Perovskite battery performance prediction method, model training method and system

Abstract

The invention provides a perovskite battery performance prediction method and a model training method and system.The training method comprises the steps that multiple sets of perovskite battery training samples are obtained, the training samples comprise preset process parameters, preset performance parameters and feature estimation parameters, the preset process parameters comprise at least one feature, and the preset performance parameters comprise at least one feature; the feature estimation parameters are obtained by performing attribute estimation on features in the preset process parameters; training is carried out in at least one basic model according to the multiple sets of training samples, the preset technological parameters and the corresponding feature estimation parameters serve as input of the basic model, and the preset performance parameters serve as output of the basic model; determining a training result of the at least one basic model; at least two basic models with training results meeting preset conditions are selected from the at least one basic model to construct a stacking model, a perovskite battery performance prediction model is obtained, and the performance of the perovskite battery can be effectively improved based on the technological parameters determined by the perovskite battery performance prediction model.

Description

Technical Field

[0001] The present disclosure relates to the field of batteries, and in particular to a method for predicting the performance of perovskite batteries, a model training method and a system. Background Art

[0002] Under today's energy challenges, perovskite photovoltaic solar cells (PSCs) as an emerging photovoltaic technology have attracted much attention from researchers. Their high power conversion efficiency (PCE) and relatively low preparation cost have made them one of the important candidates for replacing traditional energy sources.

[0003] However, current research mainly focuses on the impact of a single process on battery performance and stability. Traditional experimental methods usually require a lot of time and cost, and the battery process involves many complex steps and parameters that need to be accurately controlled and adjusted, which increases the possibility of error and limits the improvement of battery efficiency. Summary of the invention

[0004] In view of the problem that traditional experimental methods in the prior art limit the improvement of battery efficiency, the present invention proposes a prediction method, a model training method and a system for perovskite battery performance.

[0005] In a first aspect, an embodiment of the present disclosure provides a method for training a perovskite battery performance prediction model, the training method comprising:

[0006] Acquire multiple groups of training samples of perovskite cells, wherein the training samples include preset process parameters, preset performance parameters, and feature estimation parameters, the preset process parameters include at least one feature, and the feature estimation parameters are obtained by estimating the properties of the features in the preset process parameters;

[0007] Training is performed in at least one basic model according to the plurality of training samples, wherein preset process parameters and corresponding feature estimation parameters are used as inputs of the basic model, and preset performance parameters are used as outputs of the basic model;

[0008] determining a training result of at least one base model;

[0009] At least two basic models whose training results meet preset conditions are selected from at least one basic model to construct a stacking model to obtain a perovskite battery performance prediction model.

[0010] In one embodiment, the preset process parameters include: one or more of: A-site cations, X-site anions, perovskite thickness, perovskite deposition solvent, solvent mixing ratio, thermal annealing temperature, and thermal annealing time.

[0011] In one embodiment, the base model includes one or more of: decision tree, random forest, CatBoost, LassoLarsCV, histogram gradient boosting, extreme gradient boosting, K nearest neighbor, ridge regression and linear regression.

[0012] In one embodiment, before training at least one basic model based on multiple sets of training samples, the method includes:

[0013] The hyperparameters of at least one base model are optimized by 10-fold cross validation.

[0014] In one embodiment, the training method includes:

[0015] Selecting a preset proportion of training samples from multiple groups of training samples as test samples of at least one basic model;

[0016] Determine the training results of at least one base model, including:

[0017] At least one basic model is tested according to the test sample, and a training result of the at least one basic model is calculated.

[0018] In one embodiment, the training results include: one or more of: coefficient of determination, root mean square error, mean absolute error, and mean absolute percentage error.

[0019] In one embodiment, the stacked model includes a base learning layer and a meta-learning layer connected to each other;

[0020] Selecting at least two basic models whose training results meet preset conditions from at least one basic model to construct a stacking model, and obtaining a perovskite battery performance prediction model, including:

[0021] Select at least two basic models whose training results meet preset conditions from at least one basic model, and construct a basic learning layer of the stacked model according to preset weights;

[0022] At least one base model is selected from the at least two selected base models to construct a meta-learning layer of the stacked model.

[0023] In a second aspect, the present disclosure provides a method for predicting the performance of a perovskite battery, the prediction method comprising:

[0024] Obtaining a perovskite battery performance prediction model, the perovskite battery performance prediction model is trained according to the training method for the perovskite battery performance prediction model of any one of the first aspects;

[0025] Analyze the process parameters and performance parameters of the perovskite battery performance prediction model, and determine the importance of each feature in the process parameters to the performance parameters based on the analysis results;

[0026] Determine the value range of each feature in the process parameters based on importance.

[0027] In a third aspect, the present disclosure provides a training system for a perovskite battery performance prediction model, the training system comprising:

[0028] A first acquisition module is used to acquire a plurality of groups of training samples of perovskite batteries, wherein the training samples include preset process parameters, preset performance parameters and feature estimation parameters, the preset process parameters include at least one feature, and the feature estimation parameters are obtained by performing attribute estimation on the features in the preset process parameters;

[0029] A training module, used for training at least one basic model according to multiple sets of training samples, wherein preset process parameters and corresponding feature estimation parameters are used as inputs of the basic model, and preset performance parameters are used as outputs of the basic model;

[0030] A first determination module, used to determine a training result of at least one basic model;

[0031] The stacking module is used to select at least two basic models whose training results meet preset conditions from at least one basic model to construct a stacking model to obtain a perovskite battery performance prediction model.

[0032] In a fourth aspect, an embodiment of the present disclosure provides a prediction system for perovskite battery performance, the prediction system comprising:

[0033] A second acquisition module is used to acquire a perovskite battery performance prediction model, where the perovskite battery performance prediction model is trained according to a training system for a perovskite battery performance prediction model according to any one of the third aspects;

[0034] A second determination module is used to analyze the process parameters and performance parameters of the perovskite battery performance prediction model, and determine the importance of each feature in the process parameters to the performance parameters based on the analysis results;

[0035] The value module is used to determine the value range of each feature in the process parameter according to its importance.

