Photocurrent prediction method and system for perovskite / crystalline silicon two-end laminated cell

Abstract

The invention discloses a light current prediction method and system for a perovskite / crystalline silicon two-end laminated cell, relates to the technical field of cell light current calculation, and aims to solve the problem that the efficiency of the perovskite / crystalline silicon two-end laminated cell is too low when a three-dimensional optical model is simulated. The method comprises the following steps: constructing a first proxy model by taking a perovskite material, a band gap and a wavelength of the laminated cell as inputs and a complex refractive index as an output; obtaining a first result according to the three-dimensional wave optical model, and inputting the first result into a database as sample data; preprocessing the sample data, and combining a three-dimensional wave optical model and an artificial neural network with Bayesian optimization and Bayesian positive side for training to generate a second agent model; inputting the full wavelength into a second agent model to output a predicted wavelength, and then calculating to obtain the total light current of the laminated cell; according to the method, the efficient multi-agent model is established in a deep learning mode, and multi-scene application can be carried out without a high-configuration computer.

Description

Technical Field

[0001] The present invention relates to the technical field of battery photocurrent calculation, and in particular to a photocurrent prediction method and system for a perovskite / crystalline silicon two-terminal stacked battery. Background Art

[0002] Perovskite / crystalline silicon tandem cells integrate a perovskite thin film onto a crystalline silicon substrate, forming a tandem structure. This utilizes current matching and voltage stacking to improve efficiency. Currently, these cells suffer from slow efficiency when simulated using three-dimensional optical models. Specifically, these issues include: 1. The complex refractive index of the perovskite film cannot accurately and rapidly change with its band gap; and 2. The need for full-wavelength scanning to calculate the total photocurrent.

[0003] The performance of a perovskite / crystalline silicon two-terminal tandem cell depends on current matching between the subcells, which is achieved by adjusting the material, bandgap, and thickness of the perovskite layer. Bandgap adjustment can be achieved by varying the perovskite's chemical composition, but this also alters its complex refractive index (both real and imaginary parts), thereby affecting its absorbance. Because there is no direct mathematical relationship between bandgap, material, and complex refractive index, and the complex refractive index varies nonlinearly with wavelength, optimization simulations require a separate full-wavelength complex refractive index database for each bandgap, resulting in inefficient computations.

[0004] The photocurrent of a solar cell is derived by integrating the amount of light absorbed at each wavelength, so simulations require a full wavelength scan. For example, the absorption range of a perovskite / crystalline silicon tandem cell is typically 300–1200 nm. If the calculation is performed with a 1 nm step size, the 3D wave optics model would need to be run 901 times, which is computationally very expensive.

[0005] For example, the Chinese patent with publication number CN107507928A only discloses a method for regulating the total current of a battery and cannot solve the above-mentioned technical problems. Summary of the Invention

[0006] The present invention solves the problem of low efficiency of perovskite / crystalline silicon two-terminal tandem cells during three-dimensional optical model simulation, and proposes a photocurrent prediction method and system for perovskite / crystalline silicon two-terminal tandem cells. Through deep learning, an efficient multi-agent model is established, which can be applied in multiple scenarios without the need for high-configuration computers.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting the photocurrent of a perovskite / crystalline silicon two-terminal tandem cell, comprising the following steps: S1, constructing a first proxy model using the perovskite material, band gap, and wavelength of the tandem cell as input and the complex refractive index as output; S2, obtaining a first result based on the three-dimensional wave optics model, and inputting the first result into the database as sample data; the first result includes the input and output of the three-dimensional wave optics model; S3, pre-processing the sample data, combining the three-dimensional wave optics model and artificial neural network training with Bayesian optimization and Bayesian lateralization to generate the second proxy model; S4, inputting the full wavelength into the second proxy model to output the predicted wavelength, and then calculating the total photocurrent of the stacked cell.

