Photovoltaic module degradation rate prediction method based on reverse learning and computer equipment

Abstract

The invention belongs to the technical field of photovoltaic modules, and particularly relates to a photovoltaic module degradation rate prediction method based on reverse learning and computer equipment, and the method comprises the steps: inputting the degradation characteristic parameter data of a photovoltaic module before a to-be-predicted moment into a trained degradation rate prediction model, predicting to obtain a degradation rate of the photovoltaic module at a to-be-predicted moment; an initial value of a hyper-parameter in the degeneration rate prediction model is obtained through optimization by using a swarm intelligence algorithm, and a method for obtaining an initial population of the swarm intelligence algorithm comprises the following steps: selecting elite individuals in the initial population, and using a reverse learning strategy for each selected elite individual to obtain a reverse individual corresponding to each elite individual, comparing each elite individual with a corresponding reverse individual, and selecting individuals with better fitness from the elite individuals to be reserved in a reverse population; and obtaining an initial population according to the reverse population. And the quality of the initial population of the swarm intelligence algorithm is improved, so that the prediction precision and the optimization efficiency of the prediction model are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic components, and in particular relates to a photovoltaic component degradation rate prediction method based on reverse learning and a computer device. Background Art

[0002] The photoelectric conversion performance or output power of photovoltaic modules decays or degrades with the extension of service time. Accurately predicting and evaluating the field performance degradation or field life of photovoltaic modules is not only a requirement for the quality control of photovoltaic module design and manufacturing process, but also an important basis for investment and operation and maintenance decisions of photovoltaic power stations.

[0003] In order to predict the degradation rate of photovoltaic modules, the prior art usually uses the field performance degradation data of photovoltaic modules to establish a performance degradation rate prediction model to predict the degradation rate of photovoltaic modules. The selection of hyperparameters of the prediction model has a significant impact on the prediction effect of the model. In order to solve this problem, the prior art usually uses optimization algorithms such as the sparrow algorithm to optimize the hyperparameters of the prediction model. However, these optimization algorithms are prone to falling into the problem of local optimal solutions, which affects the optimization effect of the hyperparameters. When the selected hyperparameters are inappropriate, the prediction accuracy of the prediction model will be low.

[0004] In order to more reasonably optimize the hyperparameters in the prediction model, a Chinese invention patent application with application publication number CN113762078A proposes a technical solution for hyperparameter optimization using a chaotic sparrow algorithm. However, the initial population of the chaotic sparrow algorithm in this solution is directly generated using Tent chaotic perturbations, and the initial population is random. If the individual quality of the initial population is poor, it will not only affect the effect of hyperparameter optimization, but also affect the running time of the chaotic sparrow algorithm; thereby affecting the prediction accuracy and optimization efficiency of the prediction model. Summary of the invention

[0005] The purpose of the present invention is to provide a photovoltaic module degradation rate prediction method and computer equipment based on reverse learning, so as to solve the technical problems that the initial population quality of the hyperparameter optimization algorithm of the photovoltaic module degradation rate prediction model is poor, which ultimately leads to low prediction accuracy and low optimization efficiency of the prediction model.

[0006] In order to solve the above technical problems, the present invention provides a photovoltaic module degradation rate prediction method based on reverse learning, comprising: inputting degradation characteristic parameter data of the photovoltaic module within a set time period before the time to be predicted into a trained degradation rate prediction model, and predicting the degradation rate of the photovoltaic module at the time to be predicted;

[0007] The initial value of the hyperparameter in the degradation rate prediction model is obtained by optimizing the swarm intelligence algorithm, and the method for obtaining the initial population of the swarm intelligence algorithm includes:

[0008] Select elite individuals from the starting population, use the reverse learning strategy for each selected elite individual to obtain the reverse individuals corresponding to each elite individual, compare each elite individual with its corresponding reverse individual, and select individuals with better fitness to be retained in the reverse population; obtain the initial population based on the reverse population.

[0009] Furthermore, the process of obtaining the initial population according to the reverse population includes: using chaotic perturbation to perturb elite individuals to obtain new individuals, and retaining them in the chaotic population; taking individuals with better fitness in the reverse population and the chaotic population to generate the initial population.

[0010] Furthermore, the degradation characteristic parameter is a degradation characteristic parameter selected from the initially selected degradation characteristic parameters and having a higher correlation with the degradation rate.

[0011] Furthermore, the correlation is calculated using a two-dimensional Pearson correlation coefficient method.

[0012] Furthermore, the degradation characteristic parameters with a high correlation with the degradation rate include: the average temperature of the photovoltaic module, the maximum temperature difference, the relative humidity of the environment and the ultraviolet radiation. The maximum temperature difference is the temperature at the highest point of the photovoltaic module minus the temperature at the lowest point.

[0013] Furthermore, the swarm intelligence algorithm is an improved sparrow algorithm obtained by improving the traditional sparrow algorithm, and the improvement includes:

[0014] After the position of the sparrow is updated, the fitness of each sparrow and the average fitness of all sparrows are calculated first: if the fitness of a sparrow is less than the average fitness of all sparrows, it is mutated using Gaussian mutation, otherwise chaotic perturbation is added to perturb it;

[0015] All sparrows after mutation or disturbance were treated as contemporary populations for subsequent treatment.

[0016] Furthermore, the degradation rate prediction model is an LSTM neural network prediction model; the hidden layer parameters of the LSTM neural network prediction model are optimized using a swarm intelligence algorithm.

[0017] Furthermore, the historical degradation rate in the training data of the degradation rate prediction model is calculated as follows:

[0018] V=Λ(α 1 T+α 2 R H +α 3 E UV +α 4 ΔT)+Γ

[0019] Where: V represents the degradation rate of photovoltaic modules, Λ represents the constant of the influence coefficient of different geographical regions, α i represents the weight of each influencing factor variable, where i = 1, 2, 3, 4; Γ represents the fitting error, T is the average temperature of the photovoltaic module, R H is the relative humidity of the environment where the photovoltaic module is located, E UV The amount of ultraviolet radiation in the environment where the photovoltaic module is located, △T is the maximum temperature difference of the photovoltaic module.

[0020] Furthermore, the process of taking individuals with better fitness from the reverse population and the chaotic population to generate the initial population includes: taking a first set number of individuals with better fitness from the reverse population and the chaotic population to generate the initial population;

[0021] The process of selecting elite individuals from the initial population includes: selecting a second set number of individuals with higher fitness from the initial population as elite individuals, and the second set number is equal to the first set number.

[0022] Furthermore, the chaotic disturbance is a Tent chaotic disturbance.

[0023] The present invention is an improved invention creation, and its beneficial effects are as follows: the photovoltaic module degradation rate prediction method based on reverse learning of the present invention uses a reverse learning strategy to process elite individuals to obtain a reverse population, and obtains the initial population of the swarm intelligence algorithm based on the reverse population. Different from the method of reverse learning in the prior art to reversely solve and optimize poor individuals, the present invention selects the best from the elite individuals through reverse learning to obtain a better reverse population. The reverse population not only does not contain poor individuals in the entire population, but is also better than the selected elite individuals, greatly improving the quality of the initial population, being able to obtain a better solution faster, and improving the effect of hyperparameter optimization, thereby improving the prediction accuracy and optimization efficiency of the photovoltaic module degradation rate prediction model.

[0024] To solve the above technical problems, the present invention further provides a computer device, comprising a processor, which implements the steps of the photovoltaic component degradation rate prediction method based on reverse learning of the present invention when executing a computer program.

[0025] The present invention is an improved invention creation, and its beneficial effects are as follows: the computer device of the present invention, when executing the computer program, uses the reverse learning strategy to process the elite individuals to obtain the reverse population, and obtains the initial population of the swarm intelligence algorithm based on the reverse population. Different from the method of reverse learning in the prior art to reversely solve and optimize the poor individuals, the present invention selects the best from the elite individuals through reverse learning to obtain a better reverse population. The reverse population not only does not contain the poor individuals in the entire population, but is also better than the selected elite individuals, greatly improving the quality of the initial population, being able to obtain a better solution faster, and improving the effect of hyperparameter optimization, thereby improving the prediction accuracy and optimization efficiency of the photovoltaic module degradation rate prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flow chart of a photovoltaic module degradation rate prediction method based on a reverse optimization algorithm according to a method embodiment of the present invention;

[0027] Figure 2 is a schematic diagram of the degradation rate prediction model structure of the method embodiment of the present invention;

[0028] Figure 3 It is a hyperparameter optimization flow chart of the method embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention discloses a photovoltaic module degradation rate prediction method and computer device based on reverse learning, which uses a reverse learning strategy to process elite individuals to obtain a reverse population, and obtains an initial population of a swarm intelligence algorithm based on the reverse population. Different from the prior art method of reverse learning to perform reverse solution optimization on poor individuals, the present invention selects the best from the elite individuals through reverse learning to obtain a better reverse population, which not only does not contain poor individuals in the entire population, but is also better than the selected elite individuals, greatly improving the quality of the initial population, being able to obtain a better solution faster, and improving the effect of hyperparameter optimization, thereby improving the prediction accuracy and optimization efficiency of the photovoltaic module degradation rate prediction model.

[0030] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0031] Method Example:

[0032] The photovoltaic module degradation rate prediction method based on the reverse optimization algorithm of the present invention is as follows Figure 1 As shown, the following steps are included:

[0033] Step 1: Determine the input and output of the degradation rate prediction model, and generate training data for the degradation rate prediction model.

[0034] 1. Select the initial degradation characteristic parameters.

[0035] According to the current research status, the factors that affect the performance degradation of photovoltaic modules include temperature, relative humidity, wind speed, latitude, maximum temperature difference, ultraviolet radiation, etc. Based on these quantitative relationship analyses, the comprehensive factors that affect the performance degradation of photovoltaic modules and their influence weights can be determined, and then the combined prediction characteristics of the performance degradation rate of photovoltaic modules can be constructed. Therefore, these parameters are used as initial degradation characteristic parameters in this embodiment.

[0036] 2. Select degradation feature parameters with higher correlation.

[0037] In this embodiment, the two-dimensional Pearson correlation coefficient method is used to calculate the correlation between each degradation characteristic parameter and the degradation rate of the photovoltaic module, and the degradation parameter with a higher correlation is selected according to the calculated correlation. The specific calculation formula of the two-dimensional Pearson correlation coefficient method is:

[0038]

[0039] and K are random vectors φ, The average value and dimension of the elements in φ k , are random vectors φ, The value of the kth element in is a random vector φ, When the correlation coefficient is in the interval [0.8,1) it is very strongly correlated, [0.6,0.8) is strongly correlated, [0.4,0.6) is moderately correlated, and [0,0.4) is weakly correlated / uncorrelated. Through the above calculation method, the photovoltaic module temperature T and relative humidity R are finally selected. H 、Ultraviolet radiation E UV The maximum temperature difference △T is taken as the degradation characteristic parameter. The maximum temperature difference △T is equal to the highest temperature of the photovoltaic module minus the lowest temperature.

[0040] 3. Determine the input and output of the degradation rate prediction model.

[0041] The selected photovoltaic module temperature T and relative humidity R H 、Ultraviolet radiation E UV The maximum temperature difference △T is used as degradation characteristics and input into the degradation rate prediction model to predict the degradation rate of PV modules.

[0042] Select the degradation characteristic parameter J with a time length T before the predicted moment T (J T =[j 0 ,j 0 ,......,jt ]) is used as the input variable of the degradation feature prediction model, and the output is the degradation rate Y of the photovoltaic module with the prediction time T+1 .J i is the combined degradation characteristics at time i, including the PV module temperature T, relative humidity R H 、Ultraviolet radiation E UV and maximum temperature difference △T.

[0043] 4. Generate a training dataset for the degradation feature prediction model.

[0044] The degradation feature prediction model is trained using the historical degradation features and the corresponding degradation rates as training data sets. The calculation method of the degradation rate in the training data in this embodiment is as follows:

[0045] V=Λ(α 1 T+α 2 R H +α 3 E UV +α 4 ΔT)+Γ

[0046] Where: V represents the degradation rate of photovoltaic modules, Λ represents the constant of the influence coefficient of different geographical regions, α i represents the weight of each influencing factor variable, where i=1, 2, 3, 4; Γ represents the fitting error.

[0047] In this paper, we use the T, R H 、E UV The degradation model of △T and the accelerated thermal cycle test model are used to determine the weight α of each influencing factor variable. i The degradation model is:

[0048]

[0049] Where k D Indicated by T, R H 、E UV A is the degradation rate of the photovoltaic module; D , T, n represent different fitting parameters; E D Indicates that in T, R H 、E UV Activation energy under action; k B represents the Boltzmann constant.

[0050] The thermal cycle model is:

[0051]

[0052] Where k TC A represents the component degradation rate with △T as the variable; TC, θ represent fitting parameters; C N Indicates the number of cycles; E TC Represents the activation energy under the action of △T. The weights of each degradation feature in the main degradation feature are determined according to the two cycle models. The specific weight determination method is a prior art, and reference may be made to the paper "Modeling Outdoor Service Lisetime Prediction Of TV Modusts". This embodiment will not be repeated here.

[0053] The above method can be used to calculate the degradation rate of photovoltaic modules at each moment using the combined degradation features at each moment in the historical data. Then the degradation feature prediction model can be trained using the historical data.

[0054] Calculate the corresponding arithmetic mean value according to the degradation rate V of the photovoltaic module and residual error The corresponding standard deviation is further obtained. Then, the Dayla criterion is used to eliminate the points with large errors. When the residual error is greater than 3σ, the data is eliminated. Finally, the screened samples are processed to obtain the final training data set.

[0055] The training dataset is divided into training set and test set.

[0056] Selecting variables with greater correlation can effectively reduce the data input dimension, relieve training pressure, and improