Method for acquiring optimal photoelectric efficiency of perovskite solar cell
By combining neural network model and sparrow search algorithm, the problem of cumbersome combination conditions and high cost of photoelectric conversion efficiency optimization of perovskite solar cells is solved, and the efficiency and automation photoelectric conversion efficiency optimization of perovskite solar cells is achieved, and the efficiency and automation photoelectric conversion efficiency optimization of perovskite solar cells is achieved, and the Other fields of materials and chemical engineering.
Patent Information
- Application Number
- CN202411099692.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-01
AI Technical Summary
The optimization of photoelectric conversion efficiency of perovskite solar cells faces huge problems of cumbersome combination conditions and high cost, which makes it difficult for researchers to explore all possible combinations within a limited time.
Using intelligent machine learning algorithms, by combining neural network models and sparrow search algorithms, we quickly find the best components and process conditions combinations of perovskite solar cells, achieving efficient, accurate and automated material selection and process optimization.
Quickly find the maximum photoelectric conversion efficiency of perovskite solar cells in a huge search space, reducing the number of experiments and costs, improving R&D efficiency, and suitable for other materials and chemical engineering fields.
Smart Images

Figure CN118888466A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of solar cells, and in particular to a method for obtaining the optimal photoelectric conversion efficiency of a perovskite solar cell. Background Art
[0002] Solar cells, which are devices that directly convert light energy into electrical energy through the photoelectric effect or photochemical effect, have the characteristics of being clean and renewable, and have attracted much attention in the industry in recent years. Among them, perovskite solar cells have become a research hotspot in the photovoltaic field due to their advantages such as high photoelectric conversion efficiency, low manufacturing cost and solution processing, and have been rapidly developed. In just over a decade, their photoelectric conversion efficiency has rapidly increased from the initial 3.8% to 26.1%, which is comparable to the photoelectric conversion efficiency of traditional silicon-based solar cells, showing great application and development potential.
[0003] Perovskite materials refer to materials with ABX 3 A class of compounds with a structure in which the A position is an organic cation or an inorganic cation (e.g. Cs + , FA + 、MA + etc.), the B position is a metal cation (e.g. Pb 2+ etc.), X is a halogen anion (e.g. I - Br - The A position can be a single cation or a mixed cation, such as Cs + , FA + with MA + When mixed, the sum of the contents of each component must be 1. Similarly, the X position can be a single halogen anion or a mixed halogen anion, but I - With Br - The sum of must be 1, that is, the overall structure is still ABX 3structure. Since A, B, and X can be different elements or multiple combinations, different components can be formed. At the same time, in the preparation process conditions of perovskite, the perovskite deposition process includes one-step method, two-step method, etc.; the deposition methods include spin coating, vacuum method, scraping method, etc.; and the materials of each functional layer except the absorption layer also have many options, for example: the substrate includes fluorine-doped tin oxide, indium tin oxide, etc., the electron transport layer includes tin dioxide, titanium dioxide, etc., the hole transport layer includes nickel oxide, self-assembled monolayer, etc., and the electrode includes silver, gold, etc.; the annealing temperature, annealing time, anti-solvent, etc. of the preparation process all depend on the components. The combination of functional layers or transport layer materials such as substrate, electron transport layer, hole transport layer, electrode, and the choice of preparation process conditions will directly affect the battery performance (i.e., photoelectric conversion efficiency). If the above conditions are arranged and combined one by one in order to find the optimal photoelectric conversion efficiency, the number of arrangements and combinations is very large and the combination conditions are complicated. This makes it impossible for researchers and engineers to conduct experimental exploration of all combinations within a limited time (for example, changing any parameter involves thousands or even tens of thousands of arrangements and combinations). It is also costly and easy to waste a lot of experimental resources. Summary of the invention
[0004] In view of the problems existing in the above prior art, the purpose of the present invention is to provide a method for obtaining the optimal photoelectric efficiency of a perovskite solar cell. The method uses an intelligent machine learning algorithm to quickly find the optimal combination of components and process conditions of the perovskite solar cell in a huge search space based on model prediction, thereby quickly finding the maximum photoelectric conversion efficiency of the perovskite solar cell and realizing efficient, accurate and automated perovskite material selection and process optimization.
[0005] The purpose of the present invention is achieved through the following technical solutions: A method for obtaining the optimal photoelectric efficiency of a perovskite solar cell comprises the following steps: Data collection: Find the correlation between the characteristics and target attributes in perovskite solar cells, collect relevant data and establish the connection between the characteristics and target attributes; the characteristics include the composition, process conditions, functional layer or transport layer of the perovskite material, etc., and the target attribute is the photoelectric conversion efficiency; Model building: Establish a combined neural network model to predict the photoelectric conversion efficiency under different characteristic conditions, and use the trained model as the fitness function; Acquisition of optimal photoelectric conversion efficiency: data initialization, setting the continuous feature range and number of iterations, and randomly generating a combination of continuous features within the set range as the initial population (each individual in the population represents a feature combination); using the sparrow search algorithm to update the position and using the fitness function to evaluate the fitness of each combination, retaining combinations with high fitness and eliminating combinations with low fitness, adjusting the retained combinations, generating new combinations, and continuously iterating and optimizing until the number of iterations is reached or the optimal solution is found, and completing the acquisition of the combination of components and process conditions corresponding to the optimal photoelectric efficiency.
[0006] Based on further optimization of the above scheme, the combined neural network model is obtained by combining the classification model and the regression model, specifically: First, standardize continuous characteristics such as components and process conditions: ; Where: Indicates i The original value of the feature, Indicates i The standardized value of the feature, , Respectively represent i The mean and standard deviation of each feature; Perform one-hot encoding on the classification features of the functional layer or transport layer (e.g., electron transport layer, hole transport layer, etc.) and convert it into a binary vector; Then, the one-hot encoded categorical feature matrix Xc (whose dimension is the number of samples multiplied by the number of classification features) is input into a multi-layer fully connected neural network, where the first layer of the multi-layer fully connected neural network is a hidden layer, and its neurons are h c1 , The activation function is ReLU:
[0007] The activation functions from the second layer to the final output layer of the multi-layer fully connected neural network are Sigmoid , the output vector of each classification model is obtained through a multi-layer fully connected neural network C (i.e. the degree of influence of each classification feature):
[0008] Where: Wc n Indicates n The weight matrix of the layer, bc n Indicates n The bias vector of the layer, h n-1Indicates (n-1) The output of the layer; Afterwards, the output of the classification model C The standardized values of continuous features after standardization Perform splicing to obtain the feature matrix F , and establish a regression model with seven fully connected layers; the input of the regression model is the feature matrix F , the first layer of the regression model is the hidden layer, and the number of neurons is h r1 , the activation function is ReLU:
[0009] The second to sixth layers of the regression model are all hidden layers, and the number of neurons is h rm , the activation function is ReLU:
[0010] Where: Wr m Indicates m The weight matrix of the layer, br m Indicates m The bias vector of the layer, h rm Indicates m The output of the layer; The seventh layer of the regression model is the output layer, the number of neurons is 1, and a linear activation function is used to obtain the predicted value Y of the photoelectric conversion efficiency:
[0011] Based on the further optimization of the above scheme, the specific steps for obtaining the optimal photoelectric conversion efficiency are as follows: The initial population size is selected as N , the maximum number of iterations is set to T , and divide the initial population into discoverers and joiners, where the proportion of discoverers is PD The proportion of participants is SD ; First, the randomly generated initial population individuals are substituted into the combined neural network model to obtain the fitness value (i.e., photoelectric conversion efficiency) of each individual. pFit , and according to the fitness values of all individuals pFit Arrange in ascending order from small to large, retain combinations with high fitness values, and eliminate combinations with low fitness values; For each finder, generate a random number R 2 , and set the random number threshold Rd ;like , the current location is considered safe and the location is updated:
