Improved evolution multi-task photovoltaic model parameter extraction optimization method

Abstract

The invention discloses an improved evolution multi-task photovoltaic model parameter extraction optimization method, and belongs to the technical field of computers. According to the method, a plurality of independent photovoltaic conversion tasks are regarded as a task set capable of referring to experience, parameters such as current and voltage of a photovoltaic module are input into a constructed photovoltaic module model, and a to-be-optimized multi-task set is formed. And then, searching an optimal solution of each task through an improved evolutionary multi-task optimization algorithm. According to the algorithm, the knowledge migration probability between tasks is calculated through the evolution convergence speed of photovoltaic conversion tasks, and a task evolution scheme is dynamically selected through the migration probability. And the similarity between the tasks is measured by using the maximum mean value difference and the grey correlation degree. Meanwhile, an anomaly detection model based on density estimation is used for eliminating a parameter combination scheme which may be unfavorable for task convergence. In addition, a higher-quality parameter combination scheme is generated by adopting probability model sampling. Through the method, an optimal parameter combination scheme can be quickly provided for the photovoltaic module model of each task.

Description

Technical Field

[0001] The invention belongs to the field of computer technology, and in particular relates to an improved evolutionary multi-task photovoltaic model parameter extraction optimization method. Background Art

[0002] With the over-exploitation of mineral energy, energy consumption is increasing, and excessive carbon dioxide emissions are adding fuel to global warming. Therefore, people are more and more urgent to explore and utilize alternative energy. Among them, solar energy has become one of the most promising energy sources with its inexhaustible, clean and pollution-free characteristics. Solar energy is converted into electrical energy through advanced photovoltaic systems, and an accurate photovoltaic model undoubtedly plays a vital role in improving system performance. At present, the nonlinear characteristics of photovoltaic systems have been successfully described by a variety of mathematical models, among which the single diode model and the double diode model have attracted much attention for their accurate consideration of current-voltage characteristics. They provide us with important tools for in-depth insight into battery behavior. However, the parameters in the photovoltaic model have a profound impact on the system performance. Therefore, accurate parameter consistency extraction has become the key to improving efficiency and power optimization.

[0003] At present, in order to solve the problem of parameter extraction in photovoltaic models, two major categories have been developed: deterministic methods and heuristic methods. Although deterministic methods assume that the model is convex and differentiable when modeling, they are highly dependent on the initial solution. Heuristic methods draw wisdom from the principles of biological survival and are favored for their flexibility in dealing with constraints and their excellent performance in accuracy and robustness. However, although heuristic algorithms can provide us with stable solutions, they still face many challenges in determining the global optimal solution. In addition, unlike humans' multitasking capabilities, heuristic methods can usually only handle one problem at a time. However, there is often a wealth of common knowledge between models, which means that insights gained from one model may be borrowed by other models, thereby helping us solve more challenges.

[0004] Evolutionary multi-task optimization is an emerging field in the evolutionary computing community. It is a method that makes full use of existing useful information and solves multiple related tasks in parallel through evolutionary methods. It provides us with a brand new solution. Moreover, evolutionary multi-task optimization technology can improve the quality and convergence speed of solutions by transferring insights across tasks. However, traditional evolutionary multi-task optimization algorithms often rely on artificial parameters such as random pairing probability when determining the strength of knowledge transfer between task pairs. This ignores the evolving local similarities of the population during evolution and may lead to negative transfers that hinder the progress of the population. Summary of the invention

[0005] In view of the problems existing in the above technical solutions, the present invention proposes an improved adaptive evolutionary multi-task photovoltaic model parameter extraction method to provide high-quality photovoltaic models for different photovoltaic system conversion tasks to maximize the light energy conversion efficiency of the photovoltaic system. In order to achieve the above goals, the specific implementation steps of the technical solution adopted by the present invention are as follows:

[0006] 1. An improved evolutionary multi-task photovoltaic model parameter extraction optimization method, characterized in that it includes the following steps:

[0007] S1: Establish a photovoltaic module model framework based on relevant theories.

