Mechanism-data hybrid driven photovoltaic cell model construction method

Abstract

The invention belongs to the technical field of photovoltaic cell model parameter identification, and discloses a mechanism-data hybrid driven photovoltaic cell model construction method. Establishing a mechanism model of the photovoltaic cell; determining an objective function used for solving mechanism model parameters; through an AAEMDE algorithm, solving mechanism model parameters under an experiment working condition; aiming at each mechanism model parameter, constructing a correction model based on a BP neural network; the input of the correction model is irradiation intensity and temperature, the output of the correction model is a correction value of each mechanism model parameter, and the mechanism model parameters are corrected. And designing a self-adaptive differential evolution algorithm with an accelerated development mechanism, and applying the self-adaptive differential evolution algorithm to the construction of the photovoltaic cell hybrid model. According to the algorithm, the optimizing precision and the convergence speed of the basic differential evolution algorithm are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic cell model parameter identification, and in particular to a mechanism-data hybrid driven photovoltaic cell model construction method. Background Art

[0002] Energy is the solid foundation of human social progress, and its importance is self-evident. At present, the traditional energy system is still dominated by fossil fuels such as coal, oil, and natural gas. However, the over-exploitation and utilization of these resources have had many adverse effects on the environment. Therefore, the development of green, clean and efficient renewable energy has become an important task to promote social progress. There are many types of renewable energy, including wind energy, solar energy, tidal energy, etc. Among them, solar energy is particularly prominent with its unique advantages. As a model of clean energy, solar energy is not only low-cost and easy to obtain, but also has extremely rich resource reserves. In recent years, the significant advantages of solar energy have become increasingly apparent and have received widespread attention. As an important way to apply solar energy, photovoltaic power generation systems have also ushered in a good trend of rapid development.

[0003] The efficient operation of photovoltaic power generation systems depends on accurately constructed photovoltaic models, and their core role in practical applications is obvious. The photovoltaic model is composed of multiple photovoltaic cell models connected in series and parallel. Therefore, the construction of the photovoltaic cell model becomes a key link in the entire photovoltaic system modeling process. So far, researchers have proposed a variety of photovoltaic cell mechanism models. However, the accuracy of these models is strictly limited by the accuracy of the model parameters. At present, the industry generally adopts parameter identification methods, which rely on experimental data under specific working conditions and use advanced optimization algorithms to achieve accurate identification of model parameters.

[0004] The accuracy of model parameters is restricted by environmental factors such as irradiation intensity and temperature. Therefore, the model parameters obtained by parameter identification methods are only applicable to experimental conditions. For photovoltaic cells or photovoltaic arrays operating outdoors, the environmental conditions they are in are complex and changeable, which directly leads to fluctuations in model parameters, resulting in a significant decrease in model accuracy, making it difficult to meet the requirements of practical applications. Summary of the invention

[0005] The present invention aims to integrate mechanism analysis and data-driven technology, and innovatively proposes a new method for constructing photovoltaic cell models. The core steps include: first, using a unique advanced intelligent optimization algorithm, an adaptive differential evolution with accelerated exploitation mechanism (AAEMDE) algorithm is developed to accurately identify model parameters under experimental conditions and use them to construct photovoltaic models; second, a data-driven model is constructed to accurately correct model parameter changes caused by light intensity fluctuations and temperature changes, thereby constructing a mechanism-data hybrid-driven photovoltaic cell model. This method ensures that the model can maintain excellent accuracy under various environmental conditions.

[0006] The technical solution of the present invention is as follows: a photovoltaic cell model construction method driven by mechanism-data hybrid, comprising the following steps:

[0007] S1: Establish the mechanism model of photovoltaic cells;

[0008] S2: Define the objective function used to solve the parameters of the mechanism model;

[0009] S3: Obtain the mechanism model parameters under experimental conditions through the AAEMDE algorithm;

[0010] S4: For each mechanism model parameter, a correction model based on BP neural network is constructed; the input of the correction model is the irradiation intensity and temperature, and the output is the correction value of each mechanism model parameter, so as to correct the mechanism model parameters.

[0011] The objective function fitness value is calculated as follows:

[0012]

[0013] In formula (2), M represents the total number of experimental measurement data; x represents the parameter vector to be determined; m represents the mth group of experimental measurement data; f(I L ,V L ,x) represents the error function.

[0014] The error function is expressed as follows:

[0015]

[0016] I ph Represents the photocurrent value flowing through the PN junction after irradiation; I sdi Indicates the reverse saturation current value of the diode; I d Indicates the current value passing through the diode; I shIndicates the current value flowing through the shunt resistor; R s Indicates the resistance value of parallel resistance; R sh Indicates the resistance value of the series resistor; n i Represents the ideality factor of the diode, which is used to quantify the degree to which the diode deviates from the ideal behavior characteristics in actual operation; in the single-diode model, double-diode model, and triple-diode model, the value of N is 1, 2, and 3, respectively.

[0017] The steps of the AAEMDE algorithm to obtain the mechanism model parameters are as follows:

[0018] S3.1: Set the population size N p , the maximum number of iterations T max , archive matrix M CR and M K , and the upper bounds U corresponding to the parameters of the photovoltaic cell model b and the lower bound L b Generate an initial population within the range; calculate the fitness value of the objective function, record the optimal solution, and start iterative optimization;

[0019] S3.2: Determine whether to adopt an accelerated development mechanism. The specific basis for judgment is: when θ≥10 -5 , it is considered that the current stage is in the early stage of exploration and there is no need to accelerate development, and step S3.3 is performed; when θ<10 -5 , it is considered that the current stage is in the late stage of exploration, and the accelerated development mechanism is adopted to proceed to step S3.7;

[0020] The mathematical expression of θ is as follows:

[0021] θ= |Π(x i, g+1 )-Π(x i, g )| (4)

[0022] In formula (4), Π(x i,g+1 ) and Π(x i,g ) represents the fitness function value of two adjacent iterations;

[0023] S3.3: Perform mutation operations according to the following formula to promote overall evolution:

[0024]

[0025] In formula (5), r1 and r2 are two mutually unequal random integers between [1, NP], and neither of them is equal to i; K i and F i are two scale factors, which are important control parameters that determine the degree of scaling of the differential vector. i =Ki ; is from the current population P g After sorting according to fitness, select the top 100p i % a random individual within the range; x r2,g Represents a population from archive A and current population P g A randomly selected individual from the union of i Represents the ratio of elite individuals, which ranges between [0,1] and is given by the following formula;

[0026] p i =rand [p min ,n] (6)

[0027] In formula (6), the exploration of the early stage p min =2 / NP and n=0.2;

[0028] S3.4: The CR of each generation of successfully evolved individuals i and K i Recorded separately to archive file S CR and S K Calculate S CR The weighted average mean WA (S CR ) and S K The weighted Lehmer mean WL (S K );S K The weighted Lehmer mean WL (S K )The formula is as follows:

[0029]

[0030] S3.5: During the iteration process, mean WA (S CR ) and mean WL (S K ) are stored in M CR and M K In the archive, the formula is as follows:

[0031]

[0032] In equations (8) and (9), k represents the continuously updated storage location, 1≤k≤H;

[0033] S3.6: Based on the index r randomly generated from [1,H] i , respectively, using the following equations (10) and (11) to generate CR i and K i ;

[0034] CR i =randn i (M CR,ri ,0.1) (10)

[0035] K i =randc i (M K,ri ,0.1) (11)

[0036] S3.7: In the accelerated development mechanism, by adjusting the scaling factor K i 、F i and the proportion of elite groups p i , so that the mutation vector is adapted to the step size close to the first 100p i % of individuals and develop around them; at this time, take F i =0.7*K i , reduce p min and n, so that the mutation vector converges to the dominant individual faster, taking p min =1 / NP and n=0.1; finally, the M of this generation K Increase by 0.1 to increase K i The step size can improve the convergence speed;