[0008] In this technical solution, a first proxy model is constructed using the relationship between perovskite materials, band gaps, and complex refractive indexes. The first proxy model is then introduced into a three-dimensional wave optics model for sample training and verification. An artificial neural network with Bayesian optimization and Bayesian normalization is combined to finally obtain a second proxy model. Finally, the total photocurrent of the stacked cell is calculated based on the structure of the second proxy model.

[0009] The present invention is further configured as follows: the input of the three-dimensional wave optics model is the band gap, wavelength, film thickness and texture parameters of the perovskite material; the output of the three-dimensional wave optics model is the surface reflectivity at a single wavelength, and the quantum efficiency of the top perovskite and bottom crystalline silicon sub-cells.

[0010] In this technical solution, a large amount of virtual data can be provided through the three-dimensional wave optics model.

[0011] The present invention is further configured as follows: the construction process of the first proxy model is: using perovskite material, band gap, wavelength and complex refractive index as first sample data, combined with a Gaussian regression model with Bayesian optimization for training, and then testing based on the first sample data, and finally generating a first proxy model with input as material, band gap and wavelength and output as complex refractive index.

[0012] In this technical solution, a first proxy model is constructed to output the complex refractive index, which is ultimately used in the subsequent training and optimization process of the second proxy model.

[0013] The present invention is further configured as follows: before generating the second proxy model in step S3, the test data set reserved in the sample data pre-processing process is used to detect the model performance and whether there is an overfitting problem; if the model fails to accurately reflect the relationship between the input and output in the test data, Bayesian optimization needs to be re-performed; conversely, if the model can accurately predict the test data, a verified second proxy model will be obtained.

[0014] In this technical solution, the accuracy of the second agent model is further verified through the above-mentioned testing process.

[0015] The present invention is further configured as follows: the first proxy model is trained and optimized based on a Gaussian regression model with Bayesian optimization, and after the optimization is completed, the regression model parameters and hyperparameters of the perovskite material are recorded.

[0016] The present invention is further configured as follows: in step S3, the parameters to be optimized by Bayesian optimization include the number of hidden layers of the artificial neural network; the hidden layer activation function; the hidden layer size; the initial learning rate; the minimum gradient descent and the momentum of the gradient descent.

[0017] The present invention is further configured as follows: the total photocurrent of the stacked cell obtained by calculation is specifically: The photocurrents of the perovskite and crystalline silicon cells are calculated based on the vector integral corresponding to the predicted wavelength. The photocurrent of the stacked cell is the minimum photocurrent of the perovskite and crystalline silicon cells.

[0018] In this technical solution, the corresponding photocurrent is calculated based on the second proxy model.

[0019] The present invention is further configured such that after the first result is input into the database, the thickness of the perovskite layer and the velvet height in the sample data are calibrated, specifically by minimizing the error between the quantum efficiency obtained by simulation and the actual quantum efficiency, and fine-tuning the perovskite thickness and the velvet height.

[0020] In this technical solution, the input perovskite layer thickness and velvet height are obtained through a scanning electron microscope. Under the scanning electron microscope, the actual film thickness at different positions is uneven, resulting in errors, so calibration is used to further ensure the accuracy of the data.

[0021] The present invention is further configured such that: the pre-processing of the sample data includes dividing the data set of the sample data into training, verification and test data sets for training and optimization of the second proxy model.

[0022] A photocurrent prediction system for a perovskite / crystalline silicon two-terminal tandem cell, applicable to the above-mentioned photocurrent prediction method for a perovskite / crystalline silicon two-terminal tandem cell, comprises: A first proxy model construction module is configured to construct a first proxy model using the perovskite material, band gap, and wavelength of the tandem cell as input and the complex refractive index as output; A first result generating module, combining with a three-dimensional wave optics model to output a first result; a second proxy model construction module, generating a second proxy model based on the first result and the three-dimensional wave optics model training; The total photocurrent calculation module inputs the full wavelength into the second proxy model to output the predicted wavelength, and obtains the total photocurrent of the stacked cell through integral calculation.

