CN113919576A

CN113919576A - Method for predicting structural performance of solar cell

Info

Publication number: CN113919576A
Application number: CN202111203503.5A
Authority: CN
Inventors: 黄凯; 江莹; 姜卓颖; 李琳; 李澄; 李金钗; 张�荣; 康俊勇
Original assignee: Xiamen University; Tan Kah Kee Innovation Laboratory
Current assignee: Xiamen University; Tan Kah Kee Innovation Laboratory
Priority date: 2021-10-15
Filing date: 2021-10-15
Publication date: 2022-01-11

Abstract

The invention relates to the technical field of solar cells, and provides a method for predicting the performance of a solar structure, which mainly collects and extracts input characteristic parameters and corresponding output characteristic parameters of the solar cell structure, establishes a corresponding data set and preprocesses the data in the data set according to a known criterion; building a model by using a machine learning algorithm, and performing structural parameter setting and initialization training on the model; training and optimizing the model after the initial training of the structural parameters by using the preprocessed data set to obtain a prediction model; and inputting the test data of the input characteristic parameters of the solar cell structure to be predicted into the prediction model, and further obtaining the predicted value of the output characteristic parameters of the solar cell structure to be predicted. Therefore, the performance of the solar cell structure can be rapidly predicted, the operation is simple and convenient, and the accuracy is high.

Description

Method for predicting structural performance of solar cell

Technical Field

The invention relates to the technical field of solar cells, in particular to a method for predicting the structural performance of a solar cell by using a machine learning algorithm model.

Background

The solar cell is a photoelectric semiconductor sheet which directly generates electricity by using sunlight, is also called as a solar chip or a photovoltaic cell, and can output voltage instantly and generate current under the condition of a loop as long as the solar cell is illuminated under a certain illumination condition. Physically referred to as solar Photovoltaic (abbreviated PV), Photovoltaic for short. The solar cell is mainly based on semiconductor materials, and the working principle of the solar cell is that photovoltaic reaction occurs after photoelectric materials absorb light energy. Solar cells can also be said to be devices that directly convert light energy into electrical energy by the photoelectric or photochemical effect. Solar cells can convert solar energy directly into electrical energy, which is one of the most efficient forms of clean energy.

The photovoltaic cells sold in the market are mainly monocrystalline silicon photovoltaic cells produced by taking monocrystalline silicon as raw materials. With the continuous development of the solar cell industry, other various photovoltaic cell technologies are emerging. The large-scale application of photovoltaic cells requires continuous technical improvement to improve the conversion efficiency (also referred to as photoelectric conversion efficiency or power generation efficiency) of photovoltaic cells and to reduce production costs. Currently, in the research and development of new types of solar cells, the multi-junction solar cell technology has gradually become a hot spot of solar cell research due to a series of advantages of high conversion efficiency, excellent radiation resistance, stable temperature characteristics, easy mass production and the like.

However, there is still a certain blindness in the design of the multi-junction solar cell structure, and the empirical comparison is mainly relied on to improve the performance. The method has the obvious defects of low efficiency, difficulty in obtaining the optimal condition of manually selected parameters in simulation, high experimental verification cost and the like. Meanwhile, machine learning is a common research hotspot in the fields of artificial intelligence and pattern recognition, and the theory and the method thereof are widely applied to solving the complex problems in the fields of engineering application and science. In the era of rapid development of the current internet technology, if a machine learning method in artificial intelligence is applied to the structural design of a multi-junction solar cell, the efficiency improvement which can be realized may be multiplied in geometric order.

In the structural design of a new type of solar cell, how to predict the structural performance of the designed solar cell by means of a machine learning method and adjust the design scheme of the new type of solar cell structure in time by means of the prediction result to obtain a solar cell or a photovoltaic cell with better efficiency and a corresponding electronic device has become one of the problems to be actively solved by the technical personnel in the field.

Disclosure of Invention

In order to solve the problem of the structural performance prediction of the novel solar cell in the prior art, the invention provides a method for predicting the structural performance of the solar cell, which can realize the prediction of the structural performance of the solar cell by using different algorithm models in machine learning and timely adjust the scheme of the structural design of the solar cell according to the guidance of the prediction result, so that the structural design of the solar cell has more excellent performance on the whole, such as better photoelectric conversion efficiency and longer service life of a product.

In one embodiment, a method for predicting the performance of a solar cell structure may include the steps of: (S1) collecting and extracting data of input characteristic parameters and corresponding output characteristic parameters of the solar cell structure, and dividing the data into an original data set and a predicted data set; (S2) preprocessing the raw data set and the predicted data set to obtain a preprocessed raw data set and a preprocessed predicted data set; (S3) constructing an initial model using a machine learning algorithm; (S4) carrying out structural parameter setting on the initial model, and carrying out initialization training on the structural parameters to obtain an initialized model; (S5) optimizing the initialized model, and training the initialized model by using the preprocessed original data set to obtain corresponding network weight and bias so as to obtain a prediction model; (S6) forecasting, namely, inputting the preprocessed test data set in the input characteristic parameters of the solar cell structure to be forecasted into the forecasting model, and further obtaining the forecasting value of the output characteristic parameters of the solar cell structure to be forecasted.

In an embodiment, the machine learning algorithm may include, but is not limited to, one of a deep learning algorithm, a multi-layered perceptron, a decision tree, a linear regression, a gradient boosting regression, and a K-nearest neighbor algorithm. Wherein, the deep learning algorithm can include but is not limited to one of convolutional neural network, self-coding and deep confidence network.

In one embodiment, the solar cell is a multi-junction solar cell structure, and includes at least one bottom cell and a plurality of sub-cells, and the plurality of sub-cells are located above the bottom cell.

In one embodiment, the bottom cell may include a substrate, an emission layer, a window layer, and a tunnel junction along a stacking direction, and the plurality of sub-cells are stacked over the tunnel junction of the bottom cell. Each sub-cell includes, in the stacking direction, a back field layer, a base region, an emitter layer, an window layer, and a tunnel junction. And the uppermost layer of the sub-battery positioned at the topmost layer or the uppermost layer is a contact layer.

In an embodiment, the input characteristic parameters of the solar cell structure may include, but are not limited to, the thickness of each layer in the solar cell structure, the stacking manner between layers, the shape of each layer, and the composition material and material component ratio. The corresponding output characteristic parameters include, but are not limited to, a short circuit current density, an open circuit voltage, and a fill factor of the solar cell structure. The data of the input characteristic parameters and the corresponding output characteristic parameters of the solar cell structure can be screened and adjusted according to the type of the solar cell structure. In other words, the input characteristic parameters and the output characteristic parameters of the selected solar cell structure can be reduced or increased according to requirements.

In one embodiment, the method of pre-processing the raw data set and the predicted data set may comprise the steps of: (1) selecting characteristics, namely selecting input characteristic parameters of the solar cell structure according to known physical knowledge and the relation between data; (2) data processing, namely performing normalization processing on the selected feature data; (3) and data reorganization, namely reorganizing the size of the processed characteristic data.

In an embodiment, after the selected feature data is normalized, the data mean value of the feature parameter is 0 and the standard deviation is 1.

In one embodiment, in the step of optimizing the initialized model, a training result of the initialized model is determined by using a mean square error, and the mean square error formula is as follows:

wherein Predict_i、Actual_iThe predicted value and the true value of the ith sample are respectively.

Based on the above, compared with the existing simulation software such as APSYS and the like, the method for predicting the structural performance of the solar cell provided by the invention has the following effects:

1. the method for predicting the performance of the solar cell structure by using different algorithm models in machine learning can realize rapid prediction of the performance of devices with different structures without considering whether the fitting of the network structure in the model is converged, and can better guide the optimization of the design scheme of the solar cell structure according to the prediction result.

2. The neural network model adopted in the machine learning algorithm can effectively prevent or reduce overfitting of the built neural network model by using strategies such as Dropout and the like, so that the accuracy of the built neural network model for predicting the structural performance of the solar cell is improved.

3. According to the invention, the big data is subjected to machine learning, then a corresponding neural network model is built, and then the performance of the solar cell overall structure material device with different component proportions and structures is predicted by the neural network model. Therefore, the rule of a more complex physical layer in the overall structure of the solar cell can be explored from a data layer, and the operation is simple and convenient.

Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts; in the following description, the drawings are illustrated in a schematic view, and the drawings are not intended to limit the present invention.

FIG. 1 is a schematic structural diagram of a solar cell according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a classical neural network model according to the present invention;

FIG. 3 is a schematic diagram of the structure of the convolutional neural network for predicting the structural performance of the solar cell in the present invention; and

fig. 4 is a flowchart illustrating a method for predicting structural performance of a solar cell according to an embodiment of the present invention.

Reference numerals:

1-multijunction solar cell 10-bottom cell 11-substrate

12-emitting layer 13-window layer 14-tunneling junction

A-sub-battery set 20-sub-battery 21-back surface field layer

22-base region 23-emitter layer 24-window layer

25-tunneling junction 26-contact layer

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments; the technical features designed in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other; all other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In the description of the present invention, it is to be noted that all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs, and are not to be construed as limiting the present invention; it will be further understood that terms used herein should be interpreted as having a meaning that is consistent with their meaning in the context of this specification and the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

In the research and development of new solar cell technology, the multijunction solar cell is a high-efficiency solar cell. Each cell has a plurality of thin films formed by molecular beam epitaxy or metalorganic chemical vapor deposition. The different semiconductors formed by these films have different characteristic energy gaps that absorb electromagnetic energy at specific frequencies in the spectrum. The resulting semiconductor is specifically designed to absorb most of the frequencies of sunlight, thereby generating more energy. The multi-junction solar cell has been widely used in the field of space through the development of the last ten years, and the efficiency record is continuously refreshed. For convenience of explanation and understanding, in the description of the embodiment of the present invention, the structure of the multi-junction solar cell is taken as an example, and a method for predicting the performance of the multi-junction solar cell structure is explained by using different algorithm models in machine learning.

Referring to fig. 1, fig. 1 is a schematic structural diagram of a solar cell according to an embodiment of the invention. As shown, the solar cell is a multi-junction solar cell, and a structure of the multi-junction solar cell includes a plurality of PN junctions (or PN junctions). The overall structure of the multi-junction solar cell 1 mainly includes at least one bottom cell 10 and a plurality of sub-cells 20, wherein the sub-cells 20 are located above the bottom cell 10 and are sequentially stacked above the bottom cell 10. The bottom cell 10 may include a substrate 11, an emission layer 12, a window layer 13, and a tunnel junction 14 along a stacking direction (bottom-up in fig. 1). The sub-cells 20 may be formed as a sub-cell set a stacked above the tunnel junction 14 of the bottom cell 10. Each subcell 20 may include, in the stacking direction (bottom-up in fig. 1), a back field layer 21, a base region 22, an emitter layer 23, an window layer 24, and a tunnel junction 25. Wherein the tunnel junction 25 in the topmost or uppermost sub-cell 20 is replaced by a contact layer 26.

Further, as shown in fig. 1, the back field layer 21 in the sub-cell 20 adjacent to the bottom cell 10 is stacked over the tunnel junction 14 of the bottom cell 10 in contact with the tunnel junction 14 in the bottom cell 10. In the sub-cells 20, the back field layer 21 of the next sub-cell 20 is in contact with the tunnel junction 25 of the adjacent previous sub-cell 20. In order to facilitate the contacting of the multijunction solar cell 1 with other devices, the tunnel junction 25 in the topmost or uppermost subcell 20 in the subcell set a is replaced by a contact layer 26. The contact layer 26 in the subcell 20 may serve as a current guiding layer, may be a material layer of the multijunction solar cell 1 to which electrodes are connected, and may serve as an ohmic contact. The contact layer 26 can form a good ohmic contact with a specific electrode metal material, thereby reducing the electric energy loss due to the series resistance thereof.

With the development and evolution of research and development, the machine learning methods for research and publication are various, and there are various classification methods according to the difference of the emphasis side. Based on the classification of learning strategies, machine learning can be divided into machine learning that simulates the human brain and machine learning that directly employs mathematical methods. Machine learning that simulates the human brain can be further divided into symbolic learning and neural network learning (or connection learning). Machine learning that directly adopts mathematical methods is mainly statistical machine learning. Based on the classification of learning methods, machine learning can be classified into inductive learning, deductive learning, analog learning, and analytical learning. Inductive learning can be further divided into symbolic inductive learning (e.g., exemplar learning, decision tree learning) and functional inductive learning (or discovery learning, e.g., neural network learning, exemplar learning, discovery learning, statistical learning). Based on the classification of learning manners, machine learning can be classified into supervised learning (learning with a mentor), unsupervised learning (learning without a mentor), and reinforcement learning (reinforcement learning). Based on the classification of the data form, machine learning can be classified into structured learning and unstructured learning. Based on the classification of learning objectives, machine learning can be classified into concept learning, rule learning, function learning, category learning, and bayesian network learning.

Algorithms that are more commonly used in machine learning include, but are not limited to, a decision tree algorithm, a naive bayes algorithm, a support vector machine algorithm, a random forest algorithm, an artificial neural network algorithm, a Boosting and Bagging algorithm, an association rule algorithm, an EM (expectation maximization) algorithm, and deep learning. Deep Learning (DL) is a new research direction in the field of Machine Learning (ML), and can learn the intrinsic rules and representation levels of sample data. The final goal of deep learning is to make a machine capable of human-like analytical learning, and to recognize data such as characters, images, and sounds.

Different deep learning models are constructed mainly based on neural networks. A neural network is an algorithmic mathematical model that mimics the behavior of a biological neural network and produces an output upon receiving multiple inputs. With the continuous development of the neural network and the iterative update of the deep learning algorithm, the structure of the network model is also continuously adjusted and optimized, and particularly, the method in the aspects of feature extraction and feature selection has a larger improvement space. The method can map any complex nonlinear relation, has strong robustness, memory capability, self-learning capability and the like, and has wide application in the aspects of classification, prediction, pattern recognition and the like.

According to the foregoing, in the examples of the present invention, the machine learning algorithm used may be a deep learning algorithm, a Neural network model (NN) multi-layer sensor, a decision tree, a linear regression, a Gradient Boosting Regression (GBR), or a K-nearest neighbors (KNN). The deep learning algorithm can be a convolutional neural network, a cyclic neural network, self-coding and a deep confidence network. How to predict the performance of the new type of solar cell structure by using the machine learning model will be explained by using the process of predicting the performance of the multi-junction solar cell structure by using different machine learning algorithm models.

The first embodiment is as follows: and predicting the performance of the multi-junction solar cell structure by utilizing a convolution neural network algorithm in deep learning.

Referring to fig. 2, fig. 2 is a schematic diagram of a classical neural network model according to the present invention. Neural Networks (NN) are an algorithmic mathematical model that simulates the human actual Neural Networks to process information, are complex network systems formed by widely interconnecting a large number of simple processing units (called neurons), and are highly complex nonlinear dynamical learning systems. A neuron is a multiple-input, single-output information processing unit that processes information non-linearly. Neural networks are a model in machine learning.

By analogy to the human real neural network, it is understood that a neural network is composed of neurons, nodes and connections (synapses) between nodes, each neural network element, also called a perceptron, which receives a plurality of inputs and generates an output. The actual neural network decision model is often a multi-layer network composed of multiple perceptrons. As shown in fig. 2, the classical neural network model is mainly composed of an input layer, a hidden layer, and an output layer. In the figure example, LayerL₁Representing the input layer, LayerL₂Representing hidden or hidden layers, LayerL₃Representing the output layer. Based on the advantages of the neural network model in machine learning, the characteristic factors of the multi-junction solar cell structure influencing the photoelectric efficiency of the electronic device can be predicted efficiently and accurately by means of the neural network model in the process of designing the multi-junction solar cell structure, and the design scheme of the multi-junction solar cell structure is adjusted in time according to the prediction result so as to accord with the expected overall functional effect.

In a feed-forward neural network, as in fig. 2, information moves forward in only one direction, from the input node, through the hidden node (if any) to the output node. There are no loops or loops in the network. In the field of machine learning, a typical deep learning model can mainly include three types: (1) a Neural Network system based on convolution operation, namely a convolution Neural Network (CNN for short); (2) self-Coding neural networks based on multi-layer neurons, including self-Coding (Auto Encoder) and Sparse Coding (Sparse Coding); (3) and pre-training in a multilayer self-coding neural network mode, and further optimizing a Deep Belief Network (DBN) of the neural network weight by combining the identification information.

In addition to the three typical deep learning models, the deep learning model may be a Recurrent Neural network (Recurrent Neural Networks), or the like.

The Convolutional Neural Network (CNN) is a deep feedforward neural network with the characteristics of local connection, weight sharing and the like. The convolutional neural network is composed of three parts: the first part is an input layer; the second part consists of a combination of n convolutional layers and a pooling layer (also called hidden layer and hidden layer); the third part is composed of a fully-connected multi-layer perceptron classifier (also called fully-connected layer). The convolutional neural network includes a feature extractor consisting of convolutional layers and sub-sampling layers. In the convolutional layer of the convolutional neural network, one neuron is connected to only part of the neighbor neurons. In a convolutional layer of CNN, there are usually several feature planes (featuremaps), each of which is composed of some neurons arranged in a rectangle, and the neurons of the same feature plane share weights, which are convolution kernels. The convolution kernel is generally initialized in the form of a random decimal matrix, and the convolution kernel learns to obtain a reasonable weight in the training process of the network. Sharing weights (convolution kernels) brings the immediate benefit of reducing the connections between layers of the network, while reducing the risk of over-fitting. Sub-sampling, also called Pooling (Pooling), typically takes the form of both Mean sub-sampling (Mean Pooling) and maximum sub-sampling (Max Pooling). The sub-sampling Layer (Subsampling Layer), also called Pooling Layer (Pooling Layer), is used for feature selection and to reduce the number of features and thus parameters. Sub-sampling can be viewed as a special convolution process. Convolution and sub-sampling greatly simplify the complexity of the model and reduce the parameters of the model.

Referring to fig. 3 and 4 in conjunction with fig. 1 and 2, fig. 3 is a schematic diagram of a convolutional neural network structure for predicting structural performance of a solar cell in the present invention, and fig. 4 is a flowchart of an embodiment of a method for predicting structural performance of a solar cell in the present invention. In the example of fig. 3, a convolutional neural network algorithm is taken as an example for further explanation and explanation. The process steps of predicting the performance of the blue light multi-junction solar cell multi-quantum well structure by using the deep learning algorithm of the convolutional neural network are as follows.

Step S1: and collecting and extracting data of the input characteristic parameters and the corresponding output characteristic parameters of the multi-junction solar cell structure, and establishing a corresponding data set for the collected data.

Method for collecting and extracting input characteristic parameters of multi-junction solar cell structureIn the process, characteristic parameters in the multi-junction solar cell structure need to be selected. In the selection, the input characteristic parameters which mainly affect the predicted values of the output characteristic parameters in the multi-junction solar cell structure are selected for data acquisition and extraction or selection. Input characteristic parameters of the selected multijunction solar cell structure include, but are not limited to: the thickness of each layer in the solar cell structure, the stacking mode among the layers, the shape of each layer, the composition ratio of materials and the material components and the like. The output characteristic parameters of the selected multi-junction solar cell structure include, but are not limited to: short circuit current density (J)_sc) Open circuit voltage (V)_oc) And Fill Factor (FF).

Then, corresponding data sets are established for the selected input characteristic parameters, and corresponding data set parameters are designed. The data set may be divided into a raw data set and a test data set, and the data set is preprocessed. The data set parameters can represent complex structures inside the multi-junction solar cell, so that a large amount of data of the multi-junction solar cell structure can be collected and recorded. Therefore, the data of each multi-junction solar cell structure can be used as a sample, and the data of a plurality of multi-junction solar cell structures can be used as a sample set. Each sample or each set of samples may serve as an input layer in a neural network.

And step S2, preprocessing the data in the data set established in the step S1 to obtain a preprocessed original data set and a preprocessed test data set. The pretreatment method comprises the following steps:

(1) in the established data set, the input characteristic parameters of the multi-junction solar cell structure are selected according to known physical knowledge and correlation coefficients among the data.

(2) And carrying out normalized data processing on the selected feature data, wherein a specific calculation formula is as follows:

where μ is the mean of the samples and σ is the standard deviation of the samples. Normalizing the data may result in input data at eachThe mean value in dimension is 0, the standard deviation is 1, and the standard normal distribution is obeyed.

(3) And data recombination, namely recombining the sizes of the processed data and dividing the processed data into multiple batches.

It should be further noted that, since the input dimension required in the two-dimensional convolutional neural network is 4D (samples, rows, cols, channels), the original data in this example is an array read by txt. Therefore, the arrangement of the raw data needs to be adjusted to match the input size of the two-dimensional convolutional neural network.

Step S3: and (3) building a neural network model by adopting a convolutional neural network based on a machine learning algorithm.

Referring to fig. 3, the convolutional neural network model structure built by using the convolutional neural network in this example mainly includes an input layer, a plurality of convolutional layers, a plurality of full-link layers, and an output layer in sequence, and the layers are connected in sequence. The input layer may be used to input data or data sets of input characteristic parameters, such as a pre-processed raw data set, a pre-processed test data set, for a sample or a sample set of the aforementioned multi-junction solar cell structures.

In the example of the figure, two convolutional layers are connected after the input layer in the convolutional neural network model structure, and each convolutional layer contains an activation function.

The convolution layer performs corresponding convolution calculation on the convolution kernel and the feature map to extract the feature map, and the specific process of the convolution of the first convolution layer can be expressed as: x is the number of^(l)＝∑x^(l-1)*ω^(l)+b^(l)Where denotes the convolution calculation of the matrix, ω^(l)Representing the neuron weight of the l-th layer, b^(l)Representing the bias of the l-th layer. In general, if the input matrix size is ω, the convolution kernel size is k, the stride is k, and the number of zero padding layers is p, the calculation formula of the size of the feature map generated after convolution is:

in this example, zero padding is performed on the input feature map, so that the size of the feature map after convolution is unchanged.

The activation function in the convolutional layer is a linear rectification function (Relu). The mathematical formula for the linear rectification function is: and f (x) max (0, x), the nonlinear transformation of the feature map can be completed.

The second convolution layer further extracts features of the image after the activation function, and the image output by the second convolution layer is activated by a linear rectification function (Relu) in the activation layer and then is transmitted to the next part, such as a full connection layer.

The activation function in the convolutional layer is a linear rectification function (Relu). The first convolution layer performs feature extraction on an input image having a shape of (5, 6, 1), and outputs an image having a size of (5, 6, 16). The size of the convolution kernel in the first convolution layer is 3 x 3, the number of the convolution kernels is 16, and the fill pattern padding selects same. The second convolution layer performs feature extraction on the image subjected to the activation function, and outputs an image with a size of (5, 6, 32). The convolution kernel size in the second convolution layer is 3 x 3, the convolution kernel number is 32, and the filling pattern padding selects same. The output image in the second convolutional layer is passed to the next part, such as the fully-connected layer, after activation by a linear rectification function (Relu) in the active layer.

In the example of fig. 3, there are two fully-connected layers in the convolutional neural network model structure. The aforementioned image activated by the activation layer becomes (512) connected to the first fully-connected layer on which 128 neurons are arranged, after being subjected to a flattening operation (Flatten). Considering the fewer training samples used in this example, a discard (Dropout) strategy was employed to suppress or reduce overfitting, with 20% of neurons randomly inactivated per round of training. For the image that has undergone the activation function, the final prediction is done by means of the second fully-connected layer.

And step S4, setting network structure parameters of the constructed convolutional neural network model, and performing initialization training on the set network structure parameters to obtain an initialized convolutional neural network model.

The method for carrying out initialization training on the network structure parameters set in the convolutional neural network model comprises the following steps: the step size of the first convolution layer is set to 1, the number of output channels is 16, and the padding mode padding is set to same. The step size of the second convolutional layer is set to 1, the number of output channels is 32, and the padding mode padding is set to same. The weights in the first convolutional layer and the second convolutional layer are initialized to truncated normal distribution noise with mean 0 and standard deviation 0.1, and all biases in the network are initialized to a constant, which is 1. Setting a learning rate in a value interval according to the characteristics of the training samples, and determining the batch size (Batchsize) of the training samples; and (4) repeatedly training the convolutional neural network model according to the setting of the training samples, and determining the total times of repeated training, thereby finishing the initial training of the convolutional neural network model. Wherein, the learning rate is set to a value range of 0.00001-0.1, and the total number of times of repeated training is 100-500 rounds.

To further illustrate, in the example of the figure, the learning rate of the training samples is set to 0.0001. The batch size (Batchsize) of the training samples was set to 16, i.e., 16 pictures were fed into the convolutional neural network at each training and the average loss of all samples for the entire batch was calculated. The total number of training rounds is 300, and a constructed convolutional neural network model is primarily optimized by selecting a random gradient descent (SGD) algorithm.

And S5, training and optimizing the initialized convolutional neural network model by means of the preprocessed data set of the input characteristic parameters of the multi-junction solar cell structure in the step S2, obtaining and storing the network weight and the bias of the convolutional neural network model, and further obtaining a convolutional neural network prediction model. Wherein the data set is a preprocessed raw data set.

In machine learning, a loss function is used to measure the loss (gap) between a model output value and a target value. Based on this, in step S5, the loss function during the training of the convolutional neural network model is expressed by a mean square error to determine the quality of the training result of the convolutional neural network model. The mean square error formula is:

wherein Predict_i、Actual_iThe predicted value and the true value of the ith sample are respectively. The more the MSE value obtained by calculation is close to 0, the better the training and optimizing results of the convolutional neural network model are, and the output result is accurateThe higher the degree. Therefore, when the performance of the multi-junction solar cell structure is predicted by using the convolutional neural network prediction model obtained after training and optimization, the accuracy of the obtained prediction value is higher.

Step S6: and inputting the preprocessed test data set in the input characteristic parameters of the multi-junction solar cell structure to be predicted into the convolutional neural network prediction model as an input layer, so as to output the predicted value of the output characteristic parameters of the multi-junction solar cell structure to be predicted. The predicted value of the output characteristic parameter of the multi-junction solar cell structure to be predicted includes, but is not limited to, the short-circuit current density (J) of the multi-junction solar cell structure_sc) Open circuit voltage (V)_oc) And Fill Factor (FF).

Example two: the performance of the multijunction solar cell structure is predicted using Support Vector Regression (SVR).

In the process of collecting and extracting the input characteristic parameters of the multi-junction solar cell structure, the characteristic parameters in the multi-junction solar cell structure need to be selected. In the selection, the input characteristic parameters which mainly affect the predicted values of the output characteristic parameters in the multi-junction solar cell structure are selected for data acquisition and extraction or selection. Input characteristic parameters of the selected multijunction solar cell structure include, but are not limited to: the thickness of each layer in the solar cell structure, the stacking mode among the layers, the shape of each layer and the composition ratio of the materials to the material components. The output characteristic parameters of the selected multi-junction solar cell structure include, but are not limited to: short circuit current density (J)_sc) Open circuit voltage (V)_oc) And Fill Factor (FF).

where μ is the mean of the samples and σ is the standard deviation of the samples. Normalizing the data may result in input data having a mean of 0 and a standard deviation of 1 in each dimension, subject to a standard normal distribution.

Step S3: and (4) building a support vector regression model by adopting a support vector machine algorithm based on a machine learning algorithm.

And step S4, setting network structure parameters of the built support vector regression model to obtain an initialized support vector regression model.

The method for setting the hyper-parameters in the support vector regression model comprises the following steps: the algorithm uses a 'linear' kernel, the times of a polynomial kernel function are 3, the tolerance factor is 0.001, and the penalty coefficient is 0.8.

And S5, training and optimizing the initialized support vector regression model by means of the preprocessed data set of the input characteristic parameters of the multi-junction solar cell structure in the step S2, obtaining and storing the network weight and the bias of the support vector regression model, and further obtaining a support vector regression prediction model. Wherein the data set is a preprocessed raw data set.

In this example, the root mean square error is used as a measure of the loss (gap) between the model output value and the target value. Based on this, in step S5, the loss function during training of the support vector regression model is expressed by a mean square error to determine the quality of the training result of the support vector regression model. The mean square error formula is:

wherein Predict_i、Actual_iThe predicted value and the true value of the ith sample are respectively. The more the value of the MSE obtained by calculation is close to 0, the better the training and optimizing result of the support vector regression model is, and the higher the accuracy of the output result is. Therefore, when the trained and optimized support vector regression prediction model is used for performance prediction of the multi-junction solar cell structure, the accuracy of the obtained prediction value is higher.

Step S6: and inputting the preprocessed test data set in the input characteristic parameters of the multi-junction solar cell structure to be predicted into the support vector regression prediction model as an input layer, so as to output the predicted value of the output characteristic parameters of the multi-junction solar cell structure to be predicted. The predicted value of the output characteristic parameter of the multi-junction solar cell structure to be predicted includes, but is not limited to, the short-circuit current density (J) of the multi-junction solar cell structure_sc) Open circuit voltage (V)_oc) And Fill Factor (FF).

Case three: the performance of the multi-junction solar cell structure is predicted using a K-nearest neighbor algorithm (KNN).

In the process of collecting and extracting the input characteristic parameters of the multi-junction solar cell structure, the characteristic parameters in the multi-junction solar cell structure need to be selected. In the selection, the input characteristic parameters which mainly affect the predicted values of the output characteristic parameters in the multi-junction solar cell structure are selected for data acquisition and extraction or selection. Input characteristic parameters of the selected multijunction solar cell structure include, but are not limited to: the thickness of each layer in the solar cell structure, the stacking mode among the layers, the shape of each layer, the composition ratio of materials and the material components and the like. The output characteristic parameters of the selected multi-junction solar cell structure include, but are not limited to: short circuit current density (J)_sc) Open circuit voltage (V)_oc) And Fill Factor (FF).

(2) Performing normalized data processing, specifically calculating, on the selected feature dataThe formula is as follows:

Step S3: and building a KNN model by adopting a K proximity algorithm based on a machine learning algorithm.

And step S4, setting network structure parameters of the constructed KNN model to obtain an initialized KNN model.

The method for setting the hyperparameter in the KNN model comprises the following steps: the number of neighbors used for the query is 3, and the weight average of all points in each neighborhood is the same.

And S5, training and optimizing the initialized KNN model by means of the preprocessed data set of the input characteristic parameters of the multi-junction solar cell structure in the step S2, obtaining and storing the network weight and the bias of the KNN model, and further obtaining a KNN prediction model. Wherein the data set is a preprocessed raw data set.

In this example, the root mean square error is used as a measure of the loss (gap) between the model output value and the target value. Based on this, in step S5, the loss function during KNN model training is expressed by a mean square error to clarify the quality of the KNN model training result. The mean square error formula is:

wherein Predict_i、Actual_iAre respectively the ith samplePredicted value and true value of the book. The calculated MSE value is closer to 0, which shows that the training and optimizing result of the KNN model is better, and the accuracy of the output result is higher. Therefore, when the performance of the multi-junction solar cell structure is predicted by using the KNN prediction model obtained after training and optimization, the accuracy of the obtained prediction value is higher.

Step S6: and inputting the preprocessed test data set in the input characteristic parameters of the multi-junction solar cell structure to be predicted into the KNN prediction model as an input layer, so as to output the predicted value of the output characteristic parameters of the multi-junction solar cell structure to be predicted. The predicted value of the output characteristic parameter of the multi-junction solar cell structure to be predicted includes, but is not limited to, the short-circuit current density (J) of the multi-junction solar cell structure_sc) Open circuit voltage (V)_oc) And Fill Factor (FF).