[0037] S3.8: Perform a crossover operation. Each individual generates its own test vector u through the crossover operation. i,g , which is derived from the target vector x i,g and mutation vector v i,g The cross combination of is as follows:

[0038]

[0039] In formula (12), u i,d,g 、v i,d,g and x i,d,g Respectively represent u i,g 、v i,g and x i,g The dth dimension of , d = 1, 2, ..., D; CR i Determines the generated test vector u i,g From x i,g and v i,g The proportion of components inherited in rand is an integer randomly selected from 1 to D, ensuring that the test vector u i,g At least one dimension comes from the mutation vector v i,g ;

[0040] S3.9: Use greedy selection operation to select u i,g and x i,gIndividuals with better fitness values ​​survive to the next generation; for the minimization problem, the formula is as follows:

[0041]

[0042] In formula (13), Π(u i,g ) and Π(x i,g ) represent u i,g and x i,g The fitness function value of

[0043] S3.10: Determine whether the maximum number of iterations has been reached. If the maximum number of iterations has been reached, the parameter values ​​of the photovoltaic cell mechanism model under the experimental conditions are output and recorded as If the maximum number of iterations is not reached, return to step S3.2 and repeat the above operation.

[0044] The settings in step 3.1 are as follows: the photocurrent value I ph The lower bound L b The value is 0A, the upper limit is U b The value is 1A; the reverse saturation current value I sdi The lower bound L b The value is 0μA, the upper limit U b The value is 1μA; the equivalent series resistance value R S Lower bound L b The value is 0Ω, the upper limit U b The value is 1Ω; the diode ideal factor n i The lower bound L b The value is 1, the upper bound U b The value is 2; the equivalent parallel resistance value R Sh Lower bound L b The value is 0Ω, the upper limit U b The value is 100Ω;

[0045] The correction model based on BP neural network adopts data-driven modeling technology and establishes a correction model according to the number of parameter vectors to be determined in the mechanism model; the correction model based on BP neural network takes the irradiation intensity and temperature as input, and the outputs are the correction values ​​of the parameters in the photovoltaic cell mechanism model, ΔI ph , ΔI sdi , ΔR s , ΔR sh , Δn i ; An indirect training strategy is used to convert the correction model training into a parameter identification problem, and the process data obtained during the actual power generation process are used to J represents the total number of process data pairs; R represents radiation intensity; T represents temperature, and the correction model is constructed without the need for target output data.

[0046] The specific description of the indirect training strategy is as follows:

[0047] The mathematical expression of the BP neural network correction model is as follows:

[0048]

[0049] In formulas (14)-(18), They represent the parameters to be determined of the correction model based on BP neural network, and are the corresponding connection weights and thresholds;

[0050] In order to identify the connection weights and thresholds in the correction model based on the BP neural network, the following objective function is adopted:

[0051]

[0052] In formula (19), J represents the total number of process data pairs; z represents the parameter vector in the correction model based on BP neural network to be determined. j represents the jth group of process data; Represents the error function, expressed as formula (20):

[0053]

[0054] Finally, using equation (20) as the fitness function, the AAEMDE algorithm is used to solve the parameter vector in the BP neural network correction model. This completes the construction of the correction model.

[0055] Beneficial effects of the present invention:

[0056] 1) This invention proposes an innovative photovoltaic cell model construction method that combines mechanism and data-driven technology. In this method, the data-driven model part can accurately compensate for the changes in the mechanism model parameters caused by light intensity fluctuations and temperature changes, ensuring that the model exhibits excellent accuracy performance under various environmental conditions. This breakthrough effectively overcomes the limitations of existing photovoltaic cell models that are only applicable to experimental conditions and difficult to meet practical application needs.

[0057] 2) The present invention also developed an advanced intelligent optimization algorithm, namely the adaptive differential evolution algorithm with accelerated development mechanism (AAEMDE), and used it in the construction of photovoltaic cell hybrid model. This algorithm significantly improves the optimization accuracy and convergence speed of the basic differential evolution algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the structure of a photovoltaic cell model driven by mechanism-data hybrid (the figure takes the photovoltaic cell double diode model as an example).

[0059] Figure 2 It is the main flow chart of the technical solution of the present invention.

[0060] Figure 3 It is the equivalent circuit of the single diode model of a photovoltaic cell.

[0061] Figure 4 It is the optimization flow chart of AAEMDE algorithm. DETAILED DESCRIPTION

[0062] This paper combines mechanism analysis with data-driven technology and proposes a novel method for building a photovoltaic cell model. The core content covers the following two aspects:

[0063] 1) The present invention proposes a mechanism-data hybrid driven photovoltaic cell model construction method, aiming to ensure that the model can show high accuracy under various environmental conditions. The core implementation path of this method is as follows: First, the unique advanced intelligent optimization algorithm is used to accurately extract the mechanism model parameters under experimental conditions; secondly, a BP neural network model is constructed to achieve accurate correction of the changes in the mechanism model parameters caused by light intensity fluctuations and temperature changes, thereby constructing a mechanism-data hybrid driven photovoltaic cell model. In the process of constructing the BP neural network correction model, an innovative indirect training strategy is also proposed to effectively address the problem of missing target output data.

[0064] 2) The present invention also develops an advanced intelligent optimization algorithm, namely the adaptive differential evolution algorithm with accelerated development mechanism (AAEMDE), and uses it in the construction of photovoltaic cell hybrid model to further improve the accuracy of the constructed model.

[0065] like Figure 2 As shown, the photovoltaic cell model construction method proposed by the present invention, taking a single diode model as an example, includes the following steps:

[0066] S1: Establish a mechanism model of photovoltaic cells.

[0067] S2: Define the objective function used to solve the parameters of the mechanism model.

[0068] S3: Use the AAEMDE algorithm to obtain the mechanism model parameters under experimental conditions.

[0069] S4: For each mechanism model parameter, a correction model (or compensation model) based on BP neural network is constructed. The input of the model is the radiation intensity and temperature, and the output is the correction value of each parameter.

[0070] In step S1, the most commonly used single diode model is used as an example, and its equivalent circuit is as follows: Figure 3It mainly consists of the following parts: (1) a photocurrent source, which depends on the properties of the semiconductor material, the irradiation intensity and the ambient temperature; (2) a diode connected in parallel with the current source, taking into account the physical effect of the PN junction; (3) a resistor (R s ) represents the internal resistance of the photovoltaic cell, including electrode contact resistance, electrode resistance, and line resistance; (4) shunt resistance (R sh ) represents the leakage current in the semiconductor.

[0071] Based on Kirchhoff's current law and Shockley's equation, the current-voltage (IV) characteristics of the equivalent circuit of a photovoltaic cell single diode model can be expressed as:

[0072]

[0073] In formula (1), I L and V L Represent the output current and output voltage of the photovoltaic cell respectively. ph Represents the photocurrent value flowing through the PN junction after irradiation; I d Indicates the current value passing through the diode; I sh Indicates the current value flowing through the shunt resistor; I sd Represents the reverse saturation current value of the diode; R s and R sh represents the parallel resistance and series resistance respectively; n represents the ideality factor of the diode, which is used to quantify the degree to which the diode deviates from the ideal behavior characteristics in actual operation. When n=1, it means that the state of the diode is very close to the ideal state; q is the charge of the electron, which is 1.6202177×10 -19 C; k represents the Boltzmann constant, which is 1.3806503×10 -23 J / K; T represents the working temperature of the photovoltaic cell; "exp" represents an exponential function based on the natural constant e.

[0074] In summary, the single diode model contains five parameters that need to be accurately identified, namely [I ph ,I sd ,R s ,R sh ,n].

[0075] In step S2, the root mean square error (RMSE) is used to measure the difference between the two sets of data. Specifically, the smaller the RMSE value, the higher the match between the model calculation data and the experimental measurement data. For the identification of the mechanism model parameters, the mathematical expression of its objective function is as follows:

[0076]

[0077] In formula (2), M represents the total number of experimental measurement data; x represents the parameter vector to be determined; m represents the mth group of experimental measurement data; f(I L ,V L ,x) represents the error function, which can be expressed by formula (3):

[0078]

[0079] The optimization process of the AAEMDE algorithm mentioned in step S3 is as follows: Figure 4 The steps to obtain the mechanism model parameters using this algorithm are as follows:

[0080] S3.1: Set the population size N p , the maximum number of iterations T max , Archive M CR and M K Optimize the algorithm parameters and set the upper bound U corresponding to each parameter of the photovoltaic cell model b and the lower bound L b The initialization population is generated within the range. Among them, the photocurrent I ph The lower bound L b The value is 0A, the upper limit is U b The value is 1A; the reverse saturation current I sd The lower bound L b The value is 0μA, the upper limit U b The value is 1μA; the equivalent series resistance R S Lower bound L b The value is 0Ω, the upper limit U b The value is 1Ω; the lower limit of the diode ideal factor n is L b The value is 1, the upper bound U b The value is 2; the equivalent parallel resistance R Sh Lower bound L b The value is 0Ω, the upper limit U b The value is 100Ω. Then the fitness value of the objective function is calculated, the optimal solution is recorded, and the iterative optimization is started.

[0081] S3.2: Determine whether to adopt an accelerated development mechanism. The specific basis for judgment is: if θ≥10 -5 , then it is considered that the current stage is in the early stage of exploration and there is no need to accelerate development, and step S3.3 is performed; if θ<10 -5 , it is considered that the current state is in the late exploration stage, and the accelerated development mechanism is adopted to proceed to step S3.7. The mathematical expression of θ is as follows:

[0082] θ=|Π(x i,g+1 )-Π(x i,g )| (4)

[0083] In formula (4), Π(x i,g+1 ) and Π(x i,g ) represents the fitness function value.

[0084] S3.3: According to the following formula, perform mutation operations on the algorithm to promote overall evolution:

[0085]

[0086] In formula (5), r1 and r2 are two random integers between [1, NP] that are not equal to each other, and they are not equal to i. i and F i are two scale factors, which are important control parameters that determine the degree of scaling of the differential vector. i =K i . is from the current population P g After sorting according to fitness, select the top 100p i % a random individual within the range; x r2,g Represents a population from archive A and current population P g The parameter p is a randomly selected individual from the union of i Represents the ratio of elite individuals, which ranges between [0,1] and is given by the following formula.

[0087] p i =rand[p min ,n] (6)

[0088] In formula (6), the exploration of the early stage p min =2 / NP and n=0.2.

[0089] S3.4: The CR of each generation of successfully evolved individuals i and K i Record to archive S CR and S K Calculate S CR The weighted average mean WA (S CR ) and S K The weighted Lehmer mean WL (S K ). K The weighted Lehmer mean WL (S K )The formula is as follows:

[0090]

[0091] S3.5: During the iteration process, the mean WA (S CR ) and mean WL (S K ) are stored in M CR and M K In the archive, the formula is as follows:

[0092]

[0093] In equations (8) and (9), k (1≤k≤H) represents a storage location that is continuously updated.

[0094] S3.6: Based on the index r randomly generated from [1,H] i , respectively, using the following equations (10) and (11) to generate CR i and K i .

[0095] CR i =randn i (M CR,ri ,0.1) (10)

[0096] K i =randc i (M K,ri ,0.1) (11)

[0097] S3.7: In the accelerated development mechanism, by adjusting the scaling factor K i 、F i and the proportion of elite groups p i , so that the mutation vector approaches the first 100p with a suitable step size i % of individuals and develop around them. At this time, take F i =0.7*K i , the larger K i Control the convergence direction and speed up the convergence speed; smaller F i The difference vector is shrunk to reduce population diversity. Secondly, p is reduced min and n, so that the mutation vector converges to the dominant individual faster, taking p min =1 / NP and n=0.1. Finally, the M K,g Increase by 0.1 to increase K i The step size can improve the convergence speed.

[0098] S3.8: Perform a crossover operation. Each individual generates its own test vector u through the crossover operation. i,g , which is derived from the target vector x i,g and mutation vector x i,g The cross combination of is as follows:

[0099]

[0100] In formula (12), u i,d,g 、v i,d,g and x i,d,g Respectively represent u i,g 、v i,g and x i,g The dth dimension of , d = 1, 2, ..., D. CR i Determines the generated test vector u i,g From x i,g and v i,g The proportion of components inherited in the rand is an integer randomly selected from 1 to D, which ensures that the test vector u i,g At least one dimension comes from the mutation vector v i,g .

[0101] S3.9: Use greedy selection operation to select u i,g and x i,g Individuals with better fitness values ​​survive to the next generation. For the minimization problem, the formula is as follows:

[0102]

[0103] In formula (13), Π(u i,g ) and Π(x i,g ) represent u i,g and x i,g The fitness function value of .

[0104] S3.10: Determine whether the maximum number of iterations has been reached. If so, output the five parameter values ​​of the photovoltaic cell mechanism model under the experimental conditions and record them as If the maximum number of iterations has not been reached, return to step S3.2 and repeat the above operation.

[0105] In step S4, five BP neural network-based photovoltaic cell mechanism model parameter correction models are constructed using data-driven modeling technology. These BP neural network models all take environmental factors (irradiation intensity, temperature) as input, and the outputs are the correction values ​​of the five parameters in the photovoltaic cell mechanism model, namely, ΔI ph , ΔI sd , ΔR s , ΔR sh, Δn. It is worth noting that the challenge of model construction is that the correction values ​​of these five parameters are unknown during the construction process. This unknown directly leads to the lack of target output data required to build the correction model, which in turn makes it impossible for traditional training algorithms to be directly applied to the construction of BP neural network correction models. To solve the above problems, this patent innovatively proposes an indirect training strategy. This strategy converts neural network training into a parameter identification problem. It only needs to use the process data obtained in the actual power generation process to (where j = 1, 2, ..., J, J represents the total number of process data pairs; R represents the radiation intensity; T represents the temperature), the correction model can be constructed without the need for target output data. The specific description of this strategy is as follows.

[0106] The mathematical expressions of the five BP neural network correction models can be expressed as:

[0107]

[0108] In formulas (14)-(18), They represent the parameters to be determined of the five BP neural network correction models, namely the corresponding connection weights and thresholds.

[0109] In order to identify the connection weights and thresholds in the five BP neural network correction models, this patent adopts the following objective function:

[0110]

[0111] In formula (19), J represents the total number of process data pairs; z represents the parameter vectors in the five BP neural network correction models to be determined. j represents the jth group of process data; represents the error function, which can be expressed by formula (20):

[0112]

[0113] Finally, using equation (20) as the fitness function, the AAEMDE algorithm is used to solve the parameter vectors in the five BP neural network correction models. The construction of the correction model is thus completed. The solution process is similar to the process of obtaining the photovoltaic cell mechanism model parameters in step S3, but there are the following key differences: First, the decision variables are transformed from the original five mechanism model parameters to the parameter vectors in the five BP neural network correction models. Secondly, the expression of the fitness function is adjusted from formula (2) to formula (20). The specific details are not repeated here. In addition, it should be pointed out that in the process of constructing the correction model, the data set is divided into two major components: the training set and the validation set, of which the training set accounts for 80% and the validation set accounts for the remaining 20%. The function of the validation set is to determine the structure of the BP neural network correction model.

Claims

1. A photovoltaic cell model construction method driven by mechanism-data hybrid, characterized in that: The following steps are involved: S1: Establish the mechanism model of photovoltaic cells; S2: Define the objective function used to solve the parameters of the mechanism model; S3: Obtain the mechanism model parameters under experimental conditions through the AAEMDE algorithm; S4: For each mechanism model parameter, a correction model based on BP neural network is constructed; the input of the correction model is the irradiation intensity and temperature, and the output is the correction value of each mechanism model parameter, so as to correct the mechanism model parameters.

2. The photovoltaic cell model construction method driven by mechanism-data hybrid according to claim 1, characterized in that: The objective function fitness value is calculated as follows: In formula (2), M represents the total number of experimental measurement data; x represents the parameter vector to be determined; m represents the mth group of experimental measurement data; f(I L ,V L ,x) represents the error function.

3. The photovoltaic cell model construction method driven by mechanism-data hybrid according to claim 2 is characterized in that: The error function is expressed as follows: I ph Represents the photocurrent value flowing through the PN junction after irradiation; I sdi Indicates the reverse saturation current value of the diode; I d Indicates the current value passing through the diode; I sh Indicates the current value flowing through the shunt resistor; R s Indicates the resistance value of parallel resistance; R sh Indicates the resistance value of the series resistor; n i The ideality factor of a diode is used to quantify the degree to which a diode deviates from its ideal behavior in actual operation. In the single diode model, double diode model and triple diode model, the value of N is 1, 2 and 3 respectively.

4. The photovoltaic cell model construction method driven by mechanism-data hybrid according to claim 1, characterized in that: The steps of the AAEMDE algorithm to obtain the mechanism model parameters are as follows: S3.1: Set the population size N p , the maximum number of iterations T max , archive matrix M CR and M K , and the upper bounds U corresponding to the parameters of the photovoltaic cell model b and the lower bound L b Generate an initial population within the range; Calculate the fitness value of the objective function, record the optimal solution, and start iterative optimization; S3.2: Determine whether to adopt an accelerated development mechanism. The specific basis for judgment is: when θ≥10 -5 , it is considered that the current stage is in the early stage of exploration and there is no need to accelerate development, and step S3.3 is performed; when θ<10 -5 , it is considered that the current stage is in the late stage of exploration, and the accelerated development mechanism is adopted to proceed to step S3.7; The mathematical expression of θ is as follows: θ= |Π(x i,g+1 )-Π(x i,g )| (4) In formula (4), Π(x i,g+1 ) and Π(x i,g ) represents the fitness function value of two adjacent iterations; S3.3: Perform mutation operations according to the following formula to promote overall evolution: In formula (5), r1 and r2 are two mutually unequal random integers between [1, NP], and neither of them is equal to i; K i and F i are two scale factors, which are important control parameters that determine the degree of scaling of the differential vector. i =K i ; is from the current population P g After sorting according to fitness, select the top 100p i % a random individual within the range; x r2,g Represents a population from archive A and current population P g A randomly selected individual from the union of i Represents the ratio of elite individuals, which ranges between [0,1] and is given by the following formula; p i =row [p min ,n] (6) In formula (6), the exploration of the early stage p min =2 / NP and n=0.2; S3.4: The CR of each generation of successfully evolved individuals i and K i Recorded separately to archive file S CR and S K Calculate S CR The weighted average mean WA (S CR ) and S K The weighted Lehmer mean WL (S K );S K The weighted Lehmer mean WL (S K )The formula is as follows: S3.5: During the iteration process, mean WA (S CR ) and mean WL (S K ) are stored in M CR and M K In the archive, the formula is as follows: In equations (8) and (9), k represents the continuously updated storage location, 1≤k≤H; S3.6: Based on the index r randomly generated from [1,H] i , respectively, using the following equations (10) and (11) to generate CR i and K i ; CR i =randn i (M CR , ri , 0.1) (10) K i =randc i (M K,ri , 0.1) (11) S3.7: In the accelerated development mechanism, by adjusting the scaling factor K i 、F i and the proportion of elite groups p i , so that the mutation vector is adapted to the step size close to the first 100p i % of individuals and develop around them; at this time, take F i =0.7*K i , reduce p min and n, so that the mutation vector converges to the dominant individual faster, taking p min =1 / NP and n=0.1; finally, the M of this generation K Increase by 0.1 to increase K i The step size can improve the convergence speed; S3.8: Perform a crossover operation. Each individual generates its own test vector u through the crossover operation. i,g , which is derived from the target vector x i,g and mutation vector v i,g The cross combination of is as follows: In formula (12), u i,d,g 、v i,d,g and x i,d,g Respectively represent u i,g 、v i,g and x i,g The dth dimension of , d = 1, 2, ..., D; CR i Determines the generated test vector u i,g From x i,g and v i,g The proportion of components inherited in rand is an integer randomly selected from 1 to D, ensuring that the test vector u i,g At least one dimension comes from the mutation vector v i,g ; S3.9: Use greedy selection operation to select u i,g and x i,g Individuals with better fitness values ​​survive to the next generation; for the minimization problem, the formula is as follows: In formula (13), Π(u i,g ) and Π(x i,g ) represent u i,g and x i,g The fitness function value of S3.10: Determine whether the maximum number of iterations has been reached. If the maximum number of iterations has been reached, the parameter values ​​of the photovoltaic cell mechanism model under the experimental conditions are output and recorded as If the maximum number of iterations is not reached, return to step S3.2 and repeat the above operation.