[0012] Where: Indicates i The individual in j The position on a dimension (representing a component, process condition, etc.); t Indicates the current iteration number, a Represents a constant that controls the search range; And add random perturbations to the updated positions:
[0013] Where: rand 1 () Represents a random number; , Respectively represent j The upper and lower bounds of the dimension variable; represents the position after adding random perturbations; Adding random perturbations can avoid falling into the local optimum of the photoelectric conversion efficiency of perovskite solar cells, improve search efficiency, and enhance search accuracy; The population at the position after random disturbance is substituted into the combined neural network model to obtain the predicted value of photoelectric conversion efficiency corresponding to each individual in the population as the updated fitness value. ; like , the current position is considered to be a dangerous position and the position is updated:
[0014] Where: Q represents random numbers that follow a normal distribution; R a represents random numbers from a standard normal distribution; And add a wider range of random perturbations to the updated positions:
[0015] Where: rand 2 () Indicates generating a random number; represents the position after adding random perturbations; The population at the position after random disturbance is substituted into the combined neural network model to obtain the predicted value of photoelectric conversion efficiency corresponding to each individual in the population as the updated fitness value. ; After that, update the joiner position and find the current optimal solution X best , the index range of each joiner is: ;in, Indicates that between 1 and Randomly select a value in the range of
[0016] Where: Indicates that the worst solution under the current number of iterations is j The position on the dimension, A represents a random vector; Indicates that the optimal solution under the current number of iterations is j The position on the dimension, B Represents a random number; The population with the updated position of the joiner is substituted into the combined neural network model to obtain the predicted value of the photoelectric conversion efficiency corresponding to each individual in the population as the updated fitness value ; In each iteration, the best fitness value experienced by each individual in the population is recorded as its historical best fitness value. pX ; Compare the updated fitness values With the best historical fitness value pX ,like , then update , and update the historical best position of the individual accordingly; at the same time, if the optimal fitness value appears, update the global optimal fitness value (the global optimal fitness value is for the entire population, used to record the best fitness value found by all individuals in the population under the current number of iterations; that is, any individual finding the optimal fitness value will lead to the update of the global optimal fitness value) and the optimal solution X best .
[0017] Based on further optimization of the above scheme, the random number threshold .
[0018] Based on further optimization of the above solution, the value range of the constant a controlling the search range is [0,1].
[0019] Based on further optimization of the above scheme, the random number rand 1 The value range is [-1,1], the random number rand 2 The value range is [0,1].
[0020] Based on further optimization of the above scheme, the random vector AEach element of is randomly selected from {-1, 1}. B It follows a uniform distribution in [0, 1].
[0021] Based on the further optimization of the above scheme, in the process of obtaining the optimal photoelectric conversion efficiency, for each individual in the population, a method of boundary processing and classification variable updating is used to check whether the updated position of the individual exceeds the defined range; Among them, the boundary processing determines whether the updated position exceeds the preset range by setting the boundary function, specifically:
[0022] Where: Indicates i The variable in t+1 The updated value after iteration, Indicates i The variable in t The updated value after iteration; , Respectively represent the lower and upper limits of the ith variable; is the boundary function, which means x The value of is always limited to [ lb,ub ] within the scope; represents the position update amount obtained by the above method; The categorical variable updates are as follows:
[0023] Where: Indicates j The categorical variables are t+1 The updated value after iteration; It means randomly selecting an integer from it. N j Indicates j The total number of categories for a categorical variable (i.e., functional stratum or transport stratum, etc.); rand() Indicates generating a random number between [0,1]; P mutate Indicates the probability of a categorical variable mutation, usually a small value.
[0024] Based on further optimization of the above scheme, the P mutate The value range is (0,0.1].
[0025] The following are the technical effects of the technical solution of the present invention: This application constructs a combined neural network model to learn and train the existing experimental data, so as to obtain the complex relationship between material components, preparation process parameters, etc. and the performance of solar cell devices; at the same time, by introducing a specific sparrow search algorithm and combining it with the combined neural network model, on the basis of the model, the best combination of components and preparation process conditions can be quickly found in the huge search space, so as to maximize the photoelectric conversion efficiency of perovskite solar cells and realize efficient and automated perovskite material selection and solar cell device process optimization. The algorithm combination of this application has the advantages of high efficiency, fast speed, strong robustness, etc., and it is not only suitable for obtaining the optimal photoelectric efficiency of solar cell perovskite materials, but also can be effectively extended to other materials and chemical engineering fields (for example: finding the best catalyst components and preparation conditions to improve catalytic efficiency and high-efficiency catalyst optimization; screening and optimizing drug molecular structures to improve drug efficacy and reduce toxic side effects; finding the best alloy components and heat treatment processes to improve material strength, toughness and corrosion resistance of material performance optimization, etc.), with a wide range of applications, can effectively reduce the number of experiments, simplify material selection, simplify process optimization steps, thereby improving R&D efficiency, and effectively reducing unnecessary time and cost in the R&D process. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 Schematic diagram of the process of obtaining the optimal photoelectric efficiency of a perovskite solar cell in an embodiment of the present invention. DETAILED DESCRIPTION
[0027] The technical solutions in the embodiments of the present invention will be described clearly and completely below. In the following description, for the purpose of explanation rather than limitation, specific details such as specific system structures and technologies are proposed to provide a thorough understanding of the embodiments of the present invention.
[0028] Embodiment 1: A method for obtaining the optimal photoelectric efficiency of a perovskite solar cell comprises the following steps: Data collection: Find the correlation between the features and target attributes in perovskite solar cells, collect relevant data and establish the connection between the features and target attributes (the relevant data are the existing material components, process conditions, functional layers or transport layers, etc., and the photoelectric conversion efficiency obtained by the corresponding combination); the features include the components, process conditions, functional layers or transport layers of perovskite materials (wherein the components and process conditions of perovskite materials are continuous features, and the functional layers or transport layers are classified features), and the target attribute is the photoelectric conversion efficiency; Model building: Establish a combined neural network model to predict the photoelectric conversion efficiency under different characteristic conditions, and use the trained model as the fitness function; The combined neural network model is obtained by combining the classification model and the regression model, specifically: First, standardize continuous characteristics such as components and process conditions: ; Where: Indicates i The original value of the feature, Indicates i The standardized value of the feature, , Respectively represent i The mean and standard deviation of each feature; Perform one-hot encoding (using a conventional one-hot encoding method in the art) on the classification features of the functional layer or transport layer (for example, an electron transport layer, a hole transport layer, etc.) and convert it into a binary vector; Then, the one-hot encoded categorical feature matrix Xc (whose dimension is the number of samples multiplied by the number of classification features) is input into a multi-layer fully connected neural network, where the first layer of the multi-layer fully connected neural network is a hidden layer, and its neurons are h c1 , the activation function is ReLU:
[0029] The activation functions from the second layer to the final output layer of the multi-layer fully connected neural network are Sigmoid , the output vector of each classification model is obtained through a multi-layer fully connected neural network C (i.e. the degree of influence of each classification feature):
[0030] Where: Wc n Indicates n The weight matrix of the layer, bc n Indicates n The bias vector of the layer, h n-1 Indicates (n-1) The output of the layer; Afterwards, the output of the classification model C The standardized values of continuous features after standardization Perform splicing to obtain the feature matrix F(The conventional splicing method in this field can be used. If all features including continuous features such as content and process conditions, and classification features such as functional layer or transmission layer are directly combined to establish a neural network, due to the limitations of the application model and the optimal photoelectric conversion efficiency, some information will be lost, resulting in low prediction accuracy and large errors in the established network model; splicing the features output by the classification model with the standardized continuous features can effectively realize information fusion, so that the regression model can learn the relationship between features and targets more comprehensively and improve prediction accuracy), and establish a regression model containing seven fully connected layers; the input of the regression model is the feature matrix F , the first layer of the regression model is the hidden layer, and the number of neurons is h r1 , the activation function is ReLU:
[0031] The second to sixth layers of the regression model are all hidden layers, and the number of neurons is h rm , the activation function is ReLU:
[0032] Where: Wr m Indicates m The weight matrix of the layer, br m Indicates m The bias vector of the layer, h rm Indicates m The output of the layer; The seventh layer of the regression model is the output layer, the number of neurons is 1, and a linear activation function is used to obtain the predicted value Y of the photoelectric conversion efficiency:
[0033] After the combined neural network model is established, the existing experimental data are selected again to train and verify the model, and a neural network model with a small root mean square error and a high average fitting coefficient is obtained.
[0034] Optimal photoelectric conversion efficiency acquisition: Data initialization, setting the continuous feature range and number of iterations, and randomly generating a combination of continuous features within the set range as the initial population (each individual in the population represents a feature combination), the selected initial population size is N , the maximum number of iterations is set to T (Set the population size and the maximum number of iterations according to the actual situation and specific target attributes, etc.), and divide the initial population into discoverers and joiners, where the proportion of discoverers is PD The proportion of participants is SD(Set the ratio of discoverers to joiners according to actual conditions); at the same time, define the range of continuous features, for example: the content range of component 1 is [ M 1 ,M 2 ], the content range of component 2 is [ M 3 ,M 4 ], etc. The temperature range of process conditions is [ W 1 ,W 2 ], etc. (the content of components and the range of process conditions are obtained based on existing data or experimental data); The sparrow search algorithm is used to update the position and the fitness function is used to evaluate the fitness of each combination. The combinations with high fitness are retained and the combinations with low fitness are eliminated. The retained combinations are adjusted and new combinations are generated. The optimization is continuously iterated. Specifically: First, the randomly generated initial population individuals are substituted into the combined neural network model to obtain the fitness value (i.e., photoelectric conversion efficiency) of each individual. pFit , and according to the fitness values of all individuals pFit Arrange in ascending order from small to large, retain combinations with high fitness values, and eliminate combinations with low fitness values; For each finder, generate a random number R 2 , and set the random number threshold R d In this embodiment: random number threshold ;like , the current location is considered safe and the location is updated:
[0035] Where: Indicates i The individual in j The position on a dimension (representing a component, process condition, etc.); t Indicates the current iteration number, a represents a constant for controlling the search range. In this embodiment, a The value range of is [0,1]; And add random perturbations to the updated positions:
[0036] Where: rand 1 () means generating a random number between [-1,1]; , Respectively representj The upper and lower bounds of the dimension variable; represents the position after adding random perturbations; Adding random perturbations can avoid falling into the local optimum of the photoelectric conversion efficiency of perovskite solar cells, improve search efficiency, and enhance search accuracy; The population at the position after random disturbance is substituted into the combined neural network model to obtain the predicted value of photoelectric conversion efficiency corresponding to each individual in the population as the updated fitness value. ; like , the current position is considered to be a dangerous position and the position is updated:
[0037] Where: Q represents random numbers that follow a normal distribution; R a represents random numbers from a standard normal distribution; And add a wider range of random perturbations to the updated positions:
[0038] Where: rand 2 () means generating a random number between [0,1]; represents the position after adding random perturbations; The population at the position after random disturbance is substituted into the combined neural network model to obtain the predicted value of photoelectric conversion efficiency corresponding to each individual in the population as the updated fitness value. ; After that, update the joiner position and find the current optimal solution X best , the index range of each joiner is: ;in, Indicates that between 1 and Randomly select a value in the range of
[0039] Where: Indicates that the worst solution under the current number of iterations is j The position on the dimension, A represents a random vector, a random vector A Each element of is randomly selected from {-1, 1}; Indicates that the optimal solution under the current number of iterations is j The position on the dimension, B Represents a random number, a random number BIt obeys the uniform distribution of [0, 1]; The population with the updated position of the joiner is substituted into the combined neural network model to obtain the predicted value of the photoelectric conversion efficiency corresponding to each individual in the population as the updated fitness value ; In each iteration, the best fitness value experienced by each individual in the population is recorded as its historical best fitness value. pX ; Compare the updated fitness values With the best historical fitness value pX ,like , then update , and update the historical best position of the individual accordingly; at the same time, if the optimal fitness value appears, update the global optimal fitness value (the global optimal fitness value is for the entire population, used to record the best fitness value found by all individuals in the population under the current number of iterations; that is, any individual finding the optimal fitness value will lead to the update of the global optimal fitness value) and the optimal solution X best ; Until the number of iterations is reached or the optimal solution is found, the combination of components and preparation processes corresponding to the optimal photoelectric efficiency is obtained.
[0040] Embodiment 2: As a further optimization of the scheme of the present application, on the basis of the scheme of Example 1, in order to avoid the updated position exceeding the defined range, resulting in errors in the final result or search failure, in the process of obtaining the optimal photoelectric conversion efficiency, for each individual in the population, a method of boundary processing and classification variable updating is used to check whether the updated position of the individual exceeds the defined range; Among them, the boundary processing determines whether the updated position exceeds the preset range by setting the boundary function, specifically:
[0041] Where: Indicates i The variable in t+1 The updated value after iteration, Indicates i The variable in t The updated value after iteration; , Respectively represent the lower and upper limits of the ith variable; is the boundary function, which means x The value of is always limited to [ lb,ub ] within the scope; represents the position update amount obtained by the above method; The categorical variable updates are as follows:
[0042] Where: Indicates j The categorical variables are t+1 The updated value after iteration; It means randomly selecting an integer from it. N j Indicates j The total number of categories for a categorical variable (i.e., functional stratum or transport stratum, etc.); rand() Indicates generating a random number between [0,1]; P mutate Indicates the probability of a categorical variable mutation, usually a small value. In this example: P mutate The value of is 0.1.
Claims
1. A method for obtaining the optimal photoelectric efficiency of a perovskite solar cell, characterized in that: The following steps are involved: Data collection: Find the correlation between the characteristics and target attributes in perovskite solar cells, collect relevant data and establish the connection between the characteristics and target attributes; the characteristics include the composition, process conditions, functional layer or transport layer of the perovskite material, etc., and the target attribute is the photoelectric conversion efficiency; Model building: Establish a combined neural network model to predict the photoelectric conversion efficiency under different characteristic conditions, and use the trained model as the fitness function; Acquisition of optimal photoelectric conversion efficiency: data initialization, setting the continuous feature range and number of iterations, and randomly generating a combination of continuous features within the set range as the initial population; using the sparrow search algorithm to update the position and use the fitness function to evaluate the fitness of each combination, retaining combinations with high fitness and eliminating combinations with low fitness, adjusting the retained combinations, generating new combinations, and continuously iterating and optimizing until the number of iterations is reached or the optimal solution is found, completing the acquisition of the combination of material components and process conditions corresponding to the optimal photoelectric conversion efficiency.
