KR102478303B1 - Artificial intelligence model for optimization of thin film coating of perovskite solar cell and its driving method - Google Patents

Abstract

The present invention relates to a big data-based artificial intelligence model for optimization of thin film coating of a perovskite solar cell and a driving method thereof. A big data-based artificial intelligence model for optimization of thin film coating of a perovskite solar cell according to an embodiment of the present invention may include: a communication interface part that collects big data of a coating thickness according to variables affecting the thin film of the light absorption layer of a solar cell; and a control part that analyzes the collected big data using support vector machine regression to which supervised learning of machine learning is applied, predicts the thickness of the thin film of the light absorption layer, and controls the adjustment of the thickness of the thin film based on a predicted value according to the prediction.

Description

Artificial intelligence model for optimizing perovskite solar cell thin film coating and its driving method

The present invention relates to an artificial intelligence model for optimizing thin film coating of a perovskite solar cell and a method for driving the same. In other words, it is about a method of driving a perovskite solar cell thin film coating optimization artificial intelligence model that can optimize thin film coating based on the analysis result by applying a program to analyze it.

As interest in new and renewable energy is increasing in order to solve the problem of global energy depletion of fossil fuels such as oil, coal and natural gas, the 9th Basic Plan for Electricity Supply and Demand of the Ministry of Trade, Industry and Energy Looking at the share of installed capacity, it can be seen that the share of new and renewable energy has increased significantly, while the share of nuclear power and fossil fuels has drastically decreased. While various methods using solar energy sources, wind power, water power, marine energy, bioenergy, hydrogen energy, etc. are being researched as new renewable energy sources, about the part of converting light energy into electrical energy by using sunlight in solar energy Solar cells are the most widely used, and solar cells are classified into inorganic solar cells, organic solar cells, and organic-inorganic hybrid solar cells depending on the material.

Perovskite solar cell is a next-generation solar cell that replaces the existing silicon solar cell. It is a solar cell device that uses an organic-inorganic hybrid material with a perovskite structure as a photoactive layer. The skyte material was used for the first time, and in 2014, a team at the Korea Research Institute of Chemical Technology reported the fabrication of a module using the spin coating method. are growing fast

By the way, as a result of searching 1,631 papers at home and abroad for existing research related to perovskite solar cells, many research institutes are mainly conducting research on module size, but a thin film coating optimization model analyzed based on big data I can't find any studies using .

Korea Patent Registration No. 10-1998021 (2019.07.02) Korean Registered Patent Publication No. 10-2161421 (2020.09.24)

An embodiment of the present invention is, for example, a perovskite solar cell or a perovskite that can optimize thin film coating based on the analysis result by generating big data and applying an artificial intelligence model, that is, a program, in the manufacture of a photovoltaic device. The purpose is to provide an artificial intelligence model and its driving method for optimizing the thin film coating of Skye solar cells.

The perovskite solar cell thin film coating optimization artificial intelligence model according to an embodiment of the present invention includes a communication interface unit for collecting big data of the coating thickness according to factors affecting the thin film of the light absorption layer of the solar cell, and the above The collected big data is analyzed using support vector machine regression to which supervised learning of machine learning is applied to predict the thickness of the light absorbing layer thin film, and based on the predicted value according to the prediction, the thickness of the thin film It includes a control unit that controls to adjust the thickness.

The control unit may execute a regression analysis algorithm of a sigmoid kernel similar to a neural network using a nonlinear polynomial kernel or a sigmoid function that adds data to a nonlinear transformation as an analysis model of the support vector machine regression.

The control unit may form a light absorption layer thin film of a perovskite solar cell based on the predicted value.

The control unit controls a bed temperature (TMP), a coating speed (SPD), a nitrogen blowing (N ₂ blowing) interval (DIST), a nitrogen blowing height (HG), and a nitrogen blowing intensity (which affect the light absorption layer thin film). The thin film thickness data according to BAR) can be utilized.

The control unit may pre-process collected data related to the coating thickness according to the variable and perform a supervised learning operation of machine learning using the pre-processed data.

In addition, the driving method of the perovskite solar cell thin film coating optimization artificial intelligence model according to an embodiment of the present invention, the big data of the coating thickness according to the variable (factor) that the communication interface affects the thin film of the light absorption layer of the solar cell Collecting, and a control unit analyzes the collected big data using support vector machine regression to which supervised learning of machine learning is applied to predict the thickness of the light absorbing layer thin film, and based on the predicted value according to the prediction, the thickness of the thin film and controlling to adjust the thickness.

According to an embodiment of the present invention, for example, when forming a light absorption layer thin film of a perovskite solar cell, the performance of the device can be improved by forming the thin film using big data of the thickness of the thin film according to variables affecting the formation of the thin film. There will be.

In addition, the embodiment of the present invention is a support vector machine regression model (or regression algorithm) that applies (or is based on) supervised learning of machine learning when predicting the thickness of a thin film through analysis of big data. When using a Gaussian kernel, the prediction accuracy is However, when tuning is applied to parameters, prediction accuracy can be increased by using an algorithm such as a linear kernel, a nonlinear polynomial kernel, or a sigmoid kernel function.

1 is a diagram illustrating the detailed structure of an artificial intelligence model for optimizing thin film coating of a perovskite solar cell according to an embodiment of the present invention;
2 is a diagram for explaining a 3D slot-die coating method;
3 is a diagram for explaining correlation by feature);
4 is a diagram for explaining support vector regression;
5 shows plots of linearity and RBF;
6 shows plots of polynomials and sigmoids;
7 shows plots of adjusted linearity and RBF;
8 shows a plot of the adjusted polynomial and sigmoid, and
9 is a flowchart illustrating a driving process of an artificial intelligence model for optimizing the coating of a thin film of a perovskite solar cell of FIG. 1 .

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

1 is a diagram illustrating the detailed structure of an artificial intelligence model for optimizing thin film coating of a perovskite solar cell according to an embodiment of the present invention.

As shown in FIG. 1, the perovskite solar cell thin film coating optimization artificial intelligence model 100 according to an embodiment of the present invention includes a communication interface unit 110, a control unit 120 and a thin film coating optimization unit 130 and part or all of the storage unit 140 .

Here, "including some or all" means that some components such as the storage unit 140 are omitted so that the perovskite solar cell thin film coating optimization artificial intelligence model 100 is configured, or the thin film coating optimization unit 130 ), which means that some components, such as the control unit 120, can be integrated into other components such as the control unit 120, and will be described as including all of them to help a sufficient understanding of the invention.

Perovskite solar cell thin film coating optimization artificial intelligence model 100 according to an embodiment of the present invention is a process equipment for optimizing thin film coating based on big data, and is a coating device or an optimal value for optimizing thin film coating (eg : It may be composed of factory equipment or a computer that provides a coating device with a thickness prediction value) so that the coating is performed.

Similar to the manufacture of a general semiconductor device, it can be seen that the optical device of the perovskite solar cell according to the embodiment of the present invention is also formed by a plurality of process equipment provided in a clean room. Of course, a plurality of chambers may be provided inside the process equipment to perform each process to form an optical device on a wafer, quartz substrate, or glass substrate. Conventionally, a perovskite solar cell may basically consist of a structure of a back electrode, a hole conducting layer, a light absorbing layer, an electron conducting layer, a front transparent electrode, and a glass substrate, and sunlight incident from the front is absorbed by the light absorbing layer, Electron-hole pairs are formed from the light absorption layer to generate electricity.

In addition, in the case of Hanyang Solar Energy in Korea, it includes a front electrode, an electron hole layer, a perovskite light absorbing layer, a hole conducting layer, and a back electrode formed on a substrate, and the perovskite light absorbing layer is A solar cell with improved probability of moving to the front side without recombination by receiving an electric field as much as the energy band gap is bent, including the layer section, the back layer section, and the energy band gap gradient layer section disposed between the front layer section and the back layer section are presenting A perovskite solar cell with improved electron collection efficiency absorbs high-energy sunlight well in the front layer section of the light absorption layer, and perovskite light that can efficiently move long distances in the back section of the light absorption layer. A solar cell including an absorber layer has been implemented.

The communication interface unit 110 of FIG. 1 collects data related to variables affecting the formation of the light absorption layer thin film under the control of the control unit 120 . Of course, in order to collect such data, a process equipment manager or a process equipment developer may precede setting work on what type of data to collect. Therefore, the communication interface unit 110 can collect data related to thin film formation or operation data of the corresponding equipment in association with various process equipment in the vicinity. Here, the operation data includes data related to an operation state when a process is performed in the surrounding process equipment. For example, variables affecting the formation of a light absorption layer thin film of a solar cell according to an embodiment of the present invention include bed temperature, coating speed, N ₂ blowing interval, N ₂ bowing height, and N ₂ blowing intensity. can include

The control unit 120 is in charge of overall control operations of the communication interface unit 110 of FIG. 1 , the thin film coating optimization unit 130 and the storage unit 140 . Representatively, the control unit 120 may store data collected through the communication interface unit 110 in the storage unit 140 and then provide the corresponding data to the thin film coating optimization unit 130 for big data analysis. In addition, the control unit 120 receives the big data analysis results of the thin film coating optimization unit 130, and based on this, in the case of coating equipment, directly proceeds with the thin film coating process, or provides the data to a separate coating equipment connected to it. It is possible to control the communication interface unit 110 to do so. Here, the big data analysis result can be a predicted value for thin film coating. This predicted value can be obtained by running an artificial intelligence deep learning program. Estimation accuracy can be increased by using a regression analysis model (or algorithm) of a machine's nonlinear polynomial kernel or sigmoid kernel.

The thin film coating optimizer 130 may perform a preprocessing operation (eg, a data selection operation for supervised learning) on the collected data, and may perform a big data related to variables affecting the formation of the light absorbing layer thin film. The thickness of the thin film is predicted using supervised learning-based support vector machine regression of the data. Then, a predicted value can be calculated. Here, the regression model for support vector machine regression may be an artificial intelligence-based deep learning program, and in an embodiment of the present invention, a nonlinear polynomial kernel that adds data to a nonlinear transformation and a sigmoid function are used A regression algorithm with a sigmoid kernel that is a bit like a neural network can be used. Of course, the regression model for support vector machine regression may further include a linear kernel that does not return data, a Gaussian (R.B.F, Radial Basis Function) kernel that applies to various shapes, but can be used immediately, but estimation, or prediction In order to further increase the accuracy of , it may not be suitable as a regression model according to an embodiment of the present invention. This, of course, can be seen as determined by the experiment, and the details will be discussed later. However, in an embodiment of the present invention, the type of regression model will not be particularly limited. In data analysis using artificial intelligence, supervised learning means using learning data that includes correct answers. Also, support vector machine regression is used for both data classification and regression.

Machine learning can be defined as a process of forming and improving a system by learning based on a data set input for a specific purpose. It is classified into supervised learning, which learns, and unsupervised learning, which creates clusters or extracts features based on the distribution of a data set without learning a target value.

Supervised learning performs learning on the tendency and distribution of target data by mapping input data to target values, builds a prediction system, and is further divided into classification and regression depending on the type of target value used. Unsupervised learning is a method in which the target data is not included in the learning data, and learning is performed by creating clusters using input data or analyzing relationships according to data distribution or characteristics. General regression analysis uses multiple regression to estimate the presence or absence of a significant relationship between a dependent variable and one or more independent variables and the degree of influence of several variables on the dependent variable, but Support Vector Machine Regression (SVR) In addition, a loss function is introduced to reduce the difference between the actual value and the estimated value, thereby providing excellent prediction performance.

In an embodiment of the present invention, it is preferable to apply a supervised learning method.

The storage unit 140 temporarily stores various types of data processed under the control of the control unit 120 . In the embodiment of the present invention, it can be seen that data related to the formation of the light absorption layer thin film of the perovskite solar cell is stored in the storage unit 140, and typically data such as bed temperature and coating speed can be stored.

Despite the above, the communication interface unit 110, the control unit 120, the thin film coating optimization unit 130, and the storage unit 140 of FIG. 1 can perform various operations, and other details will be discussed later. Since it contains them, we want to replace them with those contents.

The communication interface unit 110, the control unit 120, the thin film coating optimization unit 130, and the storage unit 140 of FIG. 1 according to an embodiment of the present invention are composed of hardware modules physically separated from each other, but each module Software for performing the above operations may be stored therein and executed. However, since the corresponding software is a set of software modules, and each module can be formed of hardware, it will not be particularly limited to the configuration of software and hardware. For example, the storage unit 140 may be hardware storage or memory. However, since it is possible to store information in a software manner (repository), the above content will not be particularly limited.

Meanwhile, as another embodiment of the present invention, the control unit 120 may include a CPU and a memory, and may be formed as a single chip. The CPU includes a control circuit, an arithmetic unit (ALU), a command interpreter, and a registry, and the memory may include RAM. The control circuit performs control operations, the operation unit performs operation of binary bit information, and the instruction interpretation unit performs an operation of converting high-level language into machine language and machine language into high-level language, including an interpreter or compiler. , the registry may be involved in software data storage. According to the above configuration, for example, at the beginning of the operation of the perovskite solar cell thin film coating optimization artificial intelligence model 100, the program stored in the thin film coating optimization unit 130 is copied and loaded into memory, that is, RAM. Then, by executing it, the data operation processing speed can be rapidly increased. In the case of a deep learning model, it can be loaded into GPU memory rather than RAM, and can be executed by accelerating the execution speed using the GPU.

Next, a research (or experiment) process according to an embodiment of the present invention will be described.

FIG. 2 is a diagram for explaining a 3D slot-die coating method, FIG. 3 is a diagram for explaining correlation by feature, FIG. 4 is a diagram for explaining support vector regression, Figure 5 shows plots of linear and R.B.F, Figure 6 shows plots of polynomials and sigmoids, Figure 7 shows plots of tuned (or tuned) linear and R.B.F, and FIG. 8 shows a plot of the adjusted polynomial and sigmoid.

Looking first at the data collection and pre-processing process, the original data used in the embodiment of the present invention is the data tested by the 3D slot-die method as shown in FIG. 2 at Hanyang Solar Energy from August 2019 to December 2021, that is, properties After collecting the coating thickness (THK) according to this coating speed (SPD), N2 blowing interval (DIST), N2 blowing intensity (BAR), N2 blowing height (HG), and bed temperature (TMP), consult with experts to determine the ideal 1,576 data were obtained by preprocessing values or missing values, and the description of data properties including derived variables is shown in <Table 1>.

The range of each independent variable is 172 ~ 188mm/min for SPD, 22 ~ 28mm for DIST, 1.2 ~ 1.8bar for BAR, 8 ~ 12mm for HG, and 56 ~ 64℃ for TMP, and the dependent variable THK according to the change of independent variables was measured from 472 to 628 nm. When using these raw data, it was determined that the effect of the attribute could not be equally evaluated from the slope determined in the regression analysis, so the data was converted to a minimum value of 0, a maximum value of 1, and other values between 0 and 1 so that each attribute had a slope It was normalized so that the effect on the effect could be evaluated equally. In addition, the descriptive statistics for the main variables are as shown in <Table 2>, except for the five variables RST_THK, XRD, UV_t, UV_a, and UV_lev, which are derived variables generated in the experiment, because their types are logical and factor, and are not suitable for regression analysis. , the correlation is shown in FIG. 3.

On the other hand, looking at support vector regression (SVR), a support vector machine is an algorithm that can be applied to classification and prediction (regression), and is largely supported by support vector classification (SVC) for classification and support vector regression for prediction. (Support Vector Regression, SVR). Basically, support vector classification obtains a hyperplane that maximizes the margin in the input space of multidimensional data, and uses this hyperplane as a decision function to predict which group the unknown data will belong to. If the difference between the value and the predicted value is within ε, there is no penalty point, and a loss function with a penalty point for an error greater than ε is used and can be expressed as in Equation 1.

로 하면

로 표현할 수 있다. 이때 변환 함수끼리 스칼라 곱인

들을 모아놓은 집합을 커널(kernel)이라고 한다.where u is the difference between the actual value and the predicted value. The above loss function assumes that there is noise in the data by finding a regression line with a small difference between the actual value and the predicted value by reducing the size of the regression coefficient, but it does not aim to perfectly estimate the actual value with noise, so it has an appropriate range (2ε) Calculate in a way that allows the difference between the actual value and the predicted value within, learn to fit as many samples as possible inside the road within a limited margin error (sample outside the road) as shown in Figure 4, and the width of the road is the hyperparameter epsilon (epsilon). One of the advantages of this support vector regression is that it enables linear estimation and non-linear estimation by using the concept of a kernel.

If you do

can be expressed as At this time, the scalar product of the conversion functions

A collection of these is called a kernel.

Continuing to look at the prediction of the support vector regression (SVR) model, the independent variables for optimizing the thin film of the light absorption layer of a solar cell are coating speed (SPD), N2 blowing interval (DIST), N2 blowing intensity (BAR), N2 blowing height (HG), The bed temperature (TMP) was set, and the film thickness (THK) was set as the dependent variable. To learn the support vector regression (SVR) model, a prediction model was learned using 70% of the data as a training set, and the remaining 30% of the data was configured as a test set to verify the learned model. It was used to do this, and each kernel function and parameter are shown in <Table 3>.

The metric used to measure the accuracy of prediction of the regression model is Root Mean Squared Error (RMSE), which reduces sensitivity by applying a penalty for errors with large differences, and intuitively improves model performance by applying the same unit as the actual value. Mean Absolute Error (MAE), which makes comparison easy, and the coefficient of determination (R-squared, R ² ), which evaluates how well the model fits the training data. The root mean square error (RMSE) and mean absolute error (MAE), the closer the index value is to 0, the better the explanatory power of the estimation model, and the coefficient of determination (R ² ), the better the explanatory power of the estimation model, the closer to 1 is <Equation 2> to <Equation 4>.

= i번째 예측 데이터값)(n = number of data, yi = i-th actual data value,

= i th predicted data value)

Cost, which is the range within which error can be tolerated, gamma, which determines the distance and influence of one data sample, epsilon, which determines the width of the hyperplane, angle of the polynomial kernel function used only in nonlinear polynomial kernels When the option of the degree parameter is not specified, the default values are shown in <Table 4>.

<Table 5> shows the measurement results of RMSE, MAE, and R ² with the default parameters applied. As a result of comparing the difference in accuracy for each kernel function, the RMSE and MAE values of the Gaussian kernel were relatively the lowest, resulting in excellent predictive power. The coefficient of determination R ² was also the best at 99.77%, and the sigmoid kernel showed poor results in all indicators.

Looking at the graph to which the basic parameter values are applied, the linear kernel and the Gaussian (R.B.F, Radial Basis Function) kernel appear linear with a small error as shown in FIG. 5, and the nonlinear polynomial kernel and the sigmoid kernel As shown in FIG. 6, it can be seen that it appears linear with an error.

Finally, the support vector regression performance evaluation is as follows.

As shown in <Table 4>, it is difficult to determine the difference in accuracy for each kernel function based on the parameters provided by default, so to find the optimal hyperplane, the linear kernel parameters are set to cost (0.1, 1, 10), gamma to (0.02, 0.2, 2), epsilon is (0.01, 0.1, 1), nonlinear polynomial kernel parameters are cost (1, 5, 10), degree is (1, 3, 10) and gamma is (0.2, 2, 5 ), epsilon is (0.01, 0.1, 1), Gaussian kernel parameters are cost (0.1, 1, 10), gamma (0.02, 0.2, 2), and epsilon (0.01, 0.1, 1) , and the sigmoid kernel parameters change the parameters for each kernel by using cost as (0.1, 1, 10), gamma as (0.02, 0.2, 2), and epsilon as (0.01, 0.1, 1). was given, and the tuning results of parameters using 10-fold cross validation as a sampling method are shown in <Table 6>.

Comparing the basic parameter values and the tuned parameter values, it can be seen that the cost, which is the range that can tolerate errors, increases, and the gamma that exerts the influence of data samples and determines the distance and the epsilon that determines the width of the hyperplane decreases. .

는 가우시안 커널이 0.99733로 가장 좋았다. 그리고 결정계수 R

는 4가지 커널 모두 0.997 이상으로 회귀식의 적합도가 높은 것을 파악할 수 있다.As a result of comparing the difference in accuracy by kernel function by inputting the tuned parameter values into each kernel (Table 7), the RMSE was 0.00964, the Gaussian kernel had the least, and the MAE was 0.00271, the sigmoid kernel the least, and the coefficient of determination R

had the best Gaussian kernel with 0.99733. and the coefficient of determination R

It can be seen that the goodness of fit of the regression equation is higher than 0.997 for all four kernels.

Looking at the graph to which the tuned parameter values are applied, the linear kernel and the Gaussian kernel look linear with a small error, similar to the graph to which the basic parameters are applied, as shown in (Figure 7), and the nonlinear polynomial kernel and sigmoid kernel, as shown in (Figure 8). It can be seen that the error appears linearly with much improvement.

Looking at the graph to which the tuned parameter values are applied, the linear kernel and the Gaussian kernel look linear with small errors, similar to the graph to which the basic parameters are applied, as shown in FIG. It can be seen that it appears linearly.

In conclusion, in 'NEO 2018', a long-term analysis report on the future world power system, it was predicted that solar and wind power would rapidly increase to 50% of global power generation by '50, and the Ministry of Industry's 'Renewable Energy 3020 Implementation Plan ( 2017)' and '2019 Government R&D Investment Direction and Standards (2018)' of the Ministry of Science and ICT (Science and Technology Innovation Headquarters), photoactive It can be seen that the perovskite solar cell, which is light, flexible, and permeable due to the characteristics of the material, shows low cost and high efficiency, and can be applied to the solution process when manufacturing, is included in the development plan as one of the key technologies for next-generation renewable energy.

Therefore, research according to an embodiment of the present invention is a linear kernel of a support vector machine, a Gaussian kernel, a nonlinear polynomial kernel, a sigmoid kernel By comparing the estimation accuracy with the regression analysis model of , the following conclusions were obtained for each kernel function.

First, as a result of examining the estimation accuracy, the best results were obtained when using the Gaussian kernel before tuning the parameters (see <Table 5>) and after tuning (see <Table 7>) among the four kernels. In particular, the Gaussian kernel showed no difference in RMSE, MAE, and coefficient of determination R ² even when the parameters were tuned and applied.

Second, it was easy to find that there was a difference in the estimation accuracy before parameter tuning in the four kernels, but as the estimation accuracy improved after parameter tuning, the difference between each kernel was found to be very small (see <Table 7>). It was confirmed that tuning was necessary.

Third, in order to increase the accuracy of estimation, it was found that while increasing the cost, which is the range that can allow errors, the gamma that exerts the influence of the data sample and determines the distance and the epsilon that determines the width of the hyperplane become smaller.

Fourth, as a result of tuning and applying the parameter values, the estimation accuracy of the polynomial kernel and the sigmoid kernel was greatly improved.

9 is a flowchart illustrating a driving process of an artificial intelligence model for optimizing the coating of a thin film of a perovskite solar cell of FIG. 1 .

Referring to FIG. 9 together with FIG. 1 for convenience of description, the perovskite solar cell thin film coating optimization artificial intelligence model 100 according to an embodiment of the present invention affects, for example, the light absorbing layer thin film of the perovskite solar cell. Collect big data related to variables that affect it (S900). Here, the variables may include variables such as bed temperature, coating speed, N ₂ blowing interval, N ₂ blowing height, and N ₂ blowing intensity.

In addition, the perovskite solar cell thin film coating optimization artificial intelligence model 100 analyzes the collected big data using supervised learning-based (or applied, based) support vector machine regression of machine learning to predict the thickness of the thin film of the light absorption layer And, based on the predicted value, form or form a thin film of the light absorption layer of the solar cell (S910). For example, when the artificial intelligence model 100 for optimizing coating of a thin film of a perovskite solar cell is a coating device, a coating operation may be performed based on a predicted value. On the other hand, in the case of a computer device that analyzes data, the analysis result may be provided as a coating device.

Here, the regression model of support vector machine regression is a linear kernel that does not return data, a nonlinear polynomial kernel that adds data to a nonlinear transformation, a Gaussian kernel that is applied to various shapes, and a sigmoid function that is a bit similar to a neural network using a sigmoid function. It may include various types of functions or algorithms, such as a de-kernel, but in an embodiment of the present invention, it may be preferable to use a polynomial kernel or a sigmoid kernel in consideration of estimation accuracy and the like.

Above all, according to the results of the study, the Gaussian kernel showed the best results in terms of estimation accuracy when looking at the results before and after tuning. Although there is a difference in the estimation accuracy before tuning the parameters in the four kernels, the estimation accuracy improves after tuning. As a result, the difference between kernels can be very small. Therefore, in order to increase the accuracy of estimation, the parameter value can be tuned so that the cost, which is the range that can tolerate errors, is large, and the gamma that determines the distance and the epsilon that determines the width of the hyperplane are small while peddling the influence of data samples. .

