Photovoltaic system performance optimization method, system, equipment and medium

Abstract

The invention provides a photovoltaic system performance optimization method, system and device and a medium, and belongs to the technical field of photovoltaic cell simulation model performance optimizing.The method comprises the following steps that a plurality of photovoltaic cell simulation models of a photovoltaic system are constructed, each photovoltaic cell simulation model is defined as an optimization task, and the optimization task is set up; initializing a population and an incremental learning agent model for each optimization task; taking one optimization task as a target task, taking other optimization tasks as source tasks, and constructing a similarity matrix according to the similarity of the incremental learning agent models of the target task and the plurality of source tasks; in combination with the similarity matrix, using an incremental learning agent model to guide global knowledge migration in a target space, and using a multivariate Gaussian model to guide local knowledge migration in a decision space; and dynamically adjusting the migration proportion according to the success rate of generation of offspring by knowledge migration, and carrying out iteration to obtain an optimized photovoltaic cell simulation model. According to the invention, the performance of the photovoltaic cell simulation model can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of performance optimization of photovoltaic cell simulation models, and particularly relates to a method, a system, a device and a medium for optimizing the performance of a photovoltaic system. Background Art

[0002] Photovoltaic systems are crucial components in power systems. To achieve the optimal design and operation of photovoltaic systems, it is very necessary to establish photovoltaic cell simulation models. These models generate current by releasing photon energy to prompt electrons to flow under the action of an electric field, so as to simulate the behavior of real photovoltaic cells, obtain current-voltage data measured under various operating conditions, and thus be used to evaluate the performance of existing photovoltaic systems, identify potential problems and improvement spaces. A good photovoltaic cell simulation model is crucial for the simulation, design, evaluation, control and optimization of solar photovoltaic power generation systems. Therefore, optimizing the performance of photovoltaic cell simulation models has very important practical significance.

[0003] Multi-task optimization aims to study multiple optimization tasks simultaneously, and utilize the potentially common useful knowledge between related tasks to promote each other, thereby improving the ability to solve each task in parallel. As an emerging paradigm in the field of evolutionary optimization, multi-task optimization achieves the purpose of processing multiple optimization tasks simultaneously through knowledge transfer between different tasks. However, related research shows that this knowledge transfer can effectively improve the performance of task optimization algorithms in dealing with similar tasks, but for tasks with significantly different optimal solutions or fitness values, it may bring negative impacts, resulting in relatively poor performance or even worse than traditional single-task optimization. In multi-task optimization, this is also known as the negative transfer phenomenon.

[0004] When using multi-task optimization to optimize the performance of photovoltaic cell simulation models, as the population evolves, the distribution of the population and the similarity between tasks will constantly change dynamically. Inappropriate knowledge transfer may interfere with the optimization process and result in the negative transfer phenomenon, thereby leading to a decline in the performance of photovoltaic cell simulation models. Summary of the Invention

[0005] In order to overcome the deficiencies of the above-mentioned prior art, the present invention provides a method for optimizing the performance of a photovoltaic system, including the following steps:

[0006] Construct multiple photovoltaic cell simulation models of the photovoltaic system, define each photovoltaic cell simulation model as an optimization task, and initialize a population and an incremental learning proxy model for each optimization task;

[0007] Take one optimization task among multiple optimization tasks as the target task, and take the other optimization tasks as source tasks. Calculate the similarity between the target task and the incremental learning surrogate models corresponding to multiple source tasks using the CKA index, and construct a similarity matrix based on the similarity between the target task and the incremental learning surrogate models corresponding to multiple source tasks;

[0008] Take the source task with the highest similarity to the target task in the similarity matrix as the knowledge transfer object. When the probability of the knowledge transfer object undergoing transfer is less than the knowledge transfer probability, perform inter-population evolution. Use the incremental learning surrogate model corresponding to the source task with the highest similarity to the target task to guide global knowledge transfer in the target space, and use the multivariate Gaussian model to guide local knowledge transfer in the decision space;

[0009] Dynamically adjust the transfer ratio according to the success rate of generating offspring through knowledge transfer;

[0010] Repeat the processes of constructing the similarity matrix, knowledge transfer, and dynamically adjusting the transfer ratio, and continuously iterate until the number of evaluations of the objective function of the multi-task evolutionary algorithm reaches the set threshold, then the iteration ends, and an optimized photovoltaic cell simulation model is obtained.

[0011] Preferably, the calculating the similarity between the target task and the incremental learning surrogate models corresponding to multiple source tasks using the CKA index includes the following steps:

[0012] Extract the similarity metric vector X i ={x1, x2, …, x N};

[0013] Input X i into the incremental learning surrogate models corresponding to all source tasks respectively, calculate the hidden layer output feature matrix and the final output layer feature matrix of the incremental learning surrogate models, and use the linear kernel function to convert the hidden layer output feature matrix and the output layer feature matrix into corresponding kernel matrices;

[0014] Perform centering processing on the kernel matrices to remove the mean of the similarity metric vector;

[0015] Calculate the hidden layer CKA index and the output layer CKA index of the centered kernel matrices using the independence criterion, and calculate the average of the hidden layer CKA index and the output layer CKA index, and take the average as the CKA index of the target task.

[0016] Preferably, it further includes that when the probability of the knowledge transfer object undergoing transfer is greater than or equal to the knowledge transfer probability, perform intra-population evolution.

[0017] Preferably, the incremental learning surrogate model is used to guide global knowledge transfer in the target space, and the multivariate Gaussian model is used to guide local knowledge transfer in the decision space, specifically: setting a threshold for the transfer selection probability rmp, when the random number is less than the threshold of the transfer selection probability rmp, the incremental learning surrogate model is used to guide global knowledge transfer in the target space; when the random number is greater than the threshold of the transfer selection probability rmp, the multivariate Gaussian model is used to guide local knowledge transfer in the decision space.

[0018] The present invention also provides a photovoltaic system performance optimization device, including:

[0019] An initialization module, configured to construct multiple photovoltaic cell simulation models of a photovoltaic system, define each photovoltaic cell simulation model as an optimization task, and initialize a population and an incremental learning surrogate model for each optimization task;

[0020] A similarity calculation module, configured to use one optimization task among multiple optimization tasks as a target task, use other optimization tasks as source tasks, calculate the similarity between the target task and the incremental learning surrogate models corresponding to multiple source tasks by using the CKA index, and construct a similarity matrix according to the similarity between the target task and the incremental learning surrogate models corresponding to multiple source tasks;

[0021] A knowledge transfer module, configured to use the source task with the highest similarity to the target task in the similarity matrix as the knowledge transfer object, and when the probability of the knowledge transfer object to occur transfer is less than the knowledge transfer probability, perform inter-population evolution, use the incremental learning surrogate model corresponding to the source task with the highest similarity to the target task to guide global knowledge transfer in the target space, and use the multivariate Gaussian model to guide local knowledge transfer in the decision space;

[0022] A transfer ratio adjustment module, configured to dynamically adjust the transfer ratio according to the success rate of generating offspring by knowledge transfer;

[0023] A performance optimization module, configured to repeat the processes of constructing the similarity matrix, knowledge transfer, and dynamically adjusting the transfer ratio, continuously perform iteration until the number of objective function evaluations of the multi-task evolutionary algorithm reaches a set threshold, then the iteration ends, and an optimized photovoltaic cell simulation model is obtained.

[0024] The present invention also provides a computer device, including a memory and a processor; the memory stores a computer program, and the processor is configured to run the computer program in the memory to execute the photovoltaic system performance optimization method.

[0025] The present invention also provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the photovoltaic system performance optimization method.

[0026] The photovoltaic system performance optimization method provided by the present invention has the following beneficial effects: The present invention defines each photovoltaic cell simulation model as an optimization task, assigns a specific population to each optimization task, establishes an incremental learning surrogate model for each optimization task, and calculates the similarity between the target task and the incremental learning surrogate models corresponding to multiple source tasks using the CKA index. This process can quantify the dynamic similarity between tasks; according to the different similarity degrees of tasks, different knowledge transfer methods are adopted. By combining the incremental learning surrogate model and the multivariate Gaussian model, a dual-model guidance strategy is designed. The incremental learning surrogate model learns the overall structure of the task history and conducts global guidance in the target space, using limited data to continuously learn the overall knowledge; the multivariate Gaussian model learns the population distribution and guides local knowledge transfer in the decision space. By using different knowledge transfer strategies, the algorithm performance is further improved; at the same time, the knowledge transfer probability is adaptively adjusted by the success rate of generating offspring through knowledge transfer and dynamically adjusting the transfer ratio, effectively avoiding the occurrence of negative transfer phenomena. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention and their design schemes, the accompanying drawings required for this embodiment will be briefly introduced below. The accompanying drawings in the following description are only partial embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0028] Figure 1 It is a schematic diagram of the multi-task evolutionary algorithm of the present invention;

[0029] Figure 2 It is a convergence schematic diagram of the method of the present invention and 5 comparison algorithms on 9 single-objective multi-task benchmark problems proposed in the CEC2017 evolutionary multi-task optimization competition; among them, Figure 2 (a1)-(i2) in it are the convergence schematic diagrams of the method of the present invention and 5 comparison algorithms on different single-objective multi-task benchmark problems respectively.

[0030] Figure 3 It is a flowchart of the performance optimization of the present invention;

[0031] Figure 4 It is an equivalent circuit diagram of the photovoltaic cell simulation model in the optimization method of the present invention; among them, Figure 4 (a) of Figure 4 (b) of Figure 4 (c) of respectively represent the equivalent circuit diagrams of three different photovoltaic cell simulation models;

[0032] Figure 5It is the convergence graph of the performance optimization method of the photovoltaic system of the present invention and the performance optimization of the photovoltaic cell simulation model by 5 comparative algorithms; among them, Figure 5 (a) of Figure 5 (b) of Figure 5 and (c) of Specific implementation manners

[0033] In order to enable those skilled in the art to better understand the technical solution of the present invention and be able to implement it, the present invention will be described in detail below in conjunction with the drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0034] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "axial", "radial", "circumferential", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the technical solution of the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention.

[0035] In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In the description of the present invention, it should be noted that unless otherwise clearly specified or limited, the terms "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. In the description of the present invention, unless otherwise stated, the meaning of "plurality" is two or more, and details are not described herein again.

[0036] Embodiment

[0037] The present invention provides a method for optimizing the performance of a photovoltaic system, specifically as Figure 1 shown, including the following steps:

[0038] Step 1: Construct multiple photovoltaic cell simulation models of the photovoltaic system, define each photovoltaic cell simulation model as an optimization task, and initialize a population and an incremental learning agent model for each optimization task.

[0039] Step 2: Take one optimization task among multiple optimization tasks as the target task, and take the other optimization tasks as source tasks. Calculate the similarity between the target task and the incremental learning surrogate models corresponding to multiple source tasks using the CKA index, and construct a similarity matrix based on the similarity between the target task and the incremental learning surrogate models corresponding to multiple source tasks. The specific steps are as follows:

[0040] Step 2.1: Extract the similarity metric vector X i from the population P i of the target task, where X N ={x1, x2, …, x

[0041] }. i Step 2.2: Input X

[0042] into the incremental learning surrogate models corresponding to all source tasks respectively, calculate the output feature matrix of the hidden layer and the output feature matrix of the final output layer of the incremental learning surrogate models, and convert the output feature matrix of the hidden layer and the output feature matrix into corresponding kernel matrices using a linear kernel function. The kernel matrix (Gram matrix) represents the relationship between data points by calculating the similarity of a set of data points under a certain kernel function. Common kernel functions include linear kernel, Gaussian kernel (RBF kernel), polynomial kernel, etc. In the present invention, a linear kernel function is used to calculate the Gram matrix, and the kernel function is expressed as: i That is, the value of the linear kernel function is equal to the inner product of sample x j and sample x

[0043] Calculate the corresponding kernel matrices K and L of the output feature matrix of the hidden layer and the output feature matrix. The kernel matrix maps the input data to a high-dimensional feature space to represent the similarity between samples. The calculation formula is as follows:

[0044]

[0045] where K and L are N×N matrices, M i , M j are both matrices, and are the transposes of M i and M j respectively.