[0023] In this technical solution, through the collaborative operation of the first proxy model construction module, the first result generation module, the second proxy model construction module and the total photocurrent calculation module, the photocurrent of the perovskite / crystalline silicon two-terminal stacked battery can be calculated and predicted, ensuring the high-efficiency operation of the perovskite / crystalline silicon two-terminal stacked battery during three-dimensional optical model simulation.

[0024] The present invention can bring the following beneficial effects: The present application involves a method for predicting the photocurrent of a perovskite / crystalline silicon two-terminal stacked cell, which constructs a first proxy model and a second proxy model through deep learning to ensure the high-efficiency operation of the perovskite / crystalline silicon two-terminal stacked cell during three-dimensional optical model simulation, and can be applied in multiple scenarios without the need for a high-configuration computer. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of constructing a first proxy model for a photocurrent prediction method for a perovskite / crystalline silicon two-terminal tandem battery in the present application.

[0026] Figure 2 This is a flow chart of the process of generating the first result of a photocurrent prediction method for a perovskite / crystalline silicon two-terminal tandem battery in the present application.

[0027] Figure 3 This is a flow chart of a photocurrent prediction method for a perovskite / crystalline silicon two-terminal tandem cell of the present application, which constructs a second agent model and calculates the total photocurrent. DETAILED DESCRIPTION

[0028] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] Perovskite / crystalline silicon two-terminal tandem cells integrate a perovskite thin film onto a crystalline silicon substrate, forming a series structure that utilizes current matching and voltage stacking to improve efficiency. Prior art perovskite / crystalline silicon two-terminal tandem cells suffer from slow efficiency when simulated using three-dimensional optical models. Analysis of this issue reveals three key issues: 1. The complex refractive index of the perovskite film cannot accurately and rapidly change with its band gap; 2. The need for full-wavelength scanning to calculate the total photocurrent; and 3. The reliance on high-performance computers hinders convenient use and digital twin applications.

[0030] The complex refractive index of perovskite films cannot accurately and quickly change with their band gap. Specifically, the performance of perovskite / crystalline silicon two-terminal tandem cells depends on the current matching of the sub-cells, which must be achieved by adjusting the material, band gap, and thickness of the perovskite layer. Band gap adjustment can be achieved by changing the chemical ratio of the perovskite, but this also changes its complex refractive index (including both real and imaginary parts), thereby affecting the absorbance. Because there is no direct mathematical relationship between band gap, material, and complex refractive index, and the complex refractive index varies nonlinearly with wavelength, optimization simulations require a full-wavelength complex refractive index database for each band gap, resulting in low computational efficiency.

[0031] Calculating the total photocurrent requires a full wavelength scan. Specifically, the photocurrent of a solar cell is derived by integrating the absorption at each wavelength, so a full wavelength scan is required during simulation. For example, the absorption range of a perovskite / crystalline silicon tandem cell is typically 300–1200 nm. If calculated with a 1 nm step size, the 3D wave optics model would need to be run 901 times, resulting in an extremely high computational cost.

[0032] Relying on high-performance computers hinders ease of use and digital twin applications. Specifically, to accurately simulate the co-texture morphology (micrometer-scale pyramid structures on the silicon surface) of perovskite / crystalline silicon tandem cells, a three-dimensional wave optics model and finite element method are required. Because silicon thicknesses can reach hundreds of microns, and wave optics requires a grid size smaller than one-fifth of a wavelength (60-240nm), the resulting grid size is extremely large, necessitating the use of high-performance servers for computation.

[0033] Example 1 In view of the shortcomings of the existing technology, this embodiment proposes a photocurrent prediction method for a perovskite / crystalline silicon two-terminal tandem cell, referring to Figure 1 、 Figure 2 as well as Figure 3 , which includes the following steps.

[0034] Step S1, constructing a first proxy model, the first proxy model takes the perovskite material, band gap and wavelength of the perovskite / crystalline silicon two-terminal stacked cell as input, and the complex refractive index as output.