In summary, compared with the prior art, the convolutional neural network model, the multilayer sensor model and the KNN model provided by the invention can be used for designing the short-circuit current density (J) of the multi-junction solar cell structure in the overall structure design process of the multi-junction solar cell structure_sc) Open circuit voltage (V)_oc) And parameters such as a Fill Factor (FF) are predicted more accurately, so that the optimization of the multi-junction solar cell structure design scheme is guided better according to the prediction result, and the overall structure of the multi-junction solar cell with the luminous efficiency meeting the expectation is designed. In addition, the convolutional neural network prediction model provided by the invention can also predict the predicted values of the output parameters of a laser, a detector and the like.

In addition, it will be appreciated by those skilled in the art that, although there may be many problems with the prior art, each embodiment or aspect of the present invention may be improved only in one or several respects, without necessarily simultaneously solving all the technical problems listed in the prior art or in the background. It will be understood by those skilled in the art that nothing in a claim should be taken as a limitation on that claim.

Although terms such as new type of solar cell, multijunction solar cell, machine learning, neural network, etc. are used more often herein, the possibility of using other terms is not excluded. These terms are used merely to more conveniently describe and explain the nature of the present invention; they are to be construed as being without limitation to any additional limitations that may be imposed by the spirit of the present invention; the terms "first," "second," and the like in the description and in the claims, and in the drawings, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the structural performance of a solar cell is characterized by comprising the following steps: comprises the following steps:

collecting and extracting data of input characteristic parameters and corresponding output characteristic parameters of a solar cell structure, and dividing the data into an original data set and a predicted data set;

preprocessing the original data set and the prediction data set to obtain a preprocessed original data set and a preprocessed prediction data set;

constructing an initial model by applying a machine learning algorithm;

setting structural parameters of the initial model, and performing initialization training on the structural parameters to obtain an initialized model;

optimizing the initialized model, and training the initialized model by using the preprocessed original data set to obtain corresponding network weight and bias so as to obtain a prediction model;

and predicting, namely inputting the preprocessed test data set in the input characteristic parameters of the solar cell structure to be predicted into the prediction model, and further obtaining the predicted value of the output characteristic parameters of the solar cell structure to be predicted.

2. The method for predicting the structural performance of a solar cell according to claim 1, wherein: the machine learning algorithm is at least one of a deep learning algorithm, a multilayer perceptron, a decision tree, a linear regression, a gradient lifting regression and a K nearest neighbor algorithm.

3. The method for predicting the structural performance of a solar cell according to claim 2, wherein: the deep learning algorithm is at least one of a convolutional neural network, self-coding and a deep confidence network.

4. The method for predicting the structural performance of a solar cell according to claim 1, wherein: the solar cell is of a multi-junction solar cell structure and comprises at least one bottom cell and a plurality of sub-cells, wherein the sub-cells are positioned above the bottom cell.

5. The method for predicting the structural performance of the solar cell according to claim 4, wherein: the bottom battery comprises a substrate, an emitting layer, a window layer and a tunneling junction along the stacking direction, and the plurality of sub-batteries are stacked above the tunneling junction of the bottom battery; each sub-cell comprises a back field layer, a base region, an emission layer, a window layer and a tunneling junction along the stacking direction; and the uppermost layer of the sub-battery positioned at the topmost layer or the uppermost layer is a contact layer.

6. The method for predicting the structural performance of a solar cell according to claim 1, wherein: the input characteristic parameters of the solar cell structure comprise the thickness of each layer in the solar cell structure, the stacking mode among the layers, the shape of each layer and the composition ratio of the composition materials to the material components; the corresponding output characteristic parameters include a short circuit current density, an open circuit voltage, and a fill factor of the solar cell structure.

7. The method for predicting the structural performance of a solar cell according to claim 1, wherein: the method of pre-processing the raw data set and the predicted data set comprises the steps of:

selecting characteristics, namely selecting input characteristic parameters of the solar cell structure according to known physical knowledge and the relation between data;

data processing, namely performing normalization processing on the selected feature data;

and data reorganization, namely reorganizing the size of the processed characteristic data.

8. The method for predicting the structural performance of a solar cell according to claim 7, wherein: after the selected feature data are normalized, the data mean value of the feature parameters is 0 and the standard deviation is 1.

9. The method for predicting the structural performance of a solar cell according to claim 1, wherein: in the step of optimizing the initialized model, a training result of the initialized model is judged by adopting a mean square error, wherein the mean square error formula is as follows:

10. The method for predicting the structural performance of a solar cell according to claim 1, wherein: the data of the input characteristic parameters and the corresponding output characteristic parameters of the solar cell structure can be screened and adjusted according to the type of the solar cell structure.