[0008] A photovoltaic module is usually composed of a plurality of solar cells connected in series and / or in parallel. The photovoltaic module model expression is:

[0009]

[0010] In formula 1, N p and N S are the number of solar cells connected in parallel and in series, respectively. L is the battery output current, I ph is the light generated current, I sd is the reverse saturation current of the diode. s is the series resistance, R sh is the shunt resistor. V L is the output voltage. n is the diode ideal factor, K is the Boltzmann constant (1.3806503×10 -23 J / K), q = 1.60217646 × 10 -19 C is the electron charge and T is the battery temperature (°K).

[0011] The single diode current output model expression is:

[0012]

[0013] The dual diode current output model expression is as follows:

[0014]

[0015] In formula 3, I sd1 is the diffusion current, I sd2 is the saturation current. n1 is the ideal factor of the diffused diode, and n2 is the ideal factor of the composite diode.

[0016] S2: Based on the PV module model, an objective function is constructed to minimize the root mean square error (RMSE) between the measured current data and the calculated current data.

[0017] The root mean square error (RMSE) objective function expression is:

[0018]

[0019] In Equation 4, N is the total number of data points. The error function f for each pair of measured and calculated current data points in the single diode model is k (V L ,I L ,x) is:

[0020]

[0021] The error function f for each pair of measured and calculated current data points in the dual diode model k (V L ,I L ,x) is:

[0022]

[0023] S3: Obtain photovoltaic conversion tasks and corresponding data information, and input the obtained information into the photovoltaic component model and objective function to form a task set to be optimized.

[0024] S4: Use the improved multi-task evolutionary algorithm to simultaneously optimize the tasks in the optimization task set, so that each photovoltaic model obtains a set of relevant parameters with the minimum error.

[0025] S5: Obtain the optimal photovoltaic model for each photovoltaic conversion task under the current situation.

[0026] Preferably, in step S4, the steps of finding the optimal solution of each objective function using an improved evolutionary multi-task algorithm are:

[0027] S4.1 sets the maximum number of iterations G, the initial number of iterations g=1, the number of similar tasks used for knowledge transfer l, and the number of clusters C required to generate offspring.

[0028] S4.2 Population initialization: Establishing the initial population Where K1 is the total number of tasks, and the size of each subpopulation is N. i represents the i-th parameter decision plan, and each individual is assigned a skill factor, which indicates the task to which the individual belongs. Due to the different compositions of different photovoltaic modules, the dimensions of individuals in different populations may be different, so individuals in different populations will be mapped to a dimension of D = Max{D i} and perform corresponding task evaluation on each population.

[0029] S4.3 Calculate the probability of knowledge transfer: Use the adaptive migration probability strategy based on evolutionary distance to calculate the probability of each task generating offspring through knowledge transfer rmp i , according to rmp i Two different strategies are used to generate the next generation of each task. That is, a random number rand is generated. If the random number rand>rmp i When , the population generates offspring through crossover mutation by differential evolution algorithm (DE). i When , knowledge transfer is used to generate offspring. i The calculation formula is as follows:

[0030]

[0031] In formula 7, Defined as the current convergence speed of the task, It is defined as the convergence speed of the task in the previous generation, and the calculation formula is as follows:

[0032]

[0033] In formula 8, f i n,g represents the function value of the nth individual in the gth generation population. When the task converges faster, rmp i Decrease; when the task convergence speed slows down, rmp i Increase.

[0034] S4.4 Differential Evolution Algorithm Generates Offspring: When Random Number Rand>Rmp i When , the population generates offspring through crossover mutation using the differential evolution algorithm (DE).

[0035] S4.5 Knowledge transfer to generate offspring: When the random number rand ≤ rmp i When , knowledge transfer is used to generate offspring.

[0036] S4.6 Elite selection strategy: Sort all individuals in the offspring and parent generations according to the objective function value, and select N individuals with the smallest objective function value as the next generation population.

[0037] S4.7 If the algorithm runs to the maximum number of iterations, it stops iterating and outputs the optimal model parameters. Otherwise, the current number of iterations of the algorithm is increased by 1 and the cycle S4.2-S4.4 continues to search for the optimal solution.

[0038] Preferably, in step S4.4, the step of generating offspring by the differential evolution algorithm is:

[0039] S4.4.1 Mutation operation: All individuals in the population undergo mutation operation, and the individuals in the population use DE / rand / 2 as the mutation operator to generate mutant individuals v i , this strategy randomly selects the target individual x in the current task population r1 , and use 2 random difference vectors to interfere with the target individual. The mutation operator is as follows:

[0040] v i =x r1 +F·(x r2 -x r3 )+F·(x r4 -x r5 ) (Formula 9)

[0041] In formula 9, F is called the variation factor, x r1 ~x r5 is an individual randomly selected from the current task population.

[0042] S4.4.2 Crossover operation: Use binomial crossover to transform the target individual x in the current task population. i With the variant individual v i Crossover generates test individuals u i , thereby enhancing the diversity of the population, and the test individuals directly constitute the offspring population. The binomial crossover operator is defined as follows:

[0043]

[0044] In formula 10, j rand ∈[1,D], used to ensure that the experimental individual u i At least one element in i CR represents the crossover factor, which controls the proportion of elements of the experimental individuals coming from the variant individuals. Generally, CR∈[0,1].

[0045] Preferably, in step S4.5, the step of knowledge migration to generate offspring is:

[0046] S4.5.1 Calculate population similarity: Use the maximum mean difference (MMD) to represent the population similarity between tasks, and calculate the population similarity between tasks separately. The calculation formula for the MMD value between population Q1 and population Q2 is as follows:

[0047]

[0048] In formula 11, x1, x2, ..., x u are individuals of population Q1, y1, y2, ..., y vis an individual of population Q2. k(x,y) is a Gaussian kernel function that can map data from the original space to an infinite-dimensional space. k(x,y) is defined as

[0049]

[0050] In formula (12), σ is the width parameter of the Gaussian kernel function.

[0051] S4.5.2 Calculate the similarity of evolutionary trends: Use the grey correlation to represent the similarity of evolutionary trends between tasks, and calculate the similarity of evolutionary trends between tasks. Take the population evolutionary trend of the current task as the reference sequence Y, and the population evolutionary trends of other tasks as several comparison sequences T = {dir j ,j=1,2,...,|u|}, where dir is the evolutionary trend and u is the set of questions to be compared for trend similarity. Based on the reference sequence and the comparison sequence, the grey correlation analysis method can be used to calculate the similarity of evolutionary trends between different populations. Then the current task is similar to the target task T in dimension d. j The correlation coefficient is:

[0052]

[0053] In formula 13, ρ is the resolution coefficient, which is generally 0.5. Then the current task in dimension d is different from the target task T j The correlation is:

[0054]

[0055] S4.5.3 Similar task selection: Based on population similarity and evolutionary trend similarity, select l tasks for knowledge transfer and sharing of the current task. l depends on the number of photovoltaic conversion tasks, and is generally 2-6.

[0056] S4.5.4 Construct anomaly detection model: All individuals of the selected tasks are used as knowledge transfer samples, and the knowledge transfer strategy based on the current population density estimation anomaly detection model is used to generate the offspring of the current task. First, construct the anomaly detection model of the current task. The anomaly detection model p(x) is constructed as follows:

[0057]

[0058] In Formula 15, μ and Σ are the mean and covariance of the current task population, respectively, |Σ| is the determinant of the covariance Σ, and x is a certain data to be detected.

[0059] S4.5.5 Anomaly detection: Perform anomaly detection on the individuals to be transferred, remove abnormal individuals, select individuals with potential for knowledge transfer, and reduce the probability of negative knowledge transfer. The anomaly detection formula is as follows:

[0060]

[0061] In formula (16), the threshold ε depends on the user preference setting. Data less than the threshold is an outlier. The threshold ε is dynamically updated using the survival rate of the candidate transfer individuals. The update formula is as follows:

[0062]

[0063] In formula 17, |P 1 ,P 2 ,P 3 ...P l | is the total number of individuals in the selected l similar tasks, s j To select N individuals belonging to task T using an elite strategy j The number of individuals.