[0035] refer to Figure 1 The above steps of constructing the first proxy model are more detailed. The perovskite material, band gap, wavelength and complex refractive index are used as the first sample data, and the Gaussian regression model with Bayesian optimization is combined for training. The first sample data is then used for testing to finally generate the first proxy model. The input of the first proxy model is the material, band gap and wavelength, and the output is the complex refractive index.

[0036] In this technical solution, a first proxy model is constructed to output the complex refractive index, which is ultimately used in the subsequent training and optimization process of the second proxy model.

[0037] More specifically, the first proxy model is trained and optimized based on a Gaussian regression model with Bayesian optimization. After the optimization is completed, the regression model parameters and corresponding hyperparameters of the perovskite material are recorded.

[0038] The above-mentioned step S1 mainly includes a data collection process and a machine learning process.

[0039] In the data collection process, during the preparation phase, perovskite films with different band gaps are formed by varying the chemical composition of the perovskite. Spectroscopic ellipsometers are then used to confirm the film band gap and measure the material's complex refractive index at all wavelengths. This ultimately creates an initial database consisting of perovskite materials, band gaps, wavelengths, and complex refractive indices. For a single material, the data sets with different band gaps, wavelengths, and complex refractive indices are divided into training, validation, and test datasets. In this example, 100,000 sets of samples (one sample for each band gap, wavelength, and complex refractive index) are collected for training the proxy model.

[0040] The machine learning process mainly includes the pre-processing of sample data, the training process of building a Gaussian regression model with Bayesian optimization, and the model testing process.

[0041] Regarding the pre-processing process of the sample data, in more detail, in this embodiment, the band gap range is generally 1 to 2eV, and the wide band gap perovskite is 1.6 to 1.8eV; the wavelength range is 300 to 1200nm; the real part of the complex refractive index ranges from 1.5 to 3, and the imaginary part ranges from 0 to 1.5. The span of the above parameter ranges is not large, and there is no situation where the upper and lower limits differ by more than 10 times, so there is no need to force normalization. In order to train an accurate model more quickly and efficiently, a set of experimental data with the smallest band gap is used as a benchmark. This set of data will include the smallest band gap, full wavelength and corresponding complex refractive index to help machine learning quickly establish a data framework.

[0042] The training process for constructing a Gaussian regression model with Bayesian optimization involves selecting hyperparameters for the basis functions, their coefficients, the noise variance, the kernel function, and their corresponding hyperparameters. After the optimization process is complete, the regression model and its final hyperparameters for the perovskite material are recorded to prepare for a future multi-material database. For example, a model is trained for each of MAPbI, FAPbI, and CsFAPbIBr materials, matching the materials to the model hyperparameters. A human classification algorithm is then added to the proxy regression model.

[0043] Specifically, during the model testing process, the reserved test data set from the data collection process is used to examine the model's performance and determine if there are any overfitting issues. If the model fails to accurately reflect the relationship between input and output in the test data (the predicted results differ significantly from the actual responses), retraining with more high-quality data or using a more complex algorithm is necessary. In this example, the current regression model is sufficiently accurate. Conversely, if the model accurately predicts the test data, we will obtain a validated proxy model.

[0044] Through the above steps, a first proxy model can be finally obtained, which can output the film complex refractive index of the perovskite material at each wavelength instantly and accurately by inputting the band gap and the wavelength considered.

[0045] After completing the process of step S1, refer to Figure 2 , proceed to step S2, combine the three-dimensional wave optics model to obtain a first result, input the first result into the database as sample data for the subsequent second proxy model; wherein the first result includes the input and output of the above-mentioned three-dimensional wave optics model.

[0046] The input of the three-dimensional wave optics model is the band gap, wavelength, film thickness and texture parameters of the perovskite material; the output of the three-dimensional wave optics model is the surface reflectivity at a single wavelength, and the quantum efficiency of the top perovskite and bottom crystalline silicon sub-cells.

[0047] In this technical solution, a large amount of virtual data can be provided through the three-dimensional wave optics model.