2. The method for obtaining the optimal photoelectric efficiency of a perovskite solar cell according to claim 1, characterized in that: The combined neural network model is obtained by combining the classification model and the regression model, specifically: First, standardize continuous characteristics such as components and process conditions: ; Where: Indicates i The original value of the feature, Indicates i The standardized value of the feature, , Respectively represent i The mean and standard deviation of each feature; Perform one-hot encoding on the categorical features of the functional layer or the transmission layer and convert them into a binary vector; Then, the one-hot encoded categorical feature matrix XC Input into the multi-layer fully connected neural network, where the first layer of the multi-layer fully connected neural network is the hidden layer, and its neurons are h c1 , the activation function is ReLU: The activation functions from the second layer to the final output layer of the multi-layer fully connected neural network are Sigmoid , the output vector of each classification model is obtained through a multi-layer fully connected neural network C : Where: W n Indicates n The weight matrix of the layer, bc n Indicates n The bias vector of the layer, h n-1 Indicates (n-1) The output of the layer; Afterwards, the output of the classification model C The standardized values of continuous features after standardization Perform splicing to obtain the feature matrix F , and establish a regression model with seven fully connected layers; the input of the regression model is the feature matrix F , the first layer of the regression model is the hidden layer, and the number of neurons is h r1 , the activation function is ReLU: The second to sixth layers of the regression model are all hidden layers, and the number of neurons is h rm , the activation function is ReLU: Where: Wr m Indicates m The weight matrix of the layer, br m Indicates m The bias vector of the layer, h rm Indicates m The output of the layer; The seventh layer of the regression model is the output layer, the number of neurons is 1, and a linear activation function is used to obtain the predicted value Y of the photoelectric conversion efficiency:
3. A method for obtaining the optimal photoelectric efficiency of a perovskite solar cell according to claim 1 or 2, characterized in that: The specific steps of obtaining the optimal photoelectric conversion efficiency are: The initial population size is selected as N , the maximum number of iterations is set to T , and divide the initial population into discoverers and joiners, where the proportion of discoverers is PD The proportion of participants is SD ; First, the randomly generated initial population individuals are substituted into the combined neural network model to obtain the fitness value of each individual pFit , and according to the fitness values of all individuals pFit Arrange in ascending order from small to large, retain combinations with high fitness values, and eliminate combinations with low fitness values; For each finder, generate a random number R 2. Set the random number threshold R d ;like , the current location is considered safe and the location is updated: Where: Indicates i The individual in j Position on the dimension; t Indicates the current iteration number, a Represents a constant that controls the search range; And add random perturbations to the updated positions: Where: rand 1() means generating a random number; , Respectively represent j The upper and lower bounds of the dimension variable; represents the position after adding random perturbations; The population at the position after random disturbance is substituted into the combined neural network model to obtain the predicted value of photoelectric conversion efficiency corresponding to each individual in the population as the updated fitness value. ; like , the current position is considered to be a dangerous position and the position is updated: Where: Q represents random numbers that follow a normal distribution; R a represents random numbers from a standard normal distribution; And add a wider range of random perturbations to the updated positions: Where: rand 2() means generating a random number; represents the position after adding random perturbations; The population at the position after random disturbance is substituted into the combined neural network model to obtain the predicted value of photoelectric conversion efficiency corresponding to each individual in the population as the updated fitness value. ; After that, update the joiner position and find the current optimal solution X best , the index range of each joiner is: ;in, Indicates that between 1 and Randomly select a value in the range of Where: Indicates that the worst solution under the current number of iterations is j The position on the dimension, A represents a random vector; Indicates that the optimal solution under the current number of iterations is j The position on the dimension, B Represents a random number; The population with the updated position of the joiner is substituted into the combined neural network model to obtain the predicted value of the photoelectric conversion efficiency corresponding to each individual in the population as the updated fitness value ; In each iteration, the best fitness value experienced by each individual in the population is recorded as its historical best fitness value. pX ; Compare the updated fitness values With the best historical fitness value pX ,like , then update , and update the individual's historical best position accordingly; at the same time, if the optimal fitness value appears, update the global optimal fitness value and the optimal solution X best .
4. A method for obtaining the optimal photoelectric efficiency of a perovskite solar cell according to claim 1 or 3, characterized in that: The random number threshold .
5. A method for obtaining the optimal photoelectric efficiency of a perovskite solar cell according to claim 1 or 3, characterized in that: The value range of the constant a controlling the search range is [0,1].
6. A method for obtaining the optimal photoelectric efficiency of a perovskite solar cell according to claim 1 or 3, characterized in that: The random number rand The value range of 1 is [-1,1], and the random number rand The value range of 2 is [0,1].
7. A method for obtaining the optimal photoelectric efficiency of a perovskite solar cell according to claim 1 or 3, characterized in that: The random vector A Each element of is randomly selected from {-1, 1}. B It follows a uniform distribution in [0, 1].
Citation Information
Cited By
Capacity grading capacity prediction method of new energy battery
CN119716568A
Design method and system of full-perovskite laminated solar cell and solar cell
CN119808602A
