Solar cell thermal buffer layer material screening method and system based on neural network

Abstract

The invention discloses a solar cell thermal buffer layer material screening method and system based on a neural network, and relates to the technical field of materials science, and the method comprises the steps: building a solar cell thermal buffer layer material data set according to various material performance parameters in a Matweb database; based on the material data set, training a dense connection convolutional neural network model for predicting the thermal expansion coefficient and the elastic modulus of the material; according to the stress simulation data of the target solar battery pack in the specific employment environment, the tuning matching value of the thermal expansion coefficient and the elastic modulus is determined; inputting a random array into the trained model to obtain a material performance array corresponding to the tuning matching value of the thermal expansion coefficient and the elastic modulus; and according to the material performance array, based on the solar cell thermal buffer layer material data set, screening the solar cell thermal buffer layer material of the target solar cell pack in the specific employment environment. According to the invention, the material screening efficiency can be improved.

Description

Technical Field

[0001] This application relates to the technical field of materials science, and particularly to a method and system for screening materials for a thermal buffer layer of a solar cell based on a neural network. Background Art

[0002] Driven by the dual imperatives of global energy transition and environmental protection, solar energy, as a clean and sustainable energy source, is being increasingly emphasized in its development and utilization. As the core device for converting solar energy into electrical energy, the stability of the performance and the improvement of the efficiency of solar cells have always been the goals pursued by scientific researchers. However, during the operation of solar cells, the thermal stress problem has always been one of the key factors restricting the performance and long-term stable operation. For example, space solar cells are periodically affected by solar irradiation and cold black environment during their service life, experiencing strong temperature differences. The strong thermal shock, cyclic thermal load, and space charge formed during this process can lead to the degradation and even failure of the mechanical strength, structural integrity, and electrical properties of solar cells.

[0003] To alleviate the thermal stress problem of solar cells, various technical solutions have been developed. For example, a material with a thermal expansion coefficient matching that of the silicon wafer is selected as the encapsulation material to reduce the thermal stress caused by the thermal expansion and contraction effect. In the design of the battery structure, special structures such as cracks and holes are also innovatively introduced. They can serve as release paths for thermal stress, deforming or breaking when the thermal stress reaches the critical value, thereby effectively absorbing and dispersing the thermal stress and ensuring the safety and integrity of the device structure. Among them, adding an appropriate buffer layer between the solar cell and the encapsulation layer has become the most widely used solution, because its thickness, composition, and structure can be flexibly customized according to actual needs to meet the performance requirements of different application scenarios.

[0004] Currently, some possible solutions have been provided for the design of the buffer layer of solar cells. For example, Patent CN106684176A introduces a preparation method for a buffer layer material of a thin-film solar cell, using an organic zinc salt and a thio-organic compound as raw materials, completely solving the problem of cadmium metal pollution. In addition, Patent CN102544237A uses Zn(O,S) as the buffer layer of a copper indium gallium selenide thin-film solar cell, which not only makes the production of copper indium gallium selenide thin-film solar cells more environmentally friendly and economical, but also allows more incident photons to pass through the buffer layer because the bandgap width of Zn(O,S) is larger than that of CdS. The material parameters of the buffer layer are affected by various factors, such as the chemical composition, crystallization state, crystal structure, and bond strength of the material. However, using this traditional empirical trial-and-error method requires a large amount of empirical guidance in the design of new materials, consuming time and resources, and the results are uncertain and accidental. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for screening materials for the thermal buffer layer of solar cells based on neural networks. According to the thermal expansion coefficient and elastic modulus required for the materials of the thermal buffer layer of solar cells, the multi-parameter performance corresponding to the materials can be found through data learning in a short time, thereby improving the efficiency of material screening.

[0006] To achieve the above object, this application provides the following solutions:

[0007] In the first aspect, this application provides a method for screening materials for the thermal buffer layer of solar cells based on neural networks, including:

[0008] Based on the performance parameters of various materials in the Matweb database, a dataset of materials for the thermal buffer layer of solar cells is established; the material performance parameters include the thermal expansion coefficient, elastic modulus, physical properties, mechanical properties, and electrical properties of the materials;

[0009] Based on the dataset of materials for the thermal buffer layer of solar cells, a densely connected convolutional neural network model for predicting the thermal expansion coefficient and elastic modulus of materials is trained to obtain a trained densely connected convolutional neural network model; the trained densely connected convolutional neural network model takes the physical properties, mechanical properties, and electrical properties of the materials in the dataset of materials for the thermal buffer layer of solar cells as feature inputs, and takes the corresponding thermal expansion coefficient and elastic modulus of the materials in the dataset of materials for the thermal buffer layer of solar cells as labels, and is trained using a learning rate adjustment strategy.

[0010] According to the stress simulation data of the target solar cell group in a specific working environment, determine the optimized matching values of the thermal expansion coefficient and elastic modulus;

[0011] Input a random array into the trained densely connected convolutional neural network model to obtain a material performance array corresponding to the optimized matching values of the thermal expansion coefficient and elastic modulus; the material performance array is composed of the physical properties, mechanical properties, and electrical properties of the materials corresponding to the optimized matching values of the thermal expansion coefficient and elastic modulus.

[0012] According to the material performance array, based on the dataset of materials for the thermal buffer layer of solar cells, screen the materials for the thermal buffer layer of the target solar cell group in a specific working environment.

[0013] Optionally, the physical properties include the density, water absorption rate, linear shrinkage rate, and melt flow rate of the materials; the mechanical properties include the elongation at break, plastic ball indentation hardness, and notched impact strength of the materials; the electrical properties include the dielectric constant and dielectric strength of the materials.

[0014] Optionally, input a random array into the trained densely connected convolutional neural network model to obtain the material property array corresponding to the tuned matching values of the coefficient of thermal expansion and elastic modulus, specifically including:

[0015] Obtain the tuned matching values Y* of the coefficient of thermal expansion and elastic modulus and a random array [X 0 , X 1 , X 2 ... X n ;

[0016] Input the random array into the trained densely connected convolutional neural network model to obtain the predicted value Y 0 ;

[0017] Calculate the difference ΔY between the first output Y 0 and the tuned matching value Y*, and compare the difference ΔY with the difference threshold ε;

[0018] If the difference ΔY is less than the difference threshold ε, then obtain the material property array [X 0 * , X 1 * , X 2 * ... X n * corresponding to the tuned matching values of the coefficient of thermal expansion and elastic modulus;

[0019] If the difference ΔY is greater than the difference threshold ε, then use the optimizer to iteratively tune the input to obtain the first output array [X 0 ’, X 1 ’, X 2 ’... X n ’] and update the output value Y 0 until the difference ΔY is less than the difference threshold ε.

[0020] Optionally, after establishing the solar cell thermal buffer layer material dataset according to various material property parameters in the Matweb database, it further includes:

[0021] Perform an optimization operation on the established solar cell thermal buffer layer material dataset to obtain an optimized solar cell thermal buffer layer material dataset; the optimization process includes data duplicate cleaning, noise value removal, and expansion of the dataset sample quantity.

[0022] Optionally, after performing an optimization operation on the established solar cell thermal buffer layer material dataset to obtain an optimized solar cell thermal buffer layer material dataset, it further includes:

[0023] Normalize the optimized dataset of solar cell thermal buffer layer materials, specifically including:

[0024] According to the formula Normalize the optimized dataset of solar cell thermal buffer layer materials;

[0025] Among them, X a_new,i is the data of the performance a of material i after normalization, and X a,i is the data of the true performance a of material i. min(X a,* ) is the minimum value of the performance data of all materials a in the dataset, and max(X a,* ) is the maximum value of the performance data of all materials a in the dataset.

[0026] Optionally, the dense connection convolutional neural network model is composed of a feature extraction module, a batch normalization layer Bn, a max pooling layer MaxPool1d, a dense block, a transition layer, and a linear layer; the dense block is composed of a batch normalization layer Bn, a Relu activation function layer, and a one-dimensional convolutional layer Conv1d for layer-by-layer dense connection; the transition layer is composed of a batch normalization layer Bn, a Relu activation function layer, a one-dimensional convolutional layer Conv1d, and an average pooling layer AvgPool1d connected in sequence; the feature extraction module is composed of multi-scale convolutions; the multi-scale convolutions include a one-dimensional convolution with a kernel of 1, a one-dimensional convolution with a kernel of 3, a dilated convolution, and a depthwise separable convolution with a combination of kernels of 5 and 1.

[0027] Optionally, the calculation formula for the average error rate in the learning rate adjustment strategy is:

[0028]

[0029] Among them, bias is the average error rate of the dense connection convolutional neural network model, y real is the true performance data of the material, and y predict is the material performance data predicted by the dense connection convolutional neural network model.

[0030] In a second aspect, the present application provides a neural network-based screening system for solar cell thermal buffer layer materials, including:

[0031] A dataset establishment module for establishing a dataset of solar cell thermal buffer layer materials according to the performance parameters of various materials in the Matweb database; the material performance parameters include the thermal expansion coefficient, elastic modulus, physical properties, mechanical properties, and electrical properties of the materials;

[0032] A model training module, which is used to train a densely connected convolutional neural network model for predicting the thermal expansion coefficient and elastic modulus based on the solar cell thermal buffer layer material dataset, so as to obtain a trained densely connected convolutional neural network model; the trained densely connected convolutional neural network model takes the physical properties, mechanical properties and electrical properties of the materials in the solar cell thermal buffer layer material dataset as feature inputs, and takes the thermal expansion coefficient and elastic modulus corresponding to the materials in the solar cell thermal buffer layer material dataset as labels, and is a model trained by using a learning rate adjustment strategy;

[0033] A simulation module, which is used to determine the optimized matching values of the thermal expansion coefficient and elastic modulus according to the stress simulation data of the target solar cell group under a specific working environment;

[0034] A calculation module, which is used to input a random array into the trained densely connected convolutional neural network model to obtain a material property array corresponding to the optimized matching values of the thermal expansion coefficient and elastic modulus; the material property array is composed of the physical properties, mechanical properties and electrical properties of the materials corresponding to the optimized matching values of the thermal expansion coefficient and elastic modulus;

[0035] A screening module, which is used to screen the solar cell thermal buffer layer materials of the target solar cell group under a specific working environment based on the solar cell thermal buffer layer material dataset according to the material property array.

[0036] Optionally, the calculation module specifically includes:

[0037] A data acquisition sub-module, which is used to acquire the optimized matching values Y* of the thermal expansion coefficient and elastic modulus and a random array [X 0 , X 1 , X 2 ... X n ;

[0038] Input the random array into the trained densely connected convolutional neural network model to obtain a predicted value Y 0 ;

[0039] A calculation sub-module, which is used to calculate the difference ΔY between the first output quantity Y 0 and the optimized matching value Y*, and compare the difference ΔY with a difference threshold ε;

[0040] If the difference ΔY is less than the difference threshold ε, then obtain the material property array [X 0 * , X 1 * , X 2 * ... X n* ;

[0041] If the difference ΔY is greater than the difference threshold ε, then use the optimizer to iteratively tune the input to obtain the first output array [X 0 ’, X 1 ’, X 2 ’... X n ’] and update the output value Y 0 until the difference ΔY is less than the difference threshold ε.

[0042] Optionally, it further includes:

[0043] An optimization processing module for performing optimization processing operations on the established solar cell thermal buffer layer material dataset to obtain an optimized solar cell thermal buffer layer material dataset; the optimization processing includes data duplicate cleaning, noise value removal, and expansion of the dataset sample quantity.

[0044] According to the specific embodiments provided by the present application, the following technical effects are disclosed in the present application:

[0045] The present application provides a method and system for screening solar cell thermal buffer layer materials based on a neural network. First, according to various material property parameters in the Matweb database, a solar cell thermal buffer layer material dataset is established. Then, a densely connected convolutional neural network model for predicting the thermal expansion coefficient and elastic modulus of materials is trained to obtain a trained model. This step, through the method of the neural network model, uses the data in the dataset to train the model, enabling the model to predict the thermal expansion coefficient and elastic modulus of materials, providing a fast and accurate prediction means for determining the subsequent tuning matching values and avoiding the time consumption and cost of traditional experimental methods. Then, according to the stress simulation data of the target solar cell group under a specific working environment, the tuning matching values of the thermal expansion coefficient and elastic modulus are determined. This step, through simulation analysis, can quickly determine the tuning matching values of the thermal expansion coefficient and elastic modulus that meet the requirements of the solar cell group under a specific working environment, providing a clear target for subsequent material screening. Next, a random array is input into the trained densely connected convolutional neural network model to obtain a material property array corresponding to the tuning matching values. This step, through the prediction of the model, can quickly obtain a material property array that meets the physical, mechanical, and electrical properties of the tuning matching values. Finally, according to the material property array of the material, solar cell thermal buffer layer materials meeting the conditions are screened out from the solar cell thermal buffer layer material dataset. This step, through the screening of the dataset, can quickly find materials that meet the conditions of the material property array, thereby improving the efficiency of material screening. Description of the Drawings

[0046] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for use in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0047] Figure 1 It is a schematic flowchart of a method for screening materials for the thermal buffer layer of a solar cell based on a neural network provided by an embodiment of the present application.

[0048] Figure 2 It is a schematic diagram of the selection of material performance indicators provided by an embodiment of the present application.

[0049] Figure 3 It is a structural diagram of a neural network model provided by an embodiment of the present application.

[0050] Figure 4 It is a schematic diagram of the functional modules of a system for screening materials for the thermal buffer layer of a solar cell based on a neural network provided by an embodiment of the present application. Detailed implementation manners

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0052] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0053] Embodiment 1

[0054] As Figure 1 shown, this embodiment provides a method for screening materials for the thermal buffer layer of a solar cell based on a neural network, including:

[0055] Step 101: Establish a dataset of materials for the thermal buffer layer of a solar cell according to various material performance parameters in the Matweb database; the material performance parameters include the thermal expansion coefficient, elastic modulus, physical properties, mechanical properties, and electrical properties of the material;

[0056] Step 102: Based on the dataset of the solar cell thermal buffer layer material, train a densely connected convolutional neural network model for predicting the thermal expansion coefficient and elastic modulus of materials to obtain a trained densely connected convolutional neural network model; the trained densely connected convolutional neural network model takes the physical properties, mechanical properties, and electrical properties of the materials in the solar cell thermal buffer layer material dataset as feature inputs, and takes the thermal expansion coefficient and elastic modulus corresponding to the materials in the solar cell thermal buffer layer material dataset as labels, and is trained using a learning rate adjustment strategy;

[0057] Step 103: Determine the optimized matching values of the thermal expansion coefficient and elastic modulus according to the stress simulation data of the target solar cell module under a specific working environment;

[0058] Step 104: Input a random array into the trained densely connected convolutional neural network model to obtain a material property array corresponding to the optimized matching values of the thermal expansion coefficient and elastic modulus; the material property array is composed of the physical properties, mechanical properties, and electrical properties of the materials corresponding to the optimized matching values of the thermal expansion coefficient and elastic modulus;

[0059] Step 105: Based on the material property array and the solar cell thermal buffer layer material dataset, screen the solar cell thermal buffer layer materials for the target solar cell module under a specific working environment.

[0060] Among them, in some embodiments, when performing step 101, it can be specifically as follows:

[0061] Sample the properties of materials from the Matweb database. The types of materials include metal materials, polymer materials, composite materials, etc. The collected material property parameters include physical properties, mechanical properties, and electrical properties in addition to the thermal expansion coefficient and elastic modulus.

[0062] Specifically, the construction of the dataset is mainly obtained by sampling from the Matweb database. Since the Matweb database includes metal materials, polymer materials, composite materials, etc., and is easy to operate with a complete retrieval function, it is therefore selected as the main data source. In addition to the target prediction parameters of the thermal expansion coefficient and elastic modulus, the physical properties, mechanical properties, and electrical properties of the materials are also selected. As Figure 2 shown, the physical properties include the density, water absorption rate, linear shrinkage rate, and melt flow rate of the material; the mechanical properties include the elongation at break, plastic ball indentation hardness, and notch impact strength of the material; the electrical properties include the dielectric constant and dielectric strength of the material.

[0063] Data synthesis mainly utilizes the TVAE model in SDV. SDV is a Python library, so it can be called conveniently and quickly. The TVAE model is used for the virtual synthesis of tabular data, and its advantage lies in the high quality of the generated data. At the same time, the K-S test can objectively observe whether there is a large difference in the theoretical distributions of the original and the added synthetic data sets. Generally, when the p-value is less than 0.05, there is a large gap. Through the supplementation and enhancement of synthetic data, the model can learn more data information and enhance the performance of the model.

[0064] In some embodiments, after performing step 101, it further includes:

[0065] Performing an optimization operation on the established solar cell thermal buffer layer material data set to obtain an optimized solar cell thermal buffer layer material data set; the optimization process includes data duplicate cleaning, noise value removal, and expansion of the data set sample quantity.

[0066] In some embodiments, before performing step 102, it further includes the construction of a neural network model, which can be specifically as follows:

[0067] The training process of this neural network first imports the established data set and performs preprocessing operations. Since there is a situation where the dimensions of each material feature input are inconsistent, it is necessary to perform normalization processing on each column of feature input data, scale each data to between 0 and 1, and the formula used is as follows:

[0068]

[0069] In the formula, X a_new,i is the data of the performance a of material i after normalization, X a,i is the data of the true performance a of material i, min(X a,* ) is the minimum value of the performance data of all materials a in the data set, and max(X a,* ) is the maximum value of the performance data of all materials a in the data set.

[0070] The construction of the neural network model is an extension of machine learning and has a stronger target learning ability. By applying the neural network, complex data can be effectively processed. In this embodiment, the model architecture of the densely connected convolutional network is mainly used.

[0071] Such as Figure 3As shown in the figure, the model consists of input data, a feature extraction module, a batch normalization layer Bn, and a max pooling layer MaxPool1d, which are used to initially extract features and stabilize training, and then transfer them to the dense block. The dense block is mainly composed of a batch normalization layer Bn, a Relu activation function layer, and a one-dimensional convolutional layer Conv1d for layer-by-layer dense connection; the transition layer is mainly composed of a batch normalization layer Bn, a Relu activation function layer, a one-dimensional convolutional layer Conv1d, and an average pooling layer AvgPool1d connected in sequence. The role of the dense block is to integrate all the previous outputs as inputs, and the role of the transition layer is to compress features and optimize calculations. Finally, the linear layer Linear outputs the predicted value of the target.

[0072] Specifically, the feature extraction module consists of multi-scale convolutions; the multi-scale convolutions include a one-dimensional convolution with a kernel of 1, a one-dimensional convolution with a kernel of 3, a dilated convolution, and a depthwise separable convolution composed of a combination of a kernel of 5 and 1.

[0073] Among them, multi-scale convolution is a convolution process with multiple branches and different kernels:

[0074] (1) One-dimensional convolution Conv1D (kernel = 1).

[0075] (2) One-dimensional convolution Conv1D (kernel = 3).

[0076] (3) Dilated convolution Dilated_Conv1D (kernel = 3), where a certain dilation coefficient is set.

[0077] (4) A combination of depthwise convolution Depthwise_Conv1D (kernel = 5) and pointwise convolution Pointwise_Conv1D (kernel = 1) is called depthwise separable convolution, where depthwise convolution extracts features channel by channel, and pointwise convolution fuses information.

[0078] Among them, standard one-dimensional convolution kernels use kernels with sizes of 1 and 3 respectively to extract local features, that is, adjacent features; dilated convolution is mainly used to capture global features, that is, features at a farther distance; depthwise separable convolution is composed of depthwise convolution and pointwise convolution, aiming to achieve lightweight calculations.

[0079] Specifically, as Figure 3 shown, after the input data is preprocessed, it passes through branch calculations and applies different convolutional kernel layers to obtain corresponding values. Subsequently, the data dimensions and formats are adjusted and calculated to ensure the data format, such as (number of samples, output sequence length, number of output channels), where the number of samples is consistent with the number of output channels.

[0080] Specifically, the data of the four branches is processed by weighted averaging to obtain the data after feature extraction. The allocation of weights is dynamically adjusted according to the change of the mean square error value Loss during the model training process. For example, in the conventional training stage, one-dimensional convolution and depthwise separable convolution are given higher weights; when the training loss cannot continue to decrease after a long time of training, the weight of dilated convolution will be increased while the weights of others are decreased.

[0081] Next, the data is introduced into the batch normalization layer Bn and the max pooling layer MaxPool1d. The role of the batch normalization layer Bn is to accelerate the training process; the role of the max pooling layer MaxPool1d is to further reduce the data dimension and extract features. After that, the data is introduced into the dense convolution module, that is, the main network.

[0082] In the preprocessed material dataset, the sample data is randomly assigned to the training and test sets. During the training process, a learning rate adjustment strategy is used, such as the step adjustment of the learning rate. Every few training epochs, the learning rate is reduced by a certain proportion, which helps to achieve stable convergence. And an early stopping mechanism is set to determine whether the loss decreases within a certain number of epochs, which helps to prevent overfitting. In addition, the average error rate of the test set is calculated for the evaluation of the prediction model, and the calculation formula is as follows:

[0083]

[0084] where bias is the average error rate of the prediction model, y real is the true performance data of the material, which can be selected as the coefficient of thermal expansion or elastic modulus in this application, y predict is the corresponding material performance data predicted by the model.

[0085] Among them, in some embodiments, when performing step 103, it can be specifically as follows:

[0086] Simulate the coefficient of thermal expansion and elastic modulus of the battery material structure in the expected environment. Determine a certain small interval or specific value of the design battery buffer layer requirement under this condition for material selection and performance optimization guidance. For example, in a certain outer space environment, for the buffer layer required by the solar cell, after investigating materials such as the metal electrode layer and the substrate layer, it can be roughly determined that the coefficient of thermal expansion is between 23 - 55, and its unit is μm / m·°C. After the mechanical simulation performance test, it can be roughly reduced to 40.

[0087] Among them, in some embodiments, when performing step 104, it can be specifically as follows:

[0088] Obtain the tuned matching value Y* of the coefficient of thermal expansion and elastic modulus and a random array [X0 , X 1 , X 2 ... X n ;

[0089] Input the random array into the trained densely connected convolutional neural network model to obtain the predicted value Y 0 ;

[0090] Calculate the difference ΔY between the first output quantity Y 0 and the tuning matching value Y*, and compare the difference ΔY with the difference threshold ε;

[0091] If the difference ΔY is less than the difference threshold ε, obtain the material property array corresponding to the tuning matching value of the coefficient of thermal expansion and elastic modulus [X 0 * , X 1 * , X 2 * ... X n * ;

[0092] If the difference ΔY is greater than the difference threshold ε, use the optimizer to iteratively tune the input to obtain the first output array [X 0 ’, X 1 ’, X 2 ’... X n ’] and update the output value Y 0 until the difference ΔY is less than the difference threshold ε.

[0093] Among them, in some embodiments, when performing step 105, it can be specifically as follows:

[0094] Obtain the material property parameter array [X 0 *, X 1 *, X 2 *... X n *] predicted by the training model;

[0095] Introduce the existing real material data set file, and synchronize and normalize the material property parameter array [X 0 *, X 1 *, X 2 *... X n *] with the actual material data set to remove the influence of unit dimension;

[0096] Perform the Euclidean distance calculation between arrays, that is, rely on the formula:

[0097]

[0098] where d represents the distance between the predicted material properties and the actual materials, Each material property value representing the parameter array, represents the various material property values of the i-th material;

[0099] Compare and calculate the distances between each actual material and the best prediction, and verify whether the actual target values of these materials are within a certain range of the required values, and select appropriate materials for export.

[0100] Specifically, the following example is given in this embodiment:

[0101] In a certain outer space environment, for the buffer layer required by the solar cell, after investigating materials such as the metal electrode layer and the substrate layer, it can be roughly determined that the coefficient of thermal expansion is between 23 and 55, and its unit is μm / m · ℃. After mechanical simulation performance testing, it can be roughly narrowed down to 40. The specific screening is as follows:

[0102] 1) Initial import of an input array, such as [1.0901e+00, -6.4940e-02, 5.8538e-02, 9.6425e-01, 1.6035e-02, 2.8951e+01, 1.4792e+01, 1.4398e+02, 8.9423e-02];

[0103] 2) Through the first time, its prediction result is: 75.0738;

[0104] 3) Through multiple iterative optimizations, its prediction result is: 40.0013;

[0105] 4) The final optimized input parameters are obtained as: [1.1973e+00, 1.1207e-03, 7.2628e-02, 3.0674e+00, 1.1956e-02, 3.8532e+01, 1.0350e+01, 1.5141e+02, 1.0153e-01], and the material characteristics they represent are in the following order: density, water absorption rate, elongation at break, dielectric constant, linear shrinkage rate, dielectric strength, melt flow rate, plastic ball indentation hardness, notched impact strength. Through this array, specific materials can be approximately obtained from the established data set.

[0106] 5) Through multiple optimizations and searches, it can be found in the existing material data set:

[0107] (1) SABIC VALOX TMVX5011 PBT (Europe), its actual coefficient of thermal expansion is 38.35, and the parameters of the material are [1.310, 0.0015, 0.0300, 3.100, 0.00550, 18.00, 18.00, 165.0, 0.400];

[0108] (2) Covestro 9415 Polycarbonate, 10% Glass Filled, its actual coefficient of thermal expansion is 40.00, and the parameters of the material are [1.270, 0.0026, 0.1500, 3.200, 0.00500, 36.00, 7.00, 128.0, 0.100];

[0109] (3) SABIC LEXAN TM FR 505R PC (Asia), its actual coefficient of thermal expansion is 43.40, and the parameters of the material are [1.250, 0.0012, 0.0700, 2.800, 0.00400, 16.00, 7.00, 115.0, 0.900].

[0110] Therefore, in the first embodiment, according to the stress simulation data of the battery pack under a specific working environment, a certain small interval or specific value of the appropriate coefficient of thermal expansion and elastic modulus is obtained. Based on the application of the neural network, first, a random input array is used, and it is necessary to ensure that the number of columns of the input parameters is the same as that during training. The array is iteratively optimized by the neural network model optimizer to make the predicted value continuously approach the target value. Finally, an array containing multiple parameters is obtained, which represents the multi-dimensional performance of a material, and the material that best matches it is selected from the database.

[0111] The second embodiment

[0112] As Figure 4 shown, this embodiment provides a neural network-based screening system for solar cell thermal buffer layer materials, including:

[0113] A dataset establishment module 401, which is used to establish a dataset of solar cell thermal buffer layer materials according to the performance parameters of various materials in the Matweb database; the material performance parameters include the coefficient of thermal expansion, elastic modulus, physical properties, mechanical properties, and electrical properties of the material;

[0114] The model training module 402 is used to train a densely connected convolutional neural network model for predicting the thermal expansion coefficient and elastic modulus based on the solar cell thermal buffer layer material dataset, so as to obtain a trained densely connected convolutional neural network model. The trained densely connected convolutional neural network model takes the physical properties, mechanical properties, and electrical properties of the materials in the solar cell thermal buffer layer material dataset as feature inputs, and takes the corresponding thermal expansion coefficient and elastic modulus of the materials in the solar cell thermal buffer layer material dataset as labels, and is trained using a learning rate adjustment strategy.

[0115] The simulation module 403 is used to determine the optimized matching values of the thermal expansion coefficient and elastic modulus according to the stress simulation data of the target solar cell group under a specific working environment.

[0116] The calculation module 404 is used to input a random array into the trained densely connected convolutional neural network model to obtain a material property array corresponding to the optimized matching values of the thermal expansion coefficient and elastic modulus. The material property array is composed of the physical properties, mechanical properties, and electrical properties of the materials corresponding to the optimized matching values of the thermal expansion coefficient and elastic modulus.

[0117] The screening module 405 is used to screen the solar cell thermal buffer layer materials of the target solar cell group under a specific working environment based on the material property array and the solar cell thermal buffer layer material dataset.

[0118] Among them, the calculation module 404 specifically includes:

[0119] The data acquisition sub-module is used to acquire the optimized matching values Y* of the thermal expansion coefficient and elastic modulus and a random array [X 0 , X 1 , X 2 ... X n ;

[0120] Input the random array into the trained densely connected convolutional neural network model to obtain the predicted value Y 0 ;

[0121] The calculation sub-module is used to calculate the difference ΔY between the first output quantity Y 0 and the optimized matching value Y*, and compare the difference ΔY with the difference threshold ε;

[0122] If the difference ΔY is less than the difference threshold ε, then obtain the material property array [X 0 * , X 1 * , X 2 * ... Xn * ;

[0123] If the difference ΔY is greater than the difference threshold ε, then use the optimizer to iteratively optimize the input to obtain the first output array [X 0 ’, X 1 ’, X 2 ’... X n ’] and update the output value Y 0 until the difference ΔY is less than the difference threshold ε.

[0124] The system further includes:

[0125] An optimization processing module, configured to perform an optimization processing operation on the established dataset of solar cell thermal buffer layer materials to obtain an optimized dataset of solar cell thermal buffer layer materials; the optimization processing includes data duplicate cleaning, noise value removal, and expansion of the number of dataset samples.

[0126] In summary, the present application has the following beneficial effects:

[0127] 1) The present application can, for the thermal expansion coefficient and elastic modulus required for the solar cell thermal buffer layer materials in a specific environment, find the corresponding multi-dimensional properties of the materials, such as density, elongation at break, etc., in a short time through learning of the data, improving the efficiency of material screening and design.

[0128] 2) The present application establishes a neural network model, which is more efficient and faster compared to traditional experimental preparation and testing, and can predict the thermal expansion coefficient or elastic modulus through a certain input data column of material parameters.

[0129] 3) The present application combines the thermal expansion coefficient and elastic modulus data of the battery layer materials in the simulation environment, and by means of iterative optimization, changes the specific values of each column of the input array representing the material properties, so that the predicted value of the current input array continuously approaches the required value.

[0130] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0131] In this article, specific examples are used to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for screening solar cell thermal buffer layer materials based on neural network, characterized in that: The method for screening solar cell thermal buffer layer materials based on neural network includes: According to various material performance parameters in the Matweb database, a solar cell thermal buffer layer material data set is established; the material performance parameters include the thermal expansion coefficient, elastic modulus, physical properties, mechanical properties and electrical properties of the material; Based on the solar cell thermal buffer layer material data set, a densely connected convolutional neural network model for predicting the thermal expansion coefficient and elastic modulus of the material is trained to obtain a trained densely connected convolutional neural network model; the trained densely connected convolutional neural network model is a model obtained by training using the physical properties, mechanical properties and electrical properties of the materials in the solar cell thermal buffer layer material data set as feature inputs, and using the thermal expansion coefficient and elastic modulus corresponding to the materials in the solar cell thermal buffer layer material data set as labels, and adopting a learning rate adjustment strategy; Determine the optimized matching values ​​of the thermal expansion coefficient and the elastic modulus according to the stress simulation data of the target solar cell group under a specific working environment; Inputting a random array into a trained densely connected convolutional neural network model to obtain a material property array corresponding to the tuned matching values ​​of the thermal expansion coefficient and the elastic modulus; the material property array is composed of physical properties, mechanical properties, and electrical properties of the material corresponding to the tuned matching values ​​of the thermal expansion coefficient and the elastic modulus; According to the material performance array and based on the solar cell thermal buffer layer material data set, the solar cell thermal buffer layer materials for the target solar cell group under a specific working environment are screened.

2. The method for screening solar cell thermal buffer layer materials based on neural network according to claim 1, characterized in that: The physical properties include the density, water absorption, linear shrinkage and melt flow rate of the material; the mechanical properties include the elongation at break, plastic ball indentation hardness and notched impact strength of the material; and the electrical properties include the dielectric constant and dielectric strength of the material.

3. The method for screening solar cell thermal buffer layer materials based on neural network according to claim 1, characterized in that: A random array is input into the trained densely connected convolutional neural network model to obtain a material property array corresponding to the tuned matching values ​​of the thermal expansion coefficient and elastic modulus, specifically including: Get the optimized matching value Y* of the thermal expansion coefficient and elastic modulus and a random array [X0,X1,X2...X n ]; Input the random array into the trained densely connected convolutional neural network model to obtain the predicted value Y0; Calculate a difference ΔY between the first output quantity Y0 and the tuning matching value Y*, and compare the difference ΔY with a difference threshold ε; If the difference ΔY is less than the difference threshold ε, the material property array corresponding to the tuned matching value of the thermal expansion coefficient and the elastic modulus is obtained [X0 * ,X1 * ,X2 * ...X n * ]; If the difference ΔY is greater than the difference threshold ε, the optimizer is used to iteratively tune the input to obtain the first output array [X0 ’ ,X1 ’ ,X2 ’ ...X n ’ ] and updates the output value Y0 until the difference ΔY is less than the difference threshold ε.

4. The method for screening solar cell thermal buffer layer materials based on neural network according to claim 1, characterized in that: After establishing the solar cell thermal buffer material data set based on various material performance parameters in the Matweb database, it also includes: The established solar cell thermal buffer layer material data set is optimized to obtain an optimized solar cell thermal buffer layer material data set; the optimization process includes repeated data cleaning, noise value removal, and expansion of the number of data set samples.

5. The method for screening solar cell thermal buffer layer materials based on neural network according to claim 4, characterized in that: After optimizing the established solar cell thermal buffer layer material data set to obtain an optimized solar cell thermal buffer layer material data set, the method further includes: The optimized solar cell thermal buffer material data set is normalized, including: According to the formula Normalize the optimized solar cell thermal buffer material data set; Among them, X a_new,i is the data of material i’s normalized performance a, X a,i is the data of the real performance a of material i, min(X a,* ) is the minimum value of all material a performance data in the data set, max(X a,* ) is the maximum value of all material a performance data in the data set.

6. The method for screening solar cell thermal buffer layer materials based on neural network according to claim 1, characterized in that: The densely connected convolutional neural network model consists of a feature extraction module, a batch normalization layer Bn, a maximum pooling layer MaxPool1d, a dense block, a transition layer and a linear layer; the dense block is composed of a batch normalization layer Bn, a Relu activation function layer, and a one-dimensional convolution layer Conv1d that are densely connected layer by layer; the transition layer is composed of a batch normalization layer Bn, a Relu activation function layer, a one-dimensional convolution layer Conv1d and an average pooling layer AvgPool1d that are connected in sequence; the feature extraction module is composed of multi-scale convolutions; the multi-scale convolutions include a one-dimensional convolution with a convolution kernel of 1, a one-dimensional convolution with a convolution kernel of 3, a dilated convolution, and a depthwise separable convolution with a convolution kernel of 5 and 1.

7. The method for screening solar cell thermal buffer layer materials based on neural network according to claim 1, characterized in that: The calculation formula of the average error rate in the learning rate adjustment strategy is: Among them, bias is the average error rate of the densely connected convolutional neural network model, y real is the actual performance data of the material, y predict Material property data predicted by the densely connected convolutional neural network model.

8. A solar cell thermal buffer layer material screening system based on neural network, characterized in that: include: A data set establishment module is used to establish a solar cell thermal buffer layer material data set based on various material performance parameters in the Matweb database; the material performance parameters include the thermal expansion coefficient, elastic modulus, physical properties, mechanical properties and electrical properties of the material; A model training module is used to train a densely connected convolutional neural network model for predicting the thermal expansion coefficient and elastic modulus of a material based on the solar cell thermal buffer layer material data set to obtain a trained densely connected convolutional neural network model; the trained densely connected convolutional neural network model is a model trained using the physical properties, mechanical properties and electrical properties of the materials in the solar cell thermal buffer layer material data set as feature inputs, the thermal expansion coefficient and elastic modulus corresponding to the materials in the solar cell thermal buffer layer material data set as labels, and a learning rate adjustment strategy; A simulation module, used to determine the tuning matching values ​​of the thermal expansion coefficient and the elastic modulus according to the stress simulation data of the target solar cell group under a specific working environment; A calculation module, for inputting a random array into a trained densely connected convolutional neural network model to obtain a material property array corresponding to the tuned matching values ​​of the thermal expansion coefficient and the elastic modulus; the material property array is composed of physical properties, mechanical properties and electrical properties of the material corresponding to the tuned matching values ​​of the thermal expansion coefficient and the elastic modulus; The screening module is used to screen the solar cell thermal buffer layer materials of the target solar cell group under a specific working environment according to the material performance array and based on the solar cell thermal buffer layer material data set.

9. The solar cell thermal buffer layer material screening system based on neural network according to claim 8, characterized in that: The computing module specifically includes: The data acquisition submodule is used to obtain the tuning matching value Y* of the thermal expansion coefficient and elastic modulus and a random array [X0, X1, X2...X n ]; Input the random array into the trained densely connected convolutional neural network model to obtain the predicted value Y0; A calculation submodule, used for calculating a difference ΔY between the first output quantity Y0 and the tuning matching value Y*, and comparing the difference ΔY with a difference threshold ε; If the difference ΔY is less than the difference threshold ε, the material property array corresponding to the tuned matching value of the thermal expansion coefficient and the elastic modulus is obtained [X0 * ,X1 * ,X2 * ...X n * ]; If the difference ΔY is greater than the difference threshold ε, the optimizer is used to iteratively tune the input to obtain the first output array [X0 ’ ,X1 ’ ,X2 ’ ...X n ’ ] and updates the output value Y0 until the difference ΔY is less than the difference threshold ε.

10. The solar cell thermal buffer layer material screening system based on neural network according to claim 9, characterized in that: Also includes: An optimization processing module is used to perform an optimization processing operation on the established solar cell thermal buffer layer material data set to obtain an optimized solar cell thermal buffer layer material data set; The optimization process includes repeated data cleaning, noise value removal and expansion of the number of data set samples.

Images