[0064] S4.5.6 Producing offspring: Using K-means clustering method to cluster Q i Divide into C clusters C is generally set to 10. A Gaussian probability model is constructed in each dimension of each cluster, and the offspring are sampled through the probability model according to the cluster where the parent individual is located. The formula for constructing the Gaussian probability model is as follows:

[0065]

[0066] Compared with the prior art solutions, the technical solutions of the present invention have the following beneficial effects:

[0067] 1. The present invention applies the improved multi-task evolutionary algorithm to the photovoltaic model parameter extraction problem, fully explores and utilizes the knowledge and experience between different photovoltaic model parameter extraction tasks, and greatly improves the problem-solving efficiency while solving multiple model parameter extraction tasks at the same time, providing a high-quality photovoltaic model for each photovoltaic conversion task.

[0068] 2. The present invention estimates the current evolution convergence speed of each task by learning the experience in the task evolution process, and adaptively updates the knowledge transfer probability based on the task evolution convergence speed. According to the dynamic selection method of the knowledge transfer probability, the evolution is maintained within the current population when the current population has an evolutionary advantage, and the knowledge transfer evolution is carried out from other populations in a timely manner when the current population does not have an evolutionary advantage, which effectively balances the self-evolution within each task and the knowledge transfer between tasks, speeds up the task convergence speed, and reduces the probability of the task evolution falling into the local optimum.

[0069] 3. The present invention uses the maximum mean difference to measure the similarity of task populations, which solves the problem that other measurement methods in high-dimensional decision space cannot accurately reflect the differences between high-dimensional decision variables, making the population similarity measurement more accurate. The gray correlation is used to measure the similarity of task evolution trends, which measures the degree of correlation between the evolution trends of two tasks at a very low computational cost. And from the two perspectives of task population similarity and task evolution trend similarity, similar tasks are jointly determined, making the measurement of similar tasks more accurate, the sources of knowledge transfer more diverse, avoiding the influence of negative knowledge transfer to a certain extent, and accelerating the convergence speed of tasks.

[0070] 4. The present invention uses an anomaly detection model to detect anomalies in individuals during knowledge transfer, removes abnormal individuals, and effectively reduces the probability of negative knowledge transfer caused by abnormal individuals. The local probability model is used to capture the distribution of population variables from a global perspective, and sampling is used to generate offspring, which effectively improves the quality of offspring individuals and helps accelerate population convergence. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 is a video tracking flow chart of the present invention;

[0072] Figure 2 Flowchart for finding the optimal solution for the improved multi-task evolutionary algorithm.

[0073] Figure 3 Flowchart for knowledge transfer DETAILED DESCRIPTION

[0074] like Figure 1 As shown, the present invention obtains the optimal photovoltaic model parameter combination scheme through the following steps:

[0075] S1: Establish a photovoltaic module model framework based on relevant theories.

[0076] A photovoltaic module is usually composed of a plurality of solar cells connected in series and / or in parallel. The photovoltaic module model expression is:

[0077]

[0078] In formula 1, N p and N S are the number of solar cells connected in parallel and in series, respectively. L is the battery output current, I ph is the light generated current, I sd is the reverse saturation current of the diode. s is the series resistance, R sh is the shunt resistor. V L is the output voltage. n is the diode ideal factor, K is the Boltzmann constant (1.3806503×10-23 J / K), q = 1.60217646 × 10 -19 C is the electron charge and T is the battery temperature (°K).

[0079] The single diode current output model expression is:

[0080]

[0081] The dual diode current output model expression is as follows:

[0082]

[0083] In formula 3, I sd1 is the diffusion current, I sd2 is the saturation current. n1 is the ideal factor of the diffused diode, and n2 is the ideal factor of the composite diode.

[0084] S2: Based on the PV module model, an objective function is constructed to minimize the root mean square error (RMSE) between the measured current data and the calculated current data.

[0085] The root mean square error (RMSE) objective function expression is:

[0086]

[0087] In Equation 4, N is the total number of data points. The error function f for each pair of measured and calculated current data points in the single diode model is k (V L ,I L ,x) is:

[0088]

[0089] The error function f for each pair of measured and calculated current data points in the dual diode model k (V L ,I L ,x) is:

[0090]

[0091] S3: Obtain photovoltaic conversion tasks and corresponding data information, and input the obtained information into the photovoltaic component model and objective function to form a task set to be optimized.