[0048] The 3D wave optics model utilizes wave optics and a finite element solver to simulate the photoelectric field distribution of steady-state light contacting the perovskite top cell, propagating across the uneven suede surface, and finally passing through the crystalline silicon bottom cell within the unit cell structure of the tandem cell. Combined with the full-wavelength incident spectrum, the quantum efficiency and photocurrent are calculated, assisting in the current matching and structural design of the perovskite / crystalline silicon tandem cell. By combining the first proxy model with the 3D wave optics model of the perovskite / crystalline silicon tandem cell, a multi-proxy model based on machine learning can be trained to accurately reflect the photoelectric field structure of the 3D wave optics, closely resembling the actual perovskite material.

[0049] After step S2, step S3 is performed to pre-process the sample data obtained in step S2, and then train and optimize it using a three-dimensional wave optics model and an artificial neural network with Bayesian optimization and Bayesian lateralization to finally generate a second proxy model.

[0050] Among them, the parameters optimized by Bayesian optimization include the number of hidden layers in the artificial neural network; the hidden layer activation function; the hidden layer size; the initial learning rate; the minimum gradient descent and the momentum of the gradient descent.

[0051] Before generating the second proxy model, the test data is used to detect the model performance and whether there is an overfitting problem. If the model cannot accurately reflect the relationship between the input and output in the test data (the error between the predicted result and the actual response is too large), it needs to be re-optimized. If the model can accurately predict the test data, the model does not need to be re-optimized, and the second proxy model is obtained.

[0052] In addition, after the first result is input into the database, the perovskite layer thickness and velvet height in the sample data are calibrated. The calibration process is to fine-tune the perovskite thickness and velvet height by minimizing the error between the quantum efficiency obtained by simulation and the actual quantum efficiency.

[0053] In this technical solution, the input perovskite layer thickness and velvet height are obtained through a scanning electron microscope. Under the scanning electron microscope, the actual film thickness at different positions is uneven, resulting in errors. Therefore, the above-mentioned calibration is used to further ensure the accuracy of the data.

[0054] The construction of the second agent model mainly includes data collection and machine learning processes.

[0055] For the data collection process, refer to Figure 3 Specifically, a large amount of virtual data is provided through a three-dimensional wave optics model. The input includes wavelength, perovskite material and its band gap, film thickness, and texture geometry parameters; the output is the surface reflectivity at a single wavelength, and the quantum efficiency of the top perovskite and bottom crystalline silicon subcells.

[0056] It should be noted that the following factors were incorporated into the data collection process of the second agent model: Figure 1 The first proxy model shown quickly outputs the material's complex refractive index by wavelength and bandgap.

[0057] In addition, the data collection process includes virtual data supplementation, which requires calibration with experimental data (measured by a quantum efficiency meter). The parameters that need to be calibrated are the perovskite layer thickness and the velvet (pyramid) height. These two parameters are selected because the actual film thickness at different locations under a scanning electron microscope (SEM) is uneven, resulting in slight errors. The calibration method is to minimize the error between the quantum efficiency obtained by simulation and the actual quantum efficiency, and then fine-tune the perovskite thickness and velvet height.

[0058] In this embodiment, the number of calibrations is determined by the actual training data. For wavelengths between 300 and 1200 nm, using a massive training data set of 100,000 samples, at least 10 textured morphologies should be selected, with calibration performed at most once for every 100 nm of perovskite thickness and every 100 nm of wavelength, resulting in approximately 1,000 experimental calibrations. (Since the quantum efficiency meter can test all wavelengths at once, the actual number of experiments is approximately 100.) After calibration, virtual data filling is completed through simulation and interpolation.

[0059] The machine learning for constructing the second agent model includes pre-processing of sample data, establishing an artificial neural network training model with Bayesian optimization and Bayesian lateralization, and model testing process.