[0036] The present disclosure provides a method for predicting the performance of a perovskite battery, a model training method and a system, which have the following technical effects: first, by introducing characteristic parameters, the generalization ability of the model is improved, and the accuracy of the model is improved; second, multiple basic models are trained, and basic models are selected based on the training results to stack to obtain a stacked model, so as to improve the accuracy of the model. Combining the advantages of multiple basic models, the models are complementary to each other to a certain extent. The perovskite battery performance prediction model in this embodiment can accurately obtain the performance parameters according to the process parameters of the perovskite battery, and the process parameters determined based on the perovskite battery performance prediction model can effectively improve the performance of the perovskite battery.

[0037] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings:

[0039] Figure 1 A flowchart of a method for training a perovskite battery performance prediction model provided by an exemplary embodiment of the present disclosure;

[0040] Figure 2 A schematic diagram of the relationship between the characteristic quantity of a preset process parameter and a preset performance parameter provided for an exemplary embodiment of the present disclosure;

[0041] Figure 3 A schematic diagram of a model training result of whether a feature parameter is input into a model according to an exemplary embodiment of the present disclosure;

[0042] Figure 4 A schematic diagram of training results of a basic model and a stacked model provided for an exemplary embodiment of the present disclosure;

[0043] Figure 5 A flowchart of a method for predicting perovskite battery performance provided by an exemplary embodiment of the present disclosure;

[0044] Figure 6 A schematic diagram of the Pearson correlation coefficient between the characteristics of process parameters and performance parameters provided for an exemplary embodiment of the present disclosure;

[0045] Figure 7 A schematic diagram of a SHAP interpreter output value based on a stacking model provided for an exemplary embodiment of the present disclosure;

[0046] Figure 8 A violin plot of characteristics of process parameters and performance parameters provided for an exemplary embodiment of the present disclosure;

[0047] Fig. 9 A waterfall diagram showing the predicted performance parameters obtained by adding PbI2 as an excess non-stoichiometric amount for an exemplary embodiment of the present disclosure;

[0048] Fig.10 A schematic diagram of perovskite cell performance with two sets of different process parameters provided for an exemplary embodiment of the present disclosure;

[0049] Fig.11 A module diagram of a training system for a perovskite battery performance prediction model provided by an exemplary embodiment of the present disclosure;

[0050] Fig.12 A module diagram of a system for predicting perovskite battery performance provided by an exemplary embodiment of the present disclosure;

[0051] Fig.13 The present invention provides a structural diagram of an electronic device according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present disclosure may be combined with each other without conflict.

[0053] The following detailed descriptions are all exemplary descriptions, and are intended to provide further detailed descriptions of the present disclosure. Unless otherwise specified, all technical terms used in the present disclosure have the same meanings as those generally understood by those skilled in the art to which the present disclosure belongs. The terms used in the present disclosure are only for describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present disclosure.

[0054] Optimizing the preparation of perovskite films is one of the key points in the research field, because the quality of perovskite films directly affects the performance and stability of photovoltaic cells. However, due to the relatively complex preparation process of photovoltaic cells, the morphology of perovskite films is difficult to accurately control, which in turn limits the improvement of battery performance. In order to achieve better battery performance, each step of the process must be optimized. In response to this problem, researchers have conducted a lot of research on the preparation process of PSCs. According to the researchers' research results, the following parameters: molecular materials, cations / anions of similar size, excess non-stoichiometric components (Ensc) (such as non-stoichiometric PbI2, excess methylammonium bromide MABr, etc.), perovskite deposition solvent (Pds), optimization of perovskite layer thickness, thermal annealing time and thermal annealing temperature, lead cesium chloride (CspbCl3) film, electron transport layer (ETL) thickness and back contact thickness, etc., are very important for achieving efficient and stable PSCs and improving PSCs performance.

[0055] When it comes to optimizing experimental methods that require a lot of time and resources for trial and error, machine learning (ML) has demonstrated amazing capabilities in perovskite battery performance prediction and multi-scale data analysis. It can quickly shorten the experimental cycle, save a lot of resources, and reduce interference from human factors, and has huge application potential.

[0056] However, current research using machine learning (ML) technology lacks the impact of characteristic dimensions of multiple process parameters on the cell performance of perovskite solar cells, but mostly focuses on a single characteristic dimension.

[0057] Based on this, an exemplary embodiment of the present disclosure provides a training method for a perovskite battery performance prediction model, see Figure 1 , training methods include:

[0058] S11. Obtain multiple groups of perovskite battery training samples.

[0059] The training samples include preset process parameters, preset performance parameters and feature estimation parameters, and the feature estimation parameters are obtained by estimating the properties of the features in the preset process parameters.

[0060] Regarding the preset process parameters, the preset process parameters include at least one feature, including but not limited to A-site cations (Perovskite composition Aions, Pc-A), X-site anions (Perovskite composition X ions, Pc-X), perovskite thickness (Perovskite thickness, Pt), perovskite deposition solvents (Perovskitedeposition solvents, Pds), solvent mixing ratios (Solvents mixing ratios, Smr), thermal annealing temperature (Thermal annealing temperature, Ta-te), thermal annealing time (Thermal annealing time, Ta-ti), back contact material (Back contact material, Bm), back contact deposition procedure (Back contact deposition procedure, Bdp), etc. The determination of the above characteristics is based on experiments on their influence on the performance parameters. In this embodiment, the characteristics of the above 9 dimensions are preferably used, but not limited to this.

[0061] Regarding preset performance parameters, the preset performance parameters include but are not limited to Voc (open circuit voltage), Jsc (short circuit current), FF (fill factor), and preferably photoelectric conversion efficiency (PCE).