[0092] S4: Use the improved multi-task evolutionary algorithm to simultaneously optimize the tasks in the optimization task set, so that each photovoltaic model obtains a set of relevant parameters with the minimum error.

[0093] The steps of finding the optimal solution of the objective function through the improved evolutionary multi-task algorithm are as follows: Figure 2 As shown, specifically:

[0094] S4.1 sets the maximum number of iterations G, the initial number of iterations g=1, the number of similar tasks used for knowledge transfer l, and the number of clusters C required to generate offspring.

[0095] S4.2 Population initialization: Establishing the initial population Where K1 is the total number of tasks, and the size of each subpopulation is N. i represents the i-th parameter decision plan, and each individual is assigned a skill factor, which indicates the task to which the individual belongs. Due to the different compositions of different photovoltaic modules, the dimensions of individuals in different populations may be different, so individuals in different populations will be mapped to a dimension of D = Max{D i} and perform corresponding task evaluation on each population.

[0096] S4.3 Calculate the probability of knowledge transfer: Use the adaptive migration probability strategy based on evolutionary distance to calculate the probability of each task generating offspring through knowledge transfer rmp i , according to rmp i Two different strategies are used to generate the next generation of each task. That is, a random number rand is generated. If the random number rand>rmp i When , the population generates offspring through crossover mutation by differential evolution algorithm (DE). i When , knowledge transfer is used to generate offspring. The probability of knowledge transfer rmp for photovoltaic conversion task i i The calculation formula is as follows:

[0097]

[0098] In Formula 7, the photovoltaic conversion task i here is numbered in order to distinguish different photovoltaic conversion tasks and facilitate description. i is a photovoltaic conversion task in the multi-task set, which can also be understood as the i-th photovoltaic conversion task. Defined as the current convergence speed of the task, It is defined as the convergence speed of the task in the previous generation, and the calculation formula is as follows:

[0099]

[0100] In formula 8, f i n,g represents the function value of the nth individual in the gth generation population. When the task converges faster, rmp i Decrease; when the task convergence speed slows down, rmp i Increase.

[0101] In multi-task problems, if the current problem is converging faster, it means that the individuals in the current population have a great advantage. At this time, knowledge transfer may affect the convergence trend of the current task. On the contrary, if the convergence speed of the current problem is decreasing, it means that the individuals in the current population no longer have a clear advantage and may even fall into a local optimum. In addition, photovoltaic conversion tasks are often very complex and may have multiple local optimal solutions. Therefore, we took advantage of the convergence trend in the evolution of photovoltaic conversion tasks and assigned each task a knowledge transfer probability that is dynamically adjusted as the number of generations increases, in order to pursue a good balance between task self-evolution and task knowledge transfer, and to speed up the task solution.

[0102] S4.4 Differential Evolution Algorithm Generates Offspring: When Random Number Rand>Rmp i When , the population generates offspring through crossover mutation using the differential evolution algorithm (DE).

[0103] The steps of differential evolution algorithm to generate offspring are:

[0104] S4.4.1 Mutation operation: All individuals in the population undergo mutation operation, and the individuals in the population use DE / rand / 2 as the mutation operator to generate mutant individuals v i , this strategy randomly selects the target individual x in the current task population r1 , and use 2 random difference vectors to interfere with the target individual. The mutation operator is as follows:

[0105] v i =x r1 +F·(x r2 -x r3 )+F·(x r4 -x r5 ) (Formula 9)

[0106] In formula 9, F is called the variation factor, x r1 ~x r5 is an individual randomly selected from the current task population.

[0107] S4.4.2 Crossover operation: Use binomial crossover to transform the target individual x in the current task population. i With the variant individual v i Crossover generates test individuals u i , thereby enhancing the diversity of the population, and the test individuals directly constitute the offspring population. The binomial crossover operator is defined as follows:

[0108]

[0109] In formula 10, j rand ∈[1,D], used to ensure that the experimental individual u iAt least one element in i CR represents the crossover factor, which controls the proportion of elements of the experimental individuals coming from the variant individuals. Generally, CR∈[0,1].

[0110] S4.5 Knowledge transfer to generate offspring: When the random number rand ≤ rmp i When , knowledge transfer is used to generate offspring.

[0111] like Figure 3 , the steps of knowledge transfer to produce offspring are:

[0112] S4.5.1 Calculate population similarity: Use the maximum mean difference (MMD) to represent the population similarity between tasks, and calculate the population similarity between tasks separately. The calculation formula for the MMD value between population Q1 and population Q2 is as follows:

[0113]

[0114] In formula 11, x1, x2, ..., x u are individuals of population Q1, y1, y2, ..., y v is an individual of population Q2. k(x,y) is a Gaussian kernel function that can map data from the original space to an infinite-dimensional space. k(x,y) is defined as

[0115]

[0116] In formula (12), σ is the width parameter of the Gaussian kernel function.

[0117] S4.5.2 Calculate the similarity of evolutionary trends: Use the grey correlation to represent the similarity of evolutionary trends between tasks, and calculate the similarity of evolutionary trends between tasks. Take the population evolutionary trend of the current task as the reference sequence Y, and the population evolutionary trends of other tasks as several comparison sequences T = {dir j ,j=1,2,...,|u|}, where dir is the evolutionary trend and u is the set of questions to be compared for trend similarity. Based on the reference sequence and the comparison sequence, the grey correlation analysis method can be used to calculate the similarity of evolutionary trends between different populations. Then the current task is similar to the target task T in dimension d. j The correlation coefficient is:

[0118]

[0119] In formula 13, ρ is the resolution coefficient, which is generally 0.5. Then the current task in dimension d is different from the target task T j The correlation is:

[0120]

[0121] S4.5.3 Similar task selection: Based on population similarity and evolutionary trend similarity, select l tasks for knowledge transfer and sharing of the current task. l depends on the number of photovoltaic conversion tasks, and is generally 2-6.

[0122] S4.5.4 Construct anomaly detection model: All individuals of the selected tasks are used as knowledge transfer samples, and the knowledge transfer strategy based on the current population density estimation anomaly detection model is used to generate the offspring of the current task. First, construct the anomaly detection model of the current task. The anomaly detection model p(x) is constructed as follows:

[0123]

[0124] In Formula 15, μ and Σ are the mean and covariance of the current task population, respectively, |Σ| is the determinant of the covariance Σ, and x is a certain data to be detected.

[0125] S4.5.5 Anomaly detection: Perform anomaly detection on the individuals to be transferred, remove abnormal individuals, select individuals with potential for knowledge transfer, and reduce the probability of negative knowledge transfer. The anomaly detection formula is as follows:

[0126]

[0127] In formula (16), the threshold ε depends on the user preference setting. Data less than the threshold is an outlier. The threshold ε is dynamically updated using the survival rate of the candidate transfer individuals. The update formula is as follows:

[0128]

[0129] In formula 17, |P 1 ,P 2 ,P 3 ...P l | is the total number of individuals in the selected l similar tasks, s j To select N individuals belonging to task T using an elite strategy j The number of individuals.

[0130] S4.5.6 Producing offspring: Using K-means clustering method to cluster Q i Divide into C clusters C is generally set to 10. A Gaussian probability model is constructed in each dimension of each cluster, and the offspring are sampled through the probability model according to the cluster where the parent individual is located. The formula for constructing the Gaussian probability model is as follows:

[0131]

[0132] S4.6 Elite selection strategy: Sort all individuals in the offspring and parent generations according to the objective function value, and select N individuals with the smallest objective function value as the next generation population.

[0133] S4.7 If the algorithm runs to the maximum number of iterations, it stops iterating and outputs the optimal model parameters. Otherwise, the current number of iterations of the algorithm is increased by 1 and the cycle S4.2-S4.4 continues to search for the optimal solution.

[0134] S5: Obtain the optimal photovoltaic model for each photovoltaic conversion task under the current situation.

[0135] The optimization method was tested 30 times in three different 50 photovoltaic conversion tasks, including single diode, double diode, and photovoltaic module. The simulated data obtained were consistent with the actual data, which confirmed the effectiveness and accuracy of parameter estimation. In the comparative test with various photovoltaic model parameter extraction methods including classic deterministic methods, heuristic methods, and the latest evolutionary multi-task methods, the root mean square error value of the parameter extraction method in the present invention is better than that of other methods.