5. The photovoltaic cell model construction method driven by mechanism-data hybrid according to claim 4 is characterized in that: The settings in step 3.1 are as follows: the photocurrent value I ph The lower bound L b The value is 0A, the upper limit is U b The value is 1A; the reverse saturation current value I sdi The lower bound L b The value is 0μA, the upper limit U b The value is 1μA; the equivalent series resistance value R S Lower bound L b The value is 0Ω, the upper limit U b The value is 1Ω; the diode ideal factor n i The lower bound L b The value is 1, the upper bound U b The value is 2; the equivalent parallel resistance value R Sh Lower bound L b The value is 0Ω, the upper limit U b The value is 100Ω.

6. The photovoltaic cell model construction method driven by mechanism-data hybrid according to claim 4 or 5, characterized in that: The correction model based on BP neural network adopts data-driven modeling technology and establishes a correction model according to the number of parameter vectors to be determined in the mechanism model; the correction model based on BP neural network takes the irradiation intensity and temperature as input, and the outputs are the correction values ​​of the parameters in the photovoltaic cell mechanism model, ΔI ph , ΔI sdi , ΔR s , ΔR sh , Δn i ; An indirect training strategy is used to convert the correction model training into a parameter identification problem, and the process data obtained during the actual power generation process are used to J represents the total number of process data pairs; R represents radiation intensity; T represents temperature, and the correction model is constructed without the need for target output data.

7. The photovoltaic cell model construction method driven by mechanism-data hybrid according to claim 6, characterized in that: The specific description of the indirect training strategy is as follows: The mathematical expression of the BP neural network correction model is as follows: In formulas (14)-(18), They represent the parameters to be determined of the correction model based on BP neural network, and are the corresponding connection weights and thresholds; In order to identify the connection weights and thresholds in the correction model based on the BP neural network, the following objective function is adopted: In formula (19), J represents the total number of process data pairs; z represents the parameter vector in the correction model based on BP neural network to be determined. j represents the jth group of process data; Represents the error function, expressed as formula (20): Finally, using equation (20) as the fitness function, the AAEMDE algorithm is used to solve the parameter vector in the BP neural network correction model. This completes the construction of the correction model.

Images