[0060] In the second proxy model construction step, the sample data pre-processing process specifically involves dividing the collected big data into training, validation, and test datasets for training and optimizing the artificial neural network-based proxy model. Due to the large number of wavelengths considered (300-1200nm, 1nm accuracy), a large training dataset of 100,000 and validation and test datasets of 10,000 each were used. The specific reference ranges for the parameters considered in data pre-processing were: wavelength 300-1200nm, perovskite thickness 400-1500nm, maximum perovskite band gap range (depending on the material) 1-2eV, pyramid base width 2-5um, pyramid height 1-3um, and pyramid top-to-base width ratio 0.5-0.9. The upper and lower limits do not differ by more than a factor of 10, so there is no need for logarithmic or normalization. Furthermore, after converting the wavelength and perovskite thickness to μm, all parameter values ​​fall within the range of 0-5, eliminating the need for additional data pre-processing. The output parameters are the total surface reflectivity, the quantum efficiency of the perovskite layer, and the quantum efficiency of the crystalline silicon layer, all ranging from 0 to 1.

[0061] In the step of constructing the second proxy model, establishing an artificial neural network training model with Bayesian optimization and Bayesian lateralization specifically includes: establishing an artificial neural network training program with Bayesian optimization and Bayesian lateralization. The parameters to be optimized by Bayesian optimization include the number of hidden layers of the artificial neural network (range 3-5 layers, 1-2 layers are not considered because they are not enough to complete highly nonlinear regression problems, and more than 5 layers are not considered because the training time is too long); hidden layer active function - hyperbolic tangent sigmoid, radial basis function, log-sigmoid and linear function; hidden layer size (range 2-100 per layer); the maximum number of iterations is constant at 1000 (this parameter is not optimized); initial learning rate (range 0.01 to 1); minimum gradient descent (one of the termination conditions for a single neural network training, range arrive ) and the momentum of gradient descent (range 0.5 to 0.98). In addition to the minimum gradient descent, the termination conditions for a single neural network training also include early stopping (a way to avoid overfitting when the error of the training data decreases but the error of the validation data increases). The termination conditions of the overall Bayesian optimization are set to the maximum number of iterations (1000 times), the maximum time limit (usually 1 week is enough) and the minimum gradient descent (usually set to The artificial neural network uses Bayesian regularization with back-propagation, which can effectively prevent overfitting and learn the covariance of multiple output parameters.

[0062] During the second-proxy model construction step, the model testing process involves using a test dataset to examine model performance and overfitting. If the model fails to accurately reflect the relationship between inputs and outputs in the test data (the error between the predicted results and the actual responses is too large), the Bayesian optimization process must be repeated. Conversely, if the model accurately predicts the test data, the second-proxy model is validated.

[0063] In step S4, the full wavelength is input into the second proxy model to output a predicted wavelength, and the total photocurrent of the tandem cell is then calculated. Specifically, the photocurrents of the perovskite and crystalline silicon cells are calculated based on the vector integral corresponding to the predicted wavelength. The photocurrent of the tandem cell is the minimum of the photocurrents of the perovskite and crystalline silicon cells.

[0064] This technical solution ultimately resulted in a neural network-based multi-proxy model (a second proxy model trained using a three-dimensional wave optics model was added to the first proxy model). Its inputs were six parameters (wavelength, perovskite film thickness, perovskite bandgap, and three texture geometry parameters), and its outputs were the incident surface reflectivity, the perovskite layer quantum efficiency, and the crystalline silicon layer quantum efficiency. Furthermore, this model was capable of full-wavelength vectorized calculations, effectively resolving the problem of previous models requiring full-wavelength scanning. Specifically, a vector with a wavelength of 300-1200nm and single values ​​for the remaining inputs resulted in a vector output corresponding to the wavelength range of 300-1200nm. The photocurrents of the perovskite and crystalline silicon cells were then calculated through integration, with the photocurrent of the tandem cell being the lowest of the two.

[0065] Based on the above-mentioned photocurrent prediction method for a perovskite / crystalline silicon two-terminal tandem cell, this embodiment also proposes a photocurrent prediction system for a perovskite / crystalline silicon two-terminal tandem cell, which mainly includes a first agent model construction module, a first result generation module, a second agent model construction module and a total photocurrent calculation module, wherein the first agent model construction module is connected to the first result generation module, the first result generation module is connected to the second agent model construction module, and the second agent model construction module is connected to the total photocurrent calculation module.

[0066] For the first proxy model construction module, it can use the perovskite material, band gap and wavelength of the tandem cell as input and the complex refractive index as output to construct the first proxy model. The specific construction process can be referred to Figure 1 And the process of step S1 in the above prediction method.

[0067] The first result generating module can output the first result according to the three-dimensional wave optics model. For the specific generating process of the first result, reference may be made to the process of step S2 in the above prediction method.

[0068] As for the second proxy model construction module, it is mainly based on the above-mentioned first result and the three-dimensional wave optics model, and is trained and optimized to finally obtain the second proxy model.

[0069] For the total photocurrent calculation module, it can input the full wavelength into the second model that has been trained and optimized, and finally output the predicted wavelength. It can also obtain the photocurrents of perovskite and crystalline silicon cells through integral operations, and the photocurrent of the stacked cell is the lowest value of the two.

[0070] In this technical solution, through the collaborative operation of the first proxy model construction module, the first result generation module, the second proxy model construction module and the total photocurrent calculation module, the photocurrent of the perovskite / crystalline silicon two-terminal stacked battery can be calculated and predicted, ensuring the high-efficiency operation of the perovskite / crystalline silicon two-terminal stacked battery during three-dimensional optical model simulation.

[0071] 1. This embodiment solves the problem that the complex refractive index of the perovskite film cannot change accurately and quickly with its band gap: (1) Since the first proxy model based on Gaussian regression is trained, the complex refractive index of the film and the band gap of the material are mapped, that is, the complex refractive index of the full wavelength can be output by inputting the perovskite material, band gap and wavelength; (2) The time consumption of this embodiment is instant. When performing optoelectronic simulation, the total calculation time required will be greatly reduced, which is convenient for big data collection and analysis; (3) After confirming the perovskite material, the first proxy model can complete the mapping of the band gap parameter to the full wavelength (300-1200nm) complex refractive index, a total of 901*2=1802 parameters, which greatly reduces the memory and disk usage during calculation.

[0072] 2. This embodiment solves the problem of needing to perform full-wavelength scanning when calculating the total photocurrent: (1) The second proxy model finally obtained can perform full-wavelength vectorized calculation, which solves the problem of needing full-wavelength scanning and simplifies 901 simulations into one vectorized calculation simulation; (2) Because the amount of training data collected is large and the parameter range is wide, the proxy model of this embodiment can cope with velvets of different morphologies (from pointed triangles to wide pyramids) and different perovskite band gaps and thicknesses, which is conducive to global optimization.

[0073] 3. This embodiment solves the problem of relying on high-configuration computers, which is not conducive to convenient use and digital twin applications: (1) The final multi-agent model uses only six input parameters (wavelength, band gap, film thickness and three geometric parameters that determine the texture structure) and three output parameters (reflectivity and quantum efficiency of two sub-cells), which is simple and convenient. In addition, only six inputs can cope with the adjustment of the perovskite material ratio, the perovskite film preparation process and the texture treatment process of the crystalline silicon surface. Real-time calculation complements actual laboratory preparation and can be applied to digital twins; (2) Compared with traditional simulation methods, there is no need for equations and physical field construction and meshing steps, so there is no hardware requirement for computer memory, disk and CPU, and it does not rely on high-configuration computers; (3) Because it is a machine learning-based agent model, it is convenient for retraining with new data and has room for improvement to keep pace with the times.

[0074] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention and are intended to be encompassed by the claims of the present invention.

Claims

1. A method for predicting the photocurrent of a perovskite / crystalline silicon two-terminal tandem cell, characterized in that: The following steps are involved: S1, constructing a first proxy model using the perovskite material, band gap, and wavelength of the tandem cell as input and the complex refractive index as output; S2, obtaining a first result based on the three-dimensional wave optics model, and inputting the first result into the database as sample data; the first result includes the input and output of the three-dimensional wave optics model; S3, pre-processing the sample data, combining the three-dimensional wave optics model and artificial neural network training with Bayesian optimization and Bayesian lateralization to generate the second proxy model; S4, inputting the full wavelength into the second proxy model to output the predicted wavelength, and then calculating the total photocurrent of the stacked cell.

2. The method for predicting photocurrent of a perovskite / crystalline silicon two-terminal tandem cell according to claim 1, characterized in that: The input of the three-dimensional wave optics model is the band gap, wavelength, film thickness and texture parameters of the perovskite material; the output of the three-dimensional wave optics model is the surface reflectivity at a single wavelength, and the quantum efficiency of the top perovskite and bottom crystalline silicon sub-cells.

3. The method for predicting photocurrent of a perovskite / crystalline silicon two-terminal tandem cell according to claim 1 or 2, characterized in that: The construction process of the first proxy model is as follows: using perovskite material, band gap, wavelength and complex refractive index as the first sample data, combined with a Gaussian regression model with Bayesian optimization for training, and then testing based on the first sample data, and finally generating a first proxy model with input of material, band gap and wavelength and output of complex refractive index.

4. The method for predicting photocurrent of a perovskite / crystalline silicon two-terminal tandem cell according to claim 3, characterized in that: Before generating the second proxy model in step S3, the test data set reserved in the sample data pre-processing process is used to detect the model performance and whether there is an overfitting problem; if the model fails to accurately reflect the relationship between the input and output in the test data, Bayesian optimization needs to be re-performed; conversely, if the model can accurately predict the test data, a verified second proxy model will be obtained.

5. The method for predicting photocurrent of a perovskite / crystalline silicon two-terminal tandem cell according to claim 3, characterized in that: The first proxy model is trained and optimized based on a Gaussian regression model with Bayesian optimization. After the optimization is completed, the regression model parameters and hyperparameters of the perovskite material are recorded.

6. A method for predicting photocurrent of a perovskite / crystalline silicon two-terminal tandem cell according to claim 1 or 2, characterized in that: In step S3, the parameters to be optimized by Bayesian optimization include the number of hidden layers of the artificial neural network; the hidden layer activation function; the hidden layer size; the initial learning rate; the minimum gradient descent and the momentum of the gradient descent.

7. A method for predicting photocurrent of a perovskite / crystalline silicon two-terminal tandem cell according to claim 1 or 2, characterized in that: The total photocurrent of the stacked cell obtained by the calculation is specifically: The photocurrent of the perovskite and crystalline silicon cells is calculated based on the vector integral corresponding to the predicted wavelength. The photocurrent of the stacked cell is the minimum photocurrent of the perovskite and crystalline silicon cells.

8. The method for predicting photocurrent of a perovskite / crystalline silicon two-terminal tandem cell according to claim 2, characterized in that: After the first result is input into the database, the perovskite layer thickness and velvet height in the sample data are calibrated. Specifically, the perovskite thickness and velvet height are fine-tuned by minimizing the error between the quantum efficiency obtained by simulation and the actual quantum efficiency.

9. A method for predicting photocurrent of a perovskite / crystalline silicon two-terminal tandem cell according to claim 1 or 2, characterized in that: The pre-processing of the sample data includes dividing the data set of the sample data into training, verification and test data sets for training and optimization of the second proxy model.

10. A photocurrent prediction system for a perovskite / crystalline silicon two-terminal tandem cell, applicable to the photocurrent prediction method for a perovskite / crystalline silicon two-terminal tandem cell according to any one of claims 1 to 9, characterized in that: include: A first proxy model construction module is configured to construct a first proxy model using the perovskite material, band gap, and wavelength of the tandem cell as input and the complex refractive index as output; A first result generating module, combining with a three-dimensional wave optics model to output a first result; a second proxy model construction module, generating a second proxy model based on the first result and the three-dimensional wave optics model training; The total photocurrent calculation module inputs the full wavelength into the second proxy model to output the predicted wavelength, and obtains the total photocurrent of the stacked cell through integral calculation.

Images