In addition to the above, the perovskite solar cell thin film coating optimization artificial intelligence model 100 of FIG. 1 can perform various operations, and since other details have been sufficiently described above, they will be replaced with those contents.

On the other hand, even if all the components constituting the embodiment of the present invention are described as being combined or operated as one, the present invention is not necessarily limited to these embodiments. That is, within the scope of the object of the present invention, all of the components may be selectively combined with one or more to operate. In addition, although all of the components may be implemented as a single independent piece of hardware, some or all of the components are selectively combined to perform some or all of the combined functions in one or a plurality of pieces of hardware. It may be implemented as a computer program having. Codes and code segments constituting the computer program may be easily inferred by a person skilled in the art. Such a computer program may implement an embodiment of the present invention by being stored in a computer-readable non-transitory computer readable media, read and executed by a computer.

Here, the non-transitory readable recording medium means a medium that stores data semi-permanently and can be read by a device, not a medium that stores data for a short moment, such as a register, cache, or memory. . Specifically, the above-described programs may be provided by being stored in a non-transitory readable recording medium such as a CD, DVD, hard disk, Blu-ray disk, USB, memory card, or ROM.

Although the preferred embodiments of the present invention have been shown and described above, the present invention is not limited to the specific embodiments described above, and is common in the art to which the present invention pertains without departing from the gist of the present invention claimed in the claims. Of course, various modifications and implementations are possible by those with knowledge of, and these modifications should not be individually understood from the technical spirit or perspective of the present invention.

110: communication interface unit 120: control unit
130: thin film coating optimization unit 140: storage unit

Claims (6)

a communication interface unit that collects big data of coating thickness according to factors affecting the thin film of the light absorbing layer of the solar cell; and
The collected big data is analyzed using support vector machine regression to which supervised learning of machine learning is applied to predict the thickness of the light absorbing layer thin film, and based on the predicted value according to the prediction, the thin film A control unit for controlling to adjust the thickness of; including,
In the support vector regression, there is no penalty if the difference between the actual value and the predicted value is within ε. Errors greater than ε are penalized using a loss function,

(Where u is the difference between the actual value and the predicted value) satisfies the conditional expression,
The control unit forms a light absorbing layer thin film of a perovskite solar cell based on the predicted value,
The control unit controls a bed temperature (TMP), a coating speed (SPD), a nitrogen blowing (N ₂ blowing) interval (DIST), a nitrogen blowing height (HG), and a nitrogen blowing intensity (which affect the light absorption layer thin film). An artificial intelligence model for optimizing thin film coatings for perovskite solar cells using thin film thickness data according to BAR). According to claim 1,
The control unit uses a nonlinear polynomial kernel or a sigmoid function that adds data to a nonlinear transformation as an analysis model of the support vector machine regression, and performs a regression analysis algorithm of a sigmoid kernel similar to a neural network. Artificial intelligence model for optimizing thin-film coating of solar cells. delete delete According to claim 1,
The control unit pre-processes the data collected in relation to the coating thickness according to the variable and performs a supervised learning operation of machine learning using the pre-processed data. A perovskite solar cell thin film coating optimization artificial intelligence model. Collecting, by a communication interface unit, big data of the coating thickness according to a variable (factor) affecting the thin film of the light absorbing layer of the solar cell; and
The control unit analyzes the collected big data using support vector machine regression to which supervised learning of machine learning is applied, predicts the thickness of the light absorbing layer thin film, and controls the thickness of the thin film to be adjusted based on the predicted value according to the prediction. Including;
In the support vector regression, there is no penalty if the difference between the actual value and the predicted value is within ε. Errors greater than ε are penalized using a loss function,

(Where u is the difference between the actual value and the predicted value) satisfies the conditional expression,
The control step is
Forming a light absorption layer thin film of a perovskite solar cell based on the predicted value; and
Depending on the bed temperature (TMP), coating speed (SPD), nitrogen blowing (N ₂ blowing) interval (DIST), nitrogen blowing height (HG), and nitrogen blowing intensity (BAR) affecting the light absorption layer thin film Utilizing the thin film thickness data;
A driving method of an artificial intelligence model for optimizing thin film coatings of perovskite solar cells, including

Images