[0062] Regarding feature estimation parameters, feature estimation parameters are coarse property estimates (CEP), which is a strategy that uses less accurate and low-cost property estimates along with chemical descriptors to improve prediction accuracy by providing additional information to the model. Using CEP as a descriptor can effectively improve the predictive ability of machine learning models to the state-of-the-art level.

[0063] Specifically, since the data scale does not directly affect the model accuracy when processing a data set, but is adjusted by the degrees of freedom (DoF) of the model, increasing the degrees of freedom of the model can improve the accuracy, thereby enabling the model to better adapt to the complexity of the data. However, increasing the degrees of freedom also increases the risk of overfitting, so a trade-off needs to be made between the complexity and generalization ability of the model. Specifically, due to the limited number of samples in a small data set, an overly complex model may cause the model to overfit the data and lose its generalization ability. Therefore, the degrees of freedom of the model usually need to be properly controlled to avoid overfitting of the model. In order to build an accurate machine learning model on a small-scale data set, a rough attribute estimate (CEP) can be introduced in the feature space.

[0064] S12. Perform training in at least one basic model based on multiple groups of training samples.

[0065] The preset process parameters and the corresponding feature estimation parameters are used as the input of the basic model, and the preset performance parameters are used as the output of the basic model.

[0066] Specifically, the basic model includes but is not limited to: at least one of decision tree (DT), random forest (RF), CatBoost, LassoLarsCV, histogram gradient boosting (HBGR), extreme gradient boosting (XGBoost), K nearest neighbor (KNN), ridge regression (Ridge) and linear regression (Linear R).

[0067] S13. Determine the training result of at least one basic model.

[0068] The training results include but are not limited to any one or more of the following: 2 ), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE).

[0069] The calculation formula for the above training results is as follows:

[0070]

[0071] Among them, n is the total number of data, x i and x' i Represent each true value and predicted value respectively, and represent the average of the true value and the predicted value respectively.

[0072] R 2 The value is the determination coefficient of the model, which is used to evaluate the degree of fit of the model to the training samples and test samples. 2 It reflects the degree to which the model explains the variance of the observed data. Its value ranges from 0 to 1. The closer it is to 1, the better the model fits. Secondly, we calculated RMSE, MAE, and MAPE to evaluate the prediction accuracy of the model. RMSE measures the average difference between the model prediction value and the actual observation value. The smaller the RMSE value, the higher the prediction accuracy of the model. MAE measures the average absolute value of the prediction error, reflecting the fitting ability of the model. The smaller the MAE value, the higher the model fit and the higher the accuracy. MAPE measures the percentage of the prediction error relative to the actual value. The smaller the MAPE value, the higher the accuracy of the model. Through these evaluation indicators, we can fully understand the performance of the model.

[0073] S14, selecting at least two basic models whose training results meet preset conditions from at least one basic model to construct a stacking model to obtain a perovskite battery performance prediction model.

[0074] For the above basic model, there are the following advantages and disadvantages:

[0075] The DT model is based on a tree structure, which has the advantages of strong flexibility, high efficiency, and simple structure. It builds a tree by gradually segmenting the features to predict the performance of perovskite solar cells. However, when the decision tree becomes too complex, it is prone to overfitting.

[0076] The RF model adopts an integrated learning method, using multiple decision tree models for prediction at the same time and comprehensively considering their results to improve prediction accuracy. However, compared with a single DT, it is difficult for RF to understand the reasoning process behind each tree, resulting in poor interpretability.

[0077] CatBoost is a machine learning model based on gradient boosting. It has the ability to automatically process categorical variables, can reduce overfitting, and is suitable for processing complex relationships in perovskite solar cell data. HBGR boosting is an integrated learning model that builds a model by combining multiple decision trees (DTs). It runs faster and is suitable for scenarios with high real-time requirements. However, when processing small-scale data sets, it may cause overfitting problems due to the complexity of the model.

[0078] The LassoLarsCV model uses lasso regression and cross-validation methods to regularize model parameters to avoid overfitting. Since it is based on a linear model, it may perform poorly for data with nonlinear relationships.

[0079] The XGBoost model uses a gradient boosting tree model to improve prediction results by building multiple regression trees in parallel and iteratively optimizing the model. If the selected hyperparameters are inappropriate or the sample data is too small, XGBoost is prone to overfitting, resulting in poor performance of the model on the test samples. The KNN model uses an instance-based learning method to make predictions by finding the training data samples that are most similar to the new data points. It is applicable to a variety of data types, but it is sensitive to outliers and noise, which may affect the stability of the model.

[0080] By introducing regularization terms, Ridge can effectively control the complexity of the model, prevent overfitting, and improve the generalization ability of the model. The choice of regularization parameter alpha has a great impact on the performance of ridge regression, and hyperparameter tuning is required to obtain the best results.

[0081] The Linear R model is simple and intuitive, easy to understand and explain, can capture the characteristics of linear relationships, and is suitable for some simple prediction problems.

[0082] In order to overcome the problems that a single basic model is limited by its inherent structure and assumptions during training, resulting in insufficient generalization ability, sensitivity to noise, and limited scenario applicability, we built a stacked model. The stacked model is an integrated learning method with the advantages of high flexibility, superior overall performance, strong ability to combat overfitting, and easy explanation. It can effectively improve the generalization ability of the model and is applicable to various types of prediction problems. The stacked model is an interconnected basic learning layer and meta-learning layer. The basic learning layer contains multiple independent basic models, such as decision trees, random forests, support vector machines, etc. Each basic model predicts the input data and passes the prediction results to the meta-model layer. The meta-model layer receives these prediction results as input and generates the final prediction results.

[0083] In one embodiment, the CEP is determined by the non-stoichiometric composition (Ensc) and the perovskite additive compound (Pac).

[0084] For non-stoichiometric components, in perovskite solar cells, adding excessive non-stoichiometric components can make the film surface morphology smoother and prolong the carrier lifetime; Liu, J et al. showed that non-stoichiometry in the preparation of perovskite solar cells helps uniform nucleation and regulates the growth dynamics during crystallization; Xiang, S et al. added non-stoichiometry to carbon-based perovskite solar cells to achieve high performance and stability of the battery. The above studies show that non-stoichiometric components play an important role in improving the performance of batteries.

[0085] For perovskite additive compounds (Pac), the use of perovskite additive compounds (Pac) can affect the crystallization and film formation of perovskite, as well as the defect passivation in the body and on the surface, and affect the interface regulation of structure and energetics

[46] . Therefore, additives are widely used in efficient, stable and hysteresis-free perovskite solar cells, and play an important role in improving the performance of perovskite solar cells.

[0086] Therefore, in this embodiment, an excess of a non-stoichiometric component (Ensc) and a perovskite additive compound (Pac) are selected as CEP.

[0087] In order to verify the relationship between the characteristics of the input preset process parameters and the preset performance parameters, R 2 Fitting with RMSE and number of features, such as Figure 2 As shown, we can see that as the number of features increases, R 2 The R value increased from 0.408 to 0.836, while the RMSE decreased from 2.489 to 1.309. 2 From 0.383 to 0.816, and RMSE from 2.555 to 1.432. After adding the last two features as CEP, R 2 The changes of ΔE and RMSE tend to be flat, which proves that it is feasible to improve the model prediction ability by incorporating non-stoichiometric components (Ensc) and perovskite additive compounds (Pac) as CEPs.

[0088] In addition, in order to compare whether the performance parameters of the stacking model are improved by adding excess non-stoichiometric components (Ensc) and perovskite additive compounds (Pac) as CEP, we compared the performance parameter output results with and without CEP based on the stacking model, as shown in the following table and Figure 3 , showing the evaluation indicators of the model on the training set and the test set. Obviously, after adding CEP, the R 2 The improvement on both the training set and the test set was significant, and the error was also reduced. The improvement on the test set was particularly significant. 2 It was improved by 16.14%, and the errors of RMSE, MAE, and MAPE were reduced by 20.44%, 17.76%, and 20.20%, respectively.

[0089]

[0090] Table 1 Evaluation indicators of the model

[0091] In one embodiment, before step S12 trains the model, the hyperparameters of the basic model need to be processed, specifically including:

[0092] The hyperparameters of at least one base model are optimized by 10-fold cross validation.

[0093] In one embodiment, before step S12 trains the model, it is necessary to preprocess the training samples of the basic model, which specifically includes:

[0094] Data preprocessing is performed on the obtained multiple groups of training samples.

[0095] Specifically, the training samples with missing values ​​are deleted, and then the character data is processed in a sequential coding manner. For example, this embodiment obtains 454 sets of training samples, and finally retains 145 sets of training samples.

[0096] In one embodiment, before step S13, the training method further includes:

[0097] A preset proportion of training samples are selected from the multiple groups of training samples as test samples of at least one basic model.

[0098] The preset ratio can be determined according to actual conditions and is not particularly limited here. The preset ratio used in this embodiment is 20%.

[0099] Step S13 specifically includes:

[0100] At least one basic model is tested according to the test sample, and a training result of the at least one basic model is calculated.

[0101] The training result in this embodiment is obtained by comparing the output of the basic model with the preset performance parameters in the test sample. For details, please refer to the calculation method of the training result in the above embodiment.

[0102] In one embodiment, step S14 specifically includes:

[0103] At least two basic models whose training results meet preset conditions are selected from at least one basic model, and a basic learning layer of the stacked model is constructed according to preset weights.

[0104] At least one base model is selected from the at least two selected base models to construct a meta-learning layer of the stacked model.

[0105] For example, when the training result includes a determination coefficient, a root mean square error, or a mean absolute error, the preset conditions may include: the determination coefficient is greater than a first preset value, the root mean square error is less than a second preset value, and the mean absolute error is less than a third preset value. The value ranges of the first preset value, the second preset value, and the third preset value may be selected according to the specific circumstances. The preset weight may be selected according to the number of selected basic models. For example, when two basic models are selected, the preset weight may be 1:1.

[0106] The above embodiment is described below through a specific implementation method:

[0107] In order to explore the impact of perovskite technology on battery performance, we constructed a basic model based on the above 9 features and a stacking model. According to the experiment, compared with other models based on distance algorithms and linear regression algorithms, the performance of tree models is generally better, especially random forest, decision tree, Catboost, and XGBoost models. The determination coefficients of these four tree models are all greater than 0.8, while the root mean square error is less than 1.4 and the mean absolute error is less than 1.1.

[0108] For details, see Figure 4 ,like Figure 4 As shown in (a), the R2 performance of the three models DT, Catboost and XGB on the training samples is the best, all 0.837, followed by the RF model with 0.814. On the test samples, the RF model has the highest R2 of 0.833, followed by the DT model with 0.795. The error analysis of the training samples and the test samples is shown in Figure 4 As shown in Figures (b) and 4(c), the DT model has an RMSE of 1.305, a MAE of 0.865, and a MAPE of 0.06 on the training samples, which is the lowest error among the 9 single models. On the test samples, RF has the lowest error, with an RMSE of 1.331, a MAE of 1.031, and a MAPE of 0.075. From the above analysis, it can be seen that the DT model performs best on the test samples, while the RF model performs best on the training samples.