[0046] Step 2.3: Center the kernel matrix. The purpose is to ensure that the kernel matrix only reflects relative similarity and is not affected by the overall position by removing the mean influence of the similarity metric vector, making the similarity metric independent of the translation of the dataset and enhancing the robustness to outliers.

[0047] Perform centering processing on the Gram matrix. The centering matrix is:

[0048]

[0049] Among them, I n is an n×n identity matrix, 1 n is a column vector of all 1s with length n, and n represents the number of samples of the matrix. and are the K and L centering matrices respectively.

[0050] Step 2.4: Calculate the hidden layer CKA index and the output layer CKA index of the centered kernel matrix using the independence criterion, and calculate the average of the hidden layer CKA index and the output layer CKA index, and use the said average as the CKA index of the target task. The formula is as follows:

[0051] Calculate the CKA index using the independence criterion Hilbert-Schmidt (Hilbert-Schmidt Independence Criterion, HSIC). The CKA index is defined as:

[0052]

[0053] In the formula, CKA hidden is the hidden layer CKA index, CKA final is the output layer CKA index, CKA avg is the average of the hidden layer CKA index and the output layer CKA index, and HSIC(K, L) is the independence metric index used to measure the correlation degree between two variables; is the trace of the product of matrices K and L, that is, the sum of the diagonal elements of the matrix, n is the number of samples, that is, the dimensions of matrices K and L; CKA(K, L) is the similarity metric index of matrices K and L; HSIC(K, K) is the independence metric value of matrices K and K; HSIC(L, L) is the independence metric value of matrices L and L.

[0054] Fill the similarity matrix similar_matrix ij , where i represents the source task and j represents the target task, where i≠j. According to similar_matrix ij select the incremental proxy model T max_sim_index with the highest similarity as the source task for migration, and update the knowledge transfer probability TRP. During the evolution process, the source task will be continuously updated for effective knowledge transfer.

[0055] Step 3: Use the source task with the highest similarity to the target task in the similarity matrix as the knowledge transfer object. When the probability of the knowledge transfer object undergoing transfer is less than the knowledge transfer probability, perform inter-population evolution. Use the incremental learning surrogate model to guide global knowledge transfer in the target space and use the multivariate Gaussian model to guide local knowledge transfer in the decision space. The specific steps are as follows:

[0056] Step 3.1: If a random event (probability of transfer) rand < TRP i , perform inter-population evolution.

[0057] When performing inter-population evolution, it is necessary to use the incremental surrogate model for global guidance, use the multivariate Gaussian model for local guidance, and finally determine the transfer candidate solution and its quantity num. At the same time, update the incremental surrogate model and accumulate population data after the transfer occurs. Finally, adopt the elitist strategy to select the top N individuals from as the next-generation population, where P i represents the parent population, O i represents the offspring solutions generated by crossover and compilation, represents the transfer candidate solution.

[0058] Specifically, set the threshold of the transfer selection probability rmp, and use the transfer selection probability rmp to guide global and local knowledge transfer respectively to optimize the performance of the multi-task evolution algorithm. rmp changes with the evolution process. In the early stage, the population is more dispersed and the evolution is incomplete, so the objective space similarity guided globally in the early stage is more accurate. In the later stage, the evolution of the population is affected by their respective solvers and transfer strategies, and the accuracy of the target space similarity decreases, and the overall gradually tends to local optimum or global optimum. Therefore, the similarity guided by the local decision space in the later stage is more accurate. As shown in formula (7).

[0059]

[0060] In the formula, rmp is the transfer selection probability, gen is the current iteration generation; gen max is the maximum iteration generation.

[0061] (1) When the random number rand < rmp transfer selection probability, the multi-task optimization algorithm uses the incremental surrogate model for global guidance. The specific process is as follows:

[0062] First, load the incremental surrogate model and Fisher information matrix learned in the previous task to retain historical knowledge, and construct a new neural network surrogate model (incremental surrogate model) including three fully connected layers. Predict the fitness value score of the optimal source task population P max_sim_index on the surrogate model.