[0136] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. An improved evolutionary multi-task photovoltaic model parameter extraction optimization method, characterized in that: The following steps are involved: S1: Establish a photovoltaic module model framework based on relevant theories. A photovoltaic module is usually composed of a plurality of solar cells connected in series and / or in parallel. The photovoltaic module model expression is: In formula 1, N p and N S are the number of solar cells connected in parallel and in series, respectively. L is the battery output current, I ph is the light generated current, I sd is the reverse saturation current of the diode. s is the series resistance, R sh is the shunt resistor. V L is the output voltage. n is the diode ideal factor, K is the Boltzmann constant (1.3806503×10 -23 J / K), q = 1.60217646 × 10 -19 C is the electron charge and T is the battery temperature (°K). The single diode current output model expression is: The dual diode current output model expression is as follows: In formula 3, I sd1 is the diffusion current, I sd2 is the saturation current. n1 is the ideal factor of the diffused diode, and n2 is the ideal factor of the composite diode. S2: Based on the PV module model, an objective function is constructed to minimize the root mean square error (RMSE) between the measured current data and the calculated current data. The root mean square error (RMSE) objective function expression is: In Equation 4, N is the total number of data points. The error function f for each pair of measured and calculated current data points in the single diode model is k (V L ,I L ,x) is: The error function f for each pair of measured and calculated current data points in the dual diode model k (V L ,I L ,x) is: S3: Obtain photovoltaic conversion tasks and corresponding data information, and input the obtained information into the photovoltaic component model and objective function to form a task set to be optimized. S4: Use the improved multi-task evolutionary algorithm to simultaneously optimize the tasks in the optimization task set, so that each photovoltaic model obtains a set of relevant parameters with the minimum error. S5: Obtain the optimal photovoltaic model for each photovoltaic conversion task under the current situation.

2. The improved evolutionary multi-task photovoltaic model parameter extraction optimization method according to claim 1, characterized in that: In step S4, the steps of an improved evolutionary multi-task algorithm to find the optimal solution of each objective function are: S4.1 sets the maximum number of iterations G, the initial number of iterations g=1, the number of similar tasks used for knowledge transfer l, and the number of clusters C required to generate offspring. S4.2 Population initialization: Establishing the initial population Where K1 is the total number of tasks, and the size of each subpopulation is N. i represents the i-th parameter decision plan, and each individual is assigned a skill factor, which indicates the task to which the individual belongs. Due to the different compositions of different photovoltaic modules, the dimensions of individuals in different populations may be different, so individuals in different populations will be mapped to a dimension of D = Max{D i } and perform corresponding task evaluation on each population. S4.3 Calculate the probability of knowledge transfer: Use the adaptive migration probability strategy based on evolutionary distance to calculate the probability of each task generating offspring through knowledge transfer rmp i , according to rmp i Two different strategies are used to generate the next generation of each task. That is, a random number rand is generated. If the random number rand>rmp i When , the population generates offspring through crossover mutation by differential evolution algorithm (DE). i When , knowledge transfer is used to generate offspring. i The calculation formula is as follows: In formula 7, Defined as the current convergence speed of the task, It is defined as the convergence speed of the task in the previous generation, and the calculation formula is as follows: In formula 8, f i n,g represents the function value of the nth individual in the gth generation population. When the task converges faster, rmp i Decrease; when the task convergence speed slows down, rmp i Increase. S4.4 Differential Evolution Algorithm Generates Offspring: When Random Number Rand>Rmp i When , the population generates offspring through crossover mutation using the differential evolution algorithm (DE). S4.5 Knowledge transfer to generate offspring: When the random number rand ≤ rmp i When , knowledge transfer is used to generate offspring. S4.6 Elite selection strategy: Sort all individuals in the offspring and parent generations according to the objective function value, and select N individuals with the smallest objective function value as the next generation population. S4.7 If the algorithm runs to the maximum number of iterations, it stops iterating and outputs the optimal model parameters. Otherwise, the current number of iterations of the algorithm is increased by 1 and the cycle S4.2-S4.4 continues to search for the optimal solution.

3. The improved evolutionary multi-task photovoltaic model parameter extraction optimization method according to claim 2, characterized in that: In step S4.4, the steps of generating offspring by the differential evolution algorithm are: S4.4.1 Mutation operation: All individuals in the population undergo mutation operation, and the individuals in the population use DE / rand / 2 as the mutation operator to generate mutant individuals v i , this strategy randomly selects the target individual x in the current task population r1 , and use 2 random difference vectors to interfere with the target individual. The mutation operator is as follows: v i =x r1 +F·(x r2 -x r3 )+F·(x r4 -x r5 ) (Formula 9) In formula 9, F is called the variation factor, x r1 ~x r5 is an individual randomly selected from the current task population. S4.4.2 Crossover operation: Use binomial crossover to transform the target individual x in the current task population. i With the variant individual v i Crossover generates test individuals u i , thereby enhancing the diversity of the population, and the test individuals directly constitute the offspring population. The binomial crossover operator is defined as follows: In formula 10, j rand ∈[1,D], used to ensure that the experimental individual u i At least one element in i CR represents the crossover factor, which controls the proportion of elements of the experimental individuals coming from the variant individuals. Generally, CR∈[0,1].

4. The improved evolutionary multi-task photovoltaic model parameter extraction optimization method according to claim 2, characterized in that: In step S4.5, the steps of knowledge transfer to generate offspring are: S4.5.1 Calculate population similarity: Use the maximum mean difference (MMD) to represent the population similarity between tasks, and calculate the population similarity between tasks separately. The calculation formula for the MMD value between population Q1 and population Q2 is as follows: In formula 11, x1, x2, ..., x u are individuals of population Q1, y1, y2, ..., y v is an individual of population Q2. k(x,y) is a Gaussian kernel function that can map data from the original space to an infinite-dimensional space. k(x,y) is defined as In formula (12), σ is the width parameter of the Gaussian kernel function. S4.5.2 Calculate the similarity of evolutionary trends: Use the grey correlation to represent the similarity of evolutionary trends between tasks, and calculate the similarity of evolutionary trends between tasks. Take the population evolutionary trend of the current task as the reference sequence Y, and the population evolutionary trends of other tasks as several comparison sequences T = {dir j ,j=1,2,...,|u|}, where dir is the evolutionary trend and u is the set of questions to be compared for trend similarity. Based on the reference sequence and the comparison sequence, the grey correlation analysis method can be used to calculate the similarity of evolutionary trends between different populations. Then the current task is similar to the target task T in dimension d. j The correlation coefficient is: In formula 13, ρ is the resolution coefficient, which is generally 0.

5. Then the current task in dimension d is different from the target task T j The correlation is: S4.5.3 Similar task selection: Based on population similarity and evolutionary trend similarity, select l tasks for knowledge transfer and sharing of the current task. l depends on the number of photovoltaic conversion tasks, and is generally 2-6. S4.5.4 Construct anomaly detection model: All individuals of the selected tasks are used as knowledge transfer samples, and the knowledge transfer strategy based on the current population density estimation anomaly detection model is used to generate the offspring of the current task. First, construct the anomaly detection model of the current task. The anomaly detection model p(x) is constructed as follows: In Formula 15, μ and Σ are the mean and covariance of the current task population, respectively, |Σ| is the determinant of the covariance Σ, and x is a certain data to be detected. S4.5.5 Anomaly detection: Perform anomaly detection on the individuals to be transferred, remove abnormal individuals, select individuals with potential for knowledge transfer, and reduce the probability of negative knowledge transfer. The anomaly detection formula is as follows: In formula (16), the threshold ε depends on the user preference setting. Data less than the threshold is an outlier. The threshold ε is dynamically updated using the survival rate of the candidate transfer individuals. The update formula is as follows: In formula 17, |P 1 ,P 2 ,P 3 ...P l | is the total number of individuals in the selected l similar tasks, s j To select N individuals belonging to task T using an elite strategy j The number of individuals. S4.5.6 Producing offspring: Using K-means clustering method to cluster Q i Divide into C clusters C is generally set to 10. A Gaussian probability model is constructed in each dimension of each cluster, and the offspring are sampled through the probability model according to the cluster where the parent individual is located. The formula for constructing the Gaussian probability model is as follows:

Images