[0109] Therefore, when we build the stacked model, we use the decision tree and random forest models as the meta-learning layer of the stacked model. Then, we input the two prediction results of the basic model into the random forest model as the meta-learning layer with a 1:1 weight. The resulting stacked model has the most stable overall performance on training and test samples. Its R2 on training samples is 0.836, RMSE is 1.309, MAE is 0.890, and MAPE is 0.061. Its R2 on test samples is 0.806, RMSE is 1.432, MAE is 1.144, and MAPE is 0.083. Figure 4 (d) is the relationship between the performance parameters (i.e., PCE) output by the stacking model and the preset performance parameters in the test sample. It can be seen that the stacking model can combine the advantages of a single model, so that the final perovskite battery performance prediction model can be complementary to a certain extent.

[0110] In this embodiment, firstly, the generalization ability of the model is improved by introducing characteristic parameters, and the accuracy of the model is improved; secondly, multiple basic models are trained, and basic models are selected based on the training results to stack to obtain a stacked model, so as to improve the accuracy of the model. Combining the advantages of multiple basic models, the models are complementary to each other to a certain extent. The perovskite battery performance prediction model in this embodiment can accurately obtain the performance parameters according to the process parameters of the perovskite battery.

[0111] An exemplary embodiment of the present disclosure provides a method for predicting the performance of a perovskite battery, see Figure 5 , prediction methods include:

[0112] S51. Obtain a perovskite battery performance prediction model.

[0113] The perovskite battery performance prediction model is trained according to the training method of the perovskite battery performance prediction model in the above embodiment.

[0114] S52. Analyze the process parameters and performance parameters of the perovskite battery performance prediction model, and determine the importance of each feature in the process parameters to the performance parameters based on the analysis results.

[0115] Among them, the process parameters are in the same data form as the preset process parameters in the above embodiment, and the performance parameters are in the same data form as the above embodiment. Please refer to the above embodiment and will not repeat them here. In addition, the perovskite battery performance prediction model will also determine the characteristic estimation parameters of the process parameters before processing the performance parameters. Please refer to the above embodiment for the specific implementation method.

[0116] S53. Determine the value range of each feature in the process parameter according to importance.

[0117] In one embodiment, step S52 may use Pearson correlation coefficient for analysis.

[0118] Specifically, in order to explore the correlation between the process parameters and performance parameters (i.e. PCE) of the battery, see Figure 6 , correlation analysis aims to determine whether there is a certain type of association between variables, as well as the strength and direction of this association. Its coefficient value range is usually between -1 and 1, where -1 indicates a complete negative correlation, 1 indicates a complete positive correlation, and 0 indicates no linear relationship. Figure 5 It can be seen that the addition of excess non-stoichiometric components to the perovskite composition has the highest correlation coefficient with PCE, reaching 0.61. Excessive non-stoichiometric components can make the growth of perovskite crystals more complete and uniform, reduce defects and interface reflections, increase the light absorption capacity of perovskite films, and thus improve PCE.

[0119] Therefore, in order to improve PCE, we must focus on the excess non-stoichiometric components added during the preparation of perovskite films, which is consistent with our expectation that the excess non-stoichiometric components (Ensc) are set as CEP to improve the model prediction ability. In addition, the correlation coefficients between PCE and A-site cations (Pc-A), X-site anions (Pc-X), and perovskite deposition solvents (Pds) are also high, which are -0.59, -0.51, and 0.53, respectively. The selection of A-site cations (Pc-A) can effectively adjust the energy gap, lattice constant, and stability of perovskites, which directly affects PCE. Adding cations / anions of similar sizes acts as a defect passivator

[16] . In addition, there are also significant correlations between the perovskite additive compound (Pac), the perovskite deposition solvent mixing ratio (Smr), the back electrode material (Bm), and the perovskite thickness (Pt). Common additive compounds include organic small molecules, ionic liquids and polymers. Additive compounds can adjust the crystal structure and grain size of perovskite films, improve carrier transport and collection efficiency, and thus improve the PCE and long-term stability of the battery. Different solvents and mixing ratios have different solubility, volatility and surface tension. Suitable solvents and ratios can affect the morphology, crystallinity and grain size of perovskite films, which directly affect PCE. The material of the back electrode (Bm) can affect the interface quality and charge transfer efficiency between the back electrode and other layers (such as perovskite layer, electron transport layer and anode). Suitable back electrode materials can optimize the matching and contact of these interfaces, reduce the recombination of electrons and holes, and improve the charge transfer and collection efficiency. Regarding the thickness of perovskite (Pt), generally speaking, the PCE of the battery will increase with the increase of film thickness, but too thick a film will lead to an increase in the recombination rate of carriers and a decrease in the transport length of carriers. A moderate film thickness can maximize the absorption of photons and promote photoelectric conversion. These factors constitute a huge variable space, making it very challenging to explore their correlation with battery performance.

[0120] In one embodiment, step S52 may be analyzed in a visual interpretation manner.

[0121] Specifically, SHAP interpreter (a visual interpreter) is used for visual analysis.

[0122] Figure 7(a) shows the importance of the characteristics of the process parameters input into the stacking model on the performance parameters (i.e., PCE). It can be seen that among the characteristics of the 11 process parameters studied, Ensc has the highest impact score on PCE, indicating that Ensc is the most important. Excess non-stoichiometry can help form a structure that is more conducive to charge transport, thereby improving the charge transport performance in the perovskite film. Therefore, in the preparation process of PSCs, the addition of Ensc can achieve a more efficient PCE. Secondly, perovskite thickness (Pt) and thermal annealing time (Ta-ti) are also important process parameters that affect PCE. The study by B. Zhang et al. showed that when the thickness of the perovskite CH3NH3PbI3 film is around 300nm, the photovoltaic efficiency is significantly improved, which is attributed to the improved carrier transport performance brought about by the preferred crystal structure. When the film thickness deviates from 300nm, the additional charge recombination and reduced mobility cause the photovoltaic efficiency of the perovskite solar cell to decrease again. The length of the thermal annealing time will affect the crystal growth and crystallinity of the perovskite film, which in turn affects the charge transport and collection efficiency, thereby affecting the battery performance. In addition to the above important process parameters, Figure 7 (a) also shows that the solvent mixing ratio (Smr), perovskite deposition solvent (Pds) and perovskite additive compound (Pac) are also crucial to PCE. In addition, thermal annealing can promote the crystal growth and crystallinity of perovskite films, improve the mobility and diffusion length of carriers, and reduce the effects of defects and surface states on charge transport, thereby improving PCE. However, too high or too low annealing temperature may lead to unstable crystal structure or the formation of unfavorable defects, affecting charge transport and light absorption efficiency, thereby reducing battery performance. Finally, although the impact of Pc-X and Bdp on PCE is not as high as the previous process parameters, they also have a certain impact.

[0123] In addition, in order to analyze the contribution of different process parameter characteristics to the prediction results, the SHAPvalue based on the stacking model is as follows: Figure 7 (b) is shown. The biggest advantage of SHAPvalue is that it can reflect the influence of each feature in each sample, and also show the positive and negative influence. Each row (i.e., the vertical axis) represents a feature, the horizontal axis is the SHAP value, and a point represents a sample. The darker the color, the larger the feature value. Figure 7In (b), it can be observed that the excess non-stoichiometric component (Ensc) is negatively correlated when the sequence code is small and positively correlated when the sequence code is large. Therefore, Ensc with large sequence code should be focused on. When the perovskite thickness increases, the SHAP value is positive, indicating a positive effect on PCE, which reflects that a moderate increase in the thickness of the perovskite film can improve the battery performance. The thickness of the perovskite film directly affects the light absorption capacity, photoelectric conversion efficiency and carrier transport efficiency of PSCs, and thus affects the performance of the entire solar cell. On the other hand, when the annealing time is too long, the SHAP value is negatively correlated, while when the time is relatively short, the SHAP value is positively correlated. The study by Harry MCronin et al. shows that too long annealing time may lead to the degradation of perovskite materials, the generation of new defects or the increase of existing defects. Therefore, while promoting grain growth and interface improvement, it is necessary to avoid too long annealing time. Adjusting the annealing time to achieve the optimal state is a key step in improving battery efficiency. These results are basically consistent with the conclusions of the Pearson correlation analysis, further explaining the rationality of the stacking model.

[0124] In one embodiment, step S52 may be analyzed using a violin plot.

[0125] Figure 8 (a)-(d) show the violin plots of the impact of the characteristics of a single process parameter on the performance parameter (i.e., PCE). Each violin represents the data distribution of an input feature at different output values. Its width reflects the density of the data points, while the shape of the violin reflects the distribution characteristics of the data. By analyzing the position, shape, and width of different violins, the degree and trend of the impact of the input feature on the output can be intuitively reflected. Figure 8 (a) It can be seen that among several different excess non-stoichiometric components (Ensc), the effect of the additive (MAI) on PCE can reach the maximum value. However, the PCE value corresponding to the fourth additive (PbI2) in the sequence coding is large and very stable. Therefore, adding PbI2 as a non-stoichiometric component is the most reliable to improve PCE. Figure 8 (b) A violin plot showing the effect of perovskite thickness (Pt) on PCE in the dataset. The perovskite thickness distribution ranges from 200 to 840, including 20 different thicknesses. As can be seen from the figure, the PCE is maximum when the thickness is 300-550nm, and the PCE value is relatively stable at 400nm. However, the PCE is relatively low below 200nm or above 600nm. Therefore, when preparing perovskite solar cells, the thickness should be controlled between 300-550nm to obtain high cell efficiency. Figure 8(c) Violin plot showing the effect of the perovskite deposition solvent mixing ratio (Smr) on PCE in the dataset. The perovskite deposition solvent mixing ratio is mainly between 0.4285 and 9, covering 8 different mixing ratios. When the mixing ratio is 4, the PCE value is the largest and tends to be stable. The thermal annealing time (Ta-ti) contained in the dataset is distributed between 5-150 minutes, including 14 different annealing times. Figure 8 (d) Violin plot showing the effect of thermal annealing time on the output PCE. It can be seen that the PCE is large and relatively stable when the annealing time is 10 minutes and 60 minutes. If the time is increased to more than 120 minutes, the PCE generally decreases. Different annealing temperatures and perovskite compositions have different requirements for the optimal annealing time. For CsFAMA perovskite film, heating at 100°C for 45 minutes or 60 minutes is the most ideal, and the PCE can reach 19.0%, while at 100 minutes, the PCE is 18.4%. Using non-peripherally substituted octapentylphthalocyanine (C5PcH2) as a free doping HTM layer, the best power conversion efficiency of the cell was obtained by annealing at 130°C for 10 minutes

[65] . Using non-peripherally substituted octapentylphthalocyanine (C5PcH2) as a free doping HTM layer, the best PCE of the cell was obtained by annealing at 130°C for 10 minutes.

[0126] In one embodiment, there is a synergistic effect between the characteristics of each process parameter and the performance parameter.

[0127] Specifically, we selected a group of batteries with excess PbI2 as a non-stoichiometric component to quantify the quantitative contribution of each input feature to PCE. The waterfall diagram is shown in Fig. 9 As shown. According to the prediction results of the model, those that have a positive contribution to PCE are represented by red, and those that do not are represented by blue. After analysis, it was found that the excess non-stoichiometric component (Ensc) and perovskite thickness (Pt) contribute the most to PCE, which are 1.65% and 1.31% respectively. Therefore, in the process of optimizing the preparation of PSCs, adjusting these two process parameters may have the most significant effect on PCE. Secondly, the contribution values ​​of perovskite deposition solvent (Pds) and thermal annealing time (Ta-ti) are 0.43% and -0.39% respectively, and the contribution trend is consistent with Figure 7 These results partially explain the synergistic effect of each input feature on the PCE of perovskite solar cells, which means that the various process parameters interact with each other in the performance optimization of PSCs. They do not affect PCE independently, but jointly affect the performance parameters of PSCs through a synergistic effect.

[0128] In one embodiment, in order to verify the process parameters optimized in this embodiment, three PSCs with different concentrations of Ensc were prepared, with 5% excess PbI2 in the first group, 10% excess PbI2 in the second group, and no excess PbI2 in the third group. The three groups used the same solvent, solvent additive, solvent mixing ratio, perovskite thickness, thermal annealing time and temperature and other conditions. The following table summarizes the performance parameters of the cells under different excess conditions.

[0129]

[0130] Table 2 Summary of performance parameters of batteries under different overcharge conditions

[0131] Fig.10 The JV characteristic curves of three groups of batteries at 100mW cm-2 and 1.5GAM are given. Obviously, the battery with 10% excess PbI2 obtains relatively higher performance parameters than the battery with 5% excess PbI2, which shows that adding excess non-stoichiometric PbI2 can effectively improve PCE. The experimental results further verify the feasibility of the established stacking model.

[0132] In this embodiment, the model prediction ability is improved by introducing a characteristic parameter method, and a variety of basic models are used to explore the optimization of the process parameters of perovskite solar cells and their impact on the performance parameters of the battery. Then, based on the decision tree and random forest models, the stacking model is concluded to be the best overall, with an R2 of 0.836, RMSE of 1.309, MAE of 0.890, and MAPE of 0.061 on the training sample, and an R2 of 0.806, RMSE of 1.432, MAE of 1.144, and MAPE of 0.083 on the test sample. The important features of the process parameters affecting the performance parameters are further analyzed based on the stacking model, and it is concluded that among the features of the 11 process parameters studied, adding excessive non-stoichiometric amounts is most important for PCE, followed by perovskite thickness. Then, the impact of the features of a single process parameter on the output is analyzed, and it is found that adding PbI2 as an excessive non-stoichiometric component is the most reliable for improving PCE. In addition, the numerical value of the performance parameters is the largest and relatively stable when the perovskite thickness is in the range of 300-550nm. Finally, it was experimentally verified that adding an excess of non-stoichiometric components (PbI2) can effectively improve performance parameters. These findings demonstrate the great potential of perovskite battery performance prediction models and provide important guidance for the optimization of process parameters and prediction of performance parameters of perovskite batteries.

[0133] The present disclosure provides a training system for a perovskite battery performance prediction model. Fig.11 , the training system includes:

[0134] A first acquisition module 111 is used to acquire a plurality of groups of training samples of perovskite cells, wherein the training samples include preset process parameters, preset performance parameters and feature estimation parameters, the preset process parameters include at least one feature, and the feature estimation parameters are obtained by performing attribute estimation on the features in the preset process parameters;

[0135] A training module 112, configured to perform training in at least one basic model according to the plurality of training samples, wherein the preset process parameters and the corresponding feature estimation parameters are used as inputs of the basic model, and the preset performance parameters are used as outputs of the basic model;

[0136] A first determination module 113, configured to determine a training result of at least one basic model;

[0137] The stacking module 114 is used to select at least two basic models whose training results meet preset conditions from at least one basic model to construct a stacking model to obtain a perovskite battery performance prediction model.

[0138] In one embodiment, the preset process parameters include: one or more of: A-site cations, X-site anions, perovskite thickness, perovskite deposition solvent, solvent mixing ratio, thermal annealing temperature, and thermal annealing time.

[0139] In one embodiment, the base model includes one or more of: decision tree, random forest, CatBoost, LassoLarsCV, histogram gradient boosting, extreme gradient boosting, K nearest neighbor, ridge regression and linear regression.

[0140] In one embodiment, before the training module 112, an optimization module is executed;

[0141] The optimization module is used to optimize the hyperparameters of at least one base model through ten-fold cross validation.

[0142] In one embodiment, the training system includes:

[0143] A selection module, used for selecting a preset proportion of training samples from multiple groups of training samples as test samples of at least one basic model;

[0144] The first determination module 113 is further used to test at least one basic model according to the test sample and calculate the training result of at least one basic model.

[0145] In one embodiment, the training results include: one or more of: coefficient of determination, root mean square error, mean absolute error, and mean absolute percentage error.

[0146] In one embodiment, the stacked model includes a base learning layer and a meta-learning layer connected to each other;

[0147] The stacking module 114 includes:

[0148] A first selection unit is used to select at least two basic models whose training results meet preset conditions from at least one basic model, and construct a basic learning layer of the stacked model according to preset weights;

[0149] The second selection unit is used to select at least one basic model from the selected at least two basic models to construct a meta-learning layer of the stacked model.

[0150] An exemplary embodiment of the present disclosure provides a prediction system for perovskite battery performance, see Fig.12 , the prediction system includes:

[0151] A second acquisition module 121 is used to acquire a perovskite battery performance prediction model, where the perovskite battery performance prediction model is trained according to a training system for a perovskite battery performance prediction model according to any one of the third aspects;

[0152] A second determination module 122 is used to analyze the process parameters and performance parameters of the perovskite battery performance prediction model, and determine the importance of each feature in the process parameters to the performance parameters based on the analysis results;

[0153] The value taking module 123 is used to determine the value range of each feature in the process parameter according to the importance.

[0154] The system and method embodiments in the disclosed embodiments are based on the same disclosed concept.

[0155] An exemplary embodiment of the present disclosure also provides an electronic device, see Fig.13 , the electronic device 1300 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 1310 (the processor 1310 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic battery FPGA), a memory 1330 for storing data, and one or more storage media 1320 (such as one or more mass storage devices) for storing application programs 1323 or data 1322. Among them, the memory 1330 and the storage medium 1320 can be short-term storage or permanent storage. The storage medium 1320 stores at least one instruction, at least one program, code set or instruction set, and at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the above-mentioned training method of the perovskite battery performance prediction model and / or the prediction method of the perovskite battery performance.

[0156] Furthermore, the CPU 1310 may be configured to communicate with the storage medium 1320 to execute a series of instruction operations in the storage medium 1320 on the electronic device 1300. The electronic device 1300 may also include one or more power supplies 1360, one or more wired or wireless network interfaces 130, one or more input and output interfaces 1340, and / or one or more operating systems 1321, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0157] The input / output interface 1340 may be used to receive or send data via a network. The specific example of the network may include a wireless network provided by a communication provider of the electronic device 1300. In one example, the input / output interface 1340 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the input / output interface 1340 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0158] It can be understood by those skilled in the art that Fig.13 The structure shown is only for illustration and does not limit the structure of the above electronic device. Fig.13 More or fewer components as shown, or with Fig.13 Different configurations are shown.

[0159] An embodiment of the present disclosure also provides a computer-readable storage medium, which can be set in a server to store at least one instruction, at least one program, code set or instruction set related to a training method for a perovskite battery performance prediction model and / or a method for predicting perovskite battery performance in an embodiment of the present disclosure. The at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the above-mentioned training method for a perovskite battery performance prediction model and / or a method for predicting perovskite battery performance.

[0160] Optionally, in this embodiment, the storage medium may be located in at least one of the multiple network servers of the computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0161] It is known from common technical knowledge that the present disclosure can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the embodiments disclosed above are only illustrative in all respects and are not exclusive. All changes within the scope of the present disclosure or within the scope equivalent to the present disclosure are included in the present disclosure.

[0162] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0163] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0164] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0165] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure rather than to limit them. Although the present disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific implementation methods of the present disclosure may still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present disclosure shall be included within the scope of protection of the claims of the present disclosure.

Claims

1. A training method for a perovskite battery performance prediction model, characterized in that: The training method comprises: Acquire multiple groups of training samples of perovskite cells, wherein the training samples include preset process parameters, preset performance parameters, and feature estimation parameters, the preset process parameters include at least one feature, and the feature estimation parameters are obtained by performing attribute estimation on the feature in the preset process parameters; Training is performed in at least one basic model according to the plurality of groups of training samples, wherein the preset process parameters and the corresponding feature estimation parameters are used as inputs of the basic model, and the preset performance parameters are used as outputs of the basic model; Determining a training result of the at least one base model; At least two of the basic models whose training results meet preset conditions are selected from the at least one basic model to construct a stacking model to obtain a perovskite battery performance prediction model.

2. The training method according to claim 1, characterized in that: The preset process parameters include: one or more of: A-site cations, X-site anions, perovskite thickness, perovskite deposition solvent, solvent mixing ratio, thermal annealing temperature, and thermal annealing time.

3. The training method according to claim 1, characterized in that: The basic models include: one or more of decision tree, random forest, CatBoost, LassoLarsCV, histogram gradient boosting, extreme gradient boosting, K nearest neighbor, ridge regression and linear regression.

4. The training method according to claim 1, characterized in that: Before training at least one basic model according to the plurality of groups of training samples, the method comprises: The hyperparameters of the at least one base model are optimized by ten-fold cross validation.

5. The training method according to claim 1, characterized in that: The training method comprises: Selecting a preset proportion of training samples from the multiple groups of training samples as test samples of the at least one basic model; Determining a training result of the at least one base model includes: The at least one basic model is tested according to the test sample, and a training result of the at least one basic model is calculated.

6. The training method according to claim 5, characterized in that: The training results include: one or more of: determination coefficient, root mean square error, mean absolute error, and mean absolute percentage error.

7. The training method according to claim 1, characterized in that: The stacked model includes a base learning layer and a meta-learning layer connected to each other; Selecting at least two of the basic models whose training results meet preset conditions from the at least one basic model to construct a stacking model, and obtaining a perovskite battery performance prediction model, including: Selecting at least two basic models whose training results meet preset conditions from the at least one basic model, and constructing a basic learning layer of the stacked model according to preset weights; At least one of the at least two selected basic models is selected to construct a meta-learning layer of the stacked model.

8. A method for predicting the performance of a perovskite battery, characterized in that: The prediction method comprises: Obtaining a perovskite battery performance prediction model, wherein the perovskite battery performance prediction model is trained according to the training method for the perovskite battery performance prediction model according to any one of claims 1 to 7; Analyzing the process parameters and the performance parameters, and determining the importance of each feature in the process parameters to the performance parameters based on the analysis results; The value range of each feature in the process parameter is determined according to the importance.

9. A training system for a perovskite battery performance prediction model, characterized in that: The training system comprises: A first acquisition module is used to acquire a plurality of groups of training samples of perovskite batteries, wherein the training samples include preset process parameters, preset performance parameters and feature estimation parameters, the preset process parameters include at least one feature, and the feature estimation parameters are obtained by performing attribute estimation on the feature in the preset process parameters; A training module, used for training at least one basic model according to the plurality of groups of training samples, wherein the preset process parameters and the corresponding feature estimation parameters are used as inputs of the basic model, and the preset performance parameters are used as outputs of the basic model; A first determining module, used to determine the training result of the at least one basic model; A stacking module is used to select at least two basic models whose training results meet preset conditions from the at least one basic model to construct a stacking model to obtain a perovskite battery performance prediction model.

10. A system for predicting perovskite battery performance, characterized in that: The prediction system comprises: A second acquisition module is used to acquire a perovskite battery performance prediction model, wherein the perovskite battery performance prediction model is trained according to the training system for the perovskite battery performance prediction model according to claim 9; A second determination module is used to analyze the process parameters and the performance parameters, and determine the importance of each feature in the process parameters to the performance parameters based on the analysis results; A value taking module is used to determine the value range of each feature in the process parameter according to the importance.

Images