerprint data, which together constitute the original PeLED additive molecular database.

3. The automated screening method for molecular additives in perovskite light-emitting diodes and photovoltaic cells according to claim 1, characterized in that: In step 2, the model structure of the effectiveness prediction model is a multi-layer network structure with multiple inputs and a single output; the model input is the molecular descriptor in the original PeLED additive molecule database obtained in step 1, as well as the molecular fingerprint and the latent code of the molecule in the VAE model, which together constitute the representation characteristics of the additive molecule; and 0 or 1 labels are given as label inputs according to the effectiveness results of the molecules; the model output is the predicted probability of the effectiveness of the molecules.

4. The method for automated screening of molecular additives in perovskite light-emitting diodes and photovoltaic cells according to claim 3, characterized in that: The model input of the effectiveness prediction model is obtained through the following process: The CalcDescriptors() method in the RDKit library is used to calculate the values ​​of all descriptors in the Descriptors._descList list for all molecules in the PeLED additive database obtained in step 1, and the descriptors whose value is 0 in all additive molecules and the descriptors whose standard deviation is less than the set value are removed; then the Pearson correlation coefficient is calculated between the remaining descriptors and the descriptors whose correlation is higher than the set value are removed; the remaining descriptors are combined with the molecular fingerprint and the latent code of the molecule in the VAE model as the representation features of the additive molecules and the input of the effectiveness prediction model.

5. The method for automated screening of molecular additives in perovskite light-emitting diodes and photovoltaic cells according to claim 1, characterized in that: In step 2, the MACCS molecular fingerprint data in the molecular donor quantity DN data set obtained in step 1 and the molecular descriptors including the most negative atomic charge of the molecule, the molecular HOMO energy level, and the molecular HUMO-LUMO energy gap are used as model inputs, the molecular predicted DN value is used as output, and the DN value measured by the standard experimental method is used as the target variable to establish a donor quantity DN value prediction model.

6. The method for automated screening of molecular additives in perovskite light-emitting diodes and photovoltaic cells according to claim 1, characterized in that: In step 2, the donor quantity DN value prediction model is trained using five-fold cross validation, and the average value of the model predictions in five cross validations is used as the final model performance.

7. The method for automated screening of molecular additives in perovskite light-emitting diodes and photovoltaic cells according to claim 1, characterized in that: In step 3, the pre-trained molecular generation model adopts a variational autoencoder and is pre-trained on the ZINC dataset.

8. The method for automated screening of molecular additives in perovskite light-emitting diodes and photovoltaic cells according to claim 7, characterized in that: The input of the variational autoencoder is the SMILES string corresponding to each molecular structure in the PeLED additive molecule database; after the SMILES string is converted into a one-hot vector, it is further input into the fully connected layer through a multi-layer 1D convolutional layer, and the latent code is obtained after standardization. After applying random perturbations to the latent code, a series of new perturbation codes are obtained. After the new perturbation code is decoded by the decoder, a new SMILES string similar to the molecular structure of the PeLED additive molecule database can be obtained.

9. The method for automated screening of molecular additives in perovskite light-emitting diodes and photovoltaic cells according to claim 1, characterized in that: Screening in step 4 includes the following steps: Step 4.1: Structural rationality screening: Use the MolFromSmiles() method in the RDKit library to process the decoded new SMIELS string, convert the SMILES string into a Mol object in RDKit, check the chemical rationality of the SMILES during the conversion process, and remove the unreasonable parts; Step 4.2: Effectiveness screening: Input the SMIELS character string that can be successfully converted in step 4.1 into the effectiveness prediction model obtained in step 2 to obtain an effectiveness prediction probability value ranging from 0 to 1, and select the molecular structure whose effectiveness prediction probability is greater than the set threshold to form the first candidate molecule set; Step 4.3: Screening of donor number DN and hydrogen bond acceptor strength β: Input the molecules in the first candidate molecule set obtained in step 4.2 into the donor quantity DN value prediction model and hydrogen bond acceptor strength β prediction model obtained in step 2, output the DN prediction value and β prediction value of each molecule, and select the molecular structures with DN value and β value within a certain range to form the second candidate molecule set; Step 4.4: Availability and affordability screening: The PubChem API was used to obtain the supplier information of the molecules in the second candidate molecule set obtained in step 4.3, and the molecules with complete supplier information and quotations within the set range were selected to form the third candidate molecule set.

10. The method for automated screening of molecular additives in perovskite light-emitting diodes and photovoltaic cells according to claim 9, characterized in that: For the molecules in the third candidate molecule set, device preparation and experimental testing are carried out to obtain the test results of the corresponding devices; and the test results are added to the PeLED additive molecule database to update the prediction model training, similar molecule generation, and multiple rounds of screening processes with ideal additive properties as the goal.

Citation Information

Patent Citations

  • Screening method of perovskite cell modification layer material

    CN117894409A

  • Method for designing drug molecules based on DEL screening data, electronic equipment and storage medium

    CN118335237A

  • Systems and methods for predicting potential inhibitors of target protein

    US20220084627A1

Cited By

  • Perovskite solar cell photoelectric conversion efficiency model construction method and efficiency prediction method

    CN119626379A