[0063] Ensure the consistency of the model architecture for all tasks. By collecting population information, generate the model training set training_pop i and training_fitness i , and perform data preprocessing.

[0064] Finally, adopt the incremental learning technique based on elastic weight constraint to retain the historical information of the incremental surrogate model. By calculating the Fisher information matrix, measure the importance of each parameter for the previous tasks. During the learning process of the new task, use formula (9) as part of the loss function to prevent the model from forgetting the old tasks. After training, the new model and the Fisher information are archived for use in subsequent learning. The calculation formula is as follows:

[0065] The diagonal elements of the Fisher information matrix F are defined as:

[0066]

[0067] where p(y|x,θ) is the output distribution corresponding to the model parameter θ given the input x; is the derivative of the log-likelihood function with respect to the parameter θ i ; F i is the i-th diagonal element of the Fisher information matrix, representing the sensitivity of the parameter θ i to the previous tasks, is the expectation calculation for a certain random variable.

[0068] The elastic weight constraint combines the loss L new (θ) of the current task and the constraint on the weights of the previous tasks Its loss function is as follows:

[0069]

[0070] where θ is the model parameter of the current task, is the parameter learned in the previous task, F i is the i-th diagonal element of the Fisher information matrix, representing the sensitivity of the parameter θ i to the previous tasks, and λ is a hyperparameter that controls the trade-off between the loss of the new task and the memory of the previous tasks.

[0071] (2) When rand >= rmp, the multi-task optimization algorithm uses the multivariate Gaussian model for local guidance. This strategy utilizes the population information in the current iteration state, calculates the mean and variance of the population of the target task, and constructs the multivariate normal distribution density function. Through the density function p(X), select the dominant individuals from the source tasks as the migration candidate solutions to guide the local migration. The calculation formula is as follows:

[0072] For a D - dimensional dataset composed of N individuals, the multivariate Gaussian model describes the data distribution through the mean vector μ and the covariance matrix Σ.

[0073]

[0074] where x (i) is the D - dimensional vector of the i - th individual, m is the total number of individuals of individual i, and here m is equal to the N individuals in the dataset.

[0075] Based on the mean μ and covariance Σ of the data, a statistical model is established, and a statistical model p(X) is fitted to describe the central tendency and dispersion of the data, and can also reveal the mutual relationship between different variables to reflect the distribution state of the population in the decision space.

[0076]

[0077] where X is a D - dimensional random vector, μ is the D - dimensional mean vector, Σ is the D - dimensional covariance matrix that describes the variance and covariance between random variables, |Σ| is the determinant of the covariance matrix, and Σ -1 is the inverse matrix of the covariance matrix.

[0078] Step 3.2: If rand >= TRP i , then perform evolution within the population. Also adopt the elitist strategy, select the top N individuals from P i ∪O i as the next - generation population, where the offspring O i is generated through the differential evolution algorithm DE (Differential Evolution) strategy.

[0079] Step 4: Dynamically adjust the migration ratio according to the success rate of generating offspring by knowledge migration.

[0080] Specifically: After each migration, comprehensively consider the success rate of generating offspring by knowledge migration, calculate the effective migration ratio ε i , and update the migration individual constraint parameter NL. NL reflects the effective migration ratio of the previous generation, so as to adaptively adjust the selection of knowledge - migration individuals between tasks. The calculation formula of the effective migration ratio is as follows:

[0081]

[0082] where s j is the number of migration individuals that successfully enter the next generation from task T j , and m j is the total number of migration candidate solutions transferred from task T i .

[0083] Step 5: Repeat the processes of constructing the similarity matrix, knowledge transfer, and dynamically adjusting the transfer ratio, and continuously iterate until the number of evaluations of the objective function of the multi-task evolutionary algorithm reaches the set threshold, then the iteration ends, and an optimized photovoltaic cell simulation model is obtained.

[0084] To verify the effectiveness of the method of the present invention, it is further illustrated by the following simulation experiments.

[0085] Table 1 Characteristics of nine benchmark SOMTO problems in the CEC2017 test suite

[0086]

[0087] The performance of the algorithm proposed by the present invention is verified by testing a single-objective multi-task benchmark problem proposed in the CEC2017 evolutionary multi-task optimization competition with a method for optimizing the performance of a photovoltaic system of the present invention. The CEC2017 evolutionary multi-task optimization includes 9 benchmark problems, and their properties are briefly given in Table 1. The similarity recorded in Table 1 is based on the rank correlation coefficient of two tasks in the MTO problem. There are high, medium, and low similarities (denoted as HS, MS, and LS) in the MTO problem. In addition, according to the intersection degree of the global optima of the benchmark problems in the unified search space, the optimization problems are divided into: complete intersection (CI), partial intersection (PI), and no intersection (NI).

[0088] A method for optimizing the performance of a photovoltaic system (MTEA-DMSL) of the present invention is compared with the effects of five advanced multi-task evolutionary algorithms, namely MFEA, SREMTO, EMEA, MTEA-SaO, and MTEA-AD. To ensure fairness, the population size of all algorithms in the experiment is set to N = 100*k, and the total number of evaluations of the objective function maxFES is set to 100000*k. When the number of function evaluations reaches maxFES, the algorithm terminates, and the mean and standard deviation of the best objective function values obtained from 20 termination runs are used to measure the performance of the algorithm. The results are as Figure 2 and shown in Table 2. Figure 2 The (a1)-(i2) in

[0089] Table 2 Average best objective function values of MTEA-DMSL and five comparison algorithms on the CEC 2017 benchmark problems

[0090]

[0091] Table 2 presents the mean and variance of the average best objective function values obtained by running the algorithm proposed in this invention (abbreviated as MFEA-DMSL for convenience of description) and five comparison algorithms independently 20 times on the CEC 2017 test problems. In this experiment, the Wilcoxon rank-sum test with a 95% confidence level was used to analyze the results. "+", "-", and "≈" indicate that the compared algorithm is significantly better than, worse than, or has no difference from MFEA-DMSL, respectively. The bold data in the table represents the best performance in the task.

[0092] MTEA-DMSL showed better performance than other algorithms in 14 out of 18 tasks. MTEA-SaO, MTEA-AD, and MFEA-II obtained the best results in the remaining 2 tasks and 1 task respectively. Therefore, in terms of the number of best objective function means, MTEA-DMSL has strong competitiveness. According to the results obtained from the rank-sum test, MTEA-DMSL achieved better results compared to other algorithms.

[0093] From Figure 2 It can be seen that under the same number of function evaluations, MTEA-DMSL has significantly better convergence performance than other algorithms in 14 out of 18 tasks. MTEA-DMSL still needs to be improved in dealing with highly similar tasks and some complex tasks, while it has very good performance in dealing with low-similarity tasks and most complex tasks, and has strong competitive advantages.

[0094] By testing a multi-task performance optimization method for photovoltaic systems (MTEA-DMSL) based on incremental learning surrogate models of this invention on three commonly used photovoltaic cell simulation models, the performance of the algorithm proposed in this invention was verified. The performance optimization methods of these three photovoltaic cell simulation models were combined into a multi-task optimization problem, and the objective functions of the three photovoltaic cell simulation models are:

[0095] (1) SDM:

[0096]

[0097] (2) DDM:

[0098]

[0099] (3) SMM:

[0100]

[0101] Among them, the objective function is the root mean square error (RMSE) of f(x), which represents the difference between the predicted value and the actual observed value of the PV model (photovoltaic cell simulation model).

[0102] Among them, f(x) is the objective function, x is the independent variable, I ph is the photocurrent, I d is the diode current, I sh is the shunt resistance current, I o represents the diode reverse saturation current, I is the output current parameter, a is the diode ideality factor, R s and R sh are the series resistance and the shunt resistance respectively, V is the battery output voltage, V t is the junction thermal voltage, I o1 、I o2 represent the diffusion current and the saturation current respectively. a1 and a2 represent the first diode ideality factor and the second diode ideality factor respectively.

[0103] The performance optimization flowcharts of three photovoltaic cell simulation models are as Figure 3 shown, and the equivalent circuit diagrams of the cell simulation models are as Figure 4 shown, where Figure 4 (a) of Figure 4 (b) of Figure 4 (c) of

[0104] Compare the performance optimization method (MFEA-SM) of a photovoltaic cell simulation model based on multi-task optimization similarity metric of the present invention with the effects of five multi-task evolutionary algorithms MFEA, MFEA-II, EMEA, SREMTO, and MTEA-AD. In the experiment, each algorithm runs independently 20 times, the population size is set to 100*k, and the number of objective function evaluations for each task is set to 300000. The average value and standard deviation of the difference between the predicted values and the actual observed values of the PV models obtained by these algorithms are shown in Table 3, with the best results in bold, and the convergence diagrams are as Figure 5 shown, where Figure 5 (a) of Figure 5 (b) of Figure 5 (c) of

[0105] AsFigure 5 As shown, it can be seen that MTEA-DMSL achieved the best results in all three tasks. The results demonstrate that MTEA-DMSL has good performance in the performance optimization of photovoltaic cell simulation models, and prove that the proposed MTEA-DMSL has the ability to solve the problems of this practical application.

[0106] Table 3 Average objective values and standard deviations of six algorithms for 20 independent runs

[0107]

[0108] The present invention also provides a device for optimizing the performance of a photovoltaic system, including:

[0109] An initialization module, configured to construct multiple photovoltaic cell simulation models of the photovoltaic system, define each photovoltaic cell simulation model as an optimization task, and initialize a population and an incremental learning agent model for each optimization task;

[0110] A similarity calculation module, configured to use one optimization task among multiple optimization tasks as a target task, use other optimization tasks as source tasks, calculate the similarity between the target task and the incremental learning agent models corresponding to multiple source tasks by using the CKA index, and construct a similarity matrix according to the similarity between the target task and the incremental learning agent models corresponding to multiple source tasks;

[0111] A knowledge transfer module, configured to use the source task with the highest similarity to the target task in the similarity matrix as the knowledge transfer object. When the probability of the knowledge transfer object occurring is less than the knowledge transfer probability, perform inter-population evolution, use the incremental learning agent model corresponding to the source task with the highest similarity to the target task to guide global knowledge transfer in the target space, and use the multivariate Gaussian model to guide local knowledge transfer in the decision space;

[0112] A transfer ratio adjustment module, configured to dynamically adjust the transfer ratio according to the success rate of generating offspring by knowledge transfer;

[0113] A performance optimization module, configured to repeat the processes of constructing the similarity matrix, knowledge transfer, and dynamically adjusting the transfer ratio, continuously perform iteration until the number of evaluations of the objective function of the multi-task evolutionary algorithm reaches a set threshold, then the iteration ends, and an optimized photovoltaic cell simulation model is obtained.

[0114] The present invention also provides a computer device, including a memory and a processor; the memory stores a computer program, and the processor is configured to run the computer program in the memory to execute the photovoltaic system performance optimization method.

[0115] The present invention also provides a computer-readable storage medium storing a computer program, which is adapted to be loaded by a processor to execute the photovoltaic system performance optimization method.

[0116] The above embodiments are only preferred specific embodiments of the present invention, and the protection scope of the present invention is not limited thereto. Any simple changes or equivalent replacements of technical solutions that can be obviously obtained by those skilled in the art within the technical scope disclosed by the present invention all belong to the protection scope of the present invention.

Claims

1. A photovoltaic system performance optimization method, characterized in that: The steps include: Construct multiple photovoltaic cell simulation models of the photovoltaic system, define each photovoltaic cell simulation model as an optimization task, and initialize a population and an incremental learning agent model for each optimization task; One of the multiple optimization tasks is taken as the target task, and the other optimization tasks are taken as the source tasks. The CKA index is used to calculate the similarity between the incremental learning proxy models corresponding to the target task and the multiple source tasks, and a similarity matrix is ​​constructed according to the similarity between the incremental learning proxy models corresponding to the target task and the multiple source tasks. The source task with the highest similarity to the target task in the similarity matrix is ​​taken as the object of knowledge transfer. When the probability of knowledge transfer object migration is less than the knowledge transfer probability, inter-population evolution is performed. The incremental learning agent model corresponding to the source task with the highest similarity to the target task is used to guide global knowledge transfer in the target space, and the multivariate Gaussian model is used to guide local knowledge transfer in the decision space. Dynamically adjust the migration ratio based on the success rate of generating offspring through knowledge migration; The process of constructing the similarity matrix, transferring knowledge, and dynamically adjusting the migration ratio is repeated and iterated continuously until the number of evaluations of the objective function of the multi-task evolutionary algorithm reaches the set threshold. The iteration ends and the optimized photovoltaic cell simulation model is obtained.

2. The photovoltaic system performance optimization method according to claim 1, characterized in that: The method of calculating the similarity of the incremental learning agent model corresponding to the target task and the plurality of source tasks by using the CKA index comprises the following steps: Extract the similarity measure vector X from the population of the target task i ={x1,x2,…,x N }; X i Input the incremental learning proxy models corresponding to all source tasks respectively, calculate the hidden layer output feature matrix and the final output layer feature matrix of the incremental learning proxy model, and use the linear kernel function to convert the hidden layer output feature matrix and the output layer feature matrix into the corresponding kernel matrix; The kernel matrix is ​​centralized to remove the mean of the similarity measurement vector; The hidden layer CKA index and the output layer CKA index of the kernel matrix after centralization are calculated using the independence criterion, and the average value of the hidden layer CKA index and the output layer CKA index is calculated, and the average value is used as the CKA index of the target task.

3. The photovoltaic system performance optimization method according to claim 1, characterized in that: The method also includes performing intra-population evolution when the probability of knowledge transfer object migration is greater than or equal to the knowledge transfer probability.

4. The photovoltaic system performance optimization method according to claim 1, characterized in that: The incremental learning agent model is used to guide the global knowledge transfer in the target space, and the multivariate Gaussian model is used to guide the local knowledge transfer in the decision space. Specifically, a threshold of the transfer selection probability rmp is set. When the random number is less than the threshold of the transfer selection probability rmp, the incremental learning agent model is used to guide the global knowledge transfer in the target space; when the random number is greater than the threshold of the transfer selection probability rmp, the multivariate Gaussian model is used to guide the local knowledge transfer in the decision space.

5. A photovoltaic system performance optimization device, characterized in that: include: An initialization module is used to construct multiple photovoltaic cell simulation models of the photovoltaic system, define each photovoltaic cell simulation model as an optimization task, and initialize a population and an incremental learning agent model for each optimization task; A similarity calculation module is used to take one of the multiple optimization tasks as the target task and the other optimization tasks as the source tasks, calculate the similarity between the target task and the incremental learning proxy models corresponding to the multiple source tasks using the CKA index, and construct a similarity matrix based on the similarity between the target task and the incremental learning proxy models corresponding to the multiple source tasks; The knowledge transfer module is used to take the source task with the highest similarity to the target task in the similarity matrix as the knowledge transfer object. When the probability of the knowledge transfer object being transferred is less than the knowledge transfer probability, inter-population evolution is performed, and the incremental learning agent model corresponding to the source task with the highest similarity to the target task is used to guide the global knowledge transfer in the target space, and the multivariate Gaussian model is used to guide the local knowledge transfer in the decision space. The migration ratio adjustment module is used to dynamically adjust the migration ratio according to the success rate of knowledge migration to generate offspring; The performance optimization module is used to repeatedly construct the similarity matrix, transfer knowledge, and dynamically adjust the migration ratio. The iteration is continued until the number of objective function evaluations of the multi-task evolutionary algorithm reaches the set threshold. The iteration ends and the optimized photovoltaic cell simulation model is obtained.

6. A computer device, characterized in that: It comprises a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the photovoltaic system performance optimization method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the photovoltaic system performance optimization method according to any one of claims 1 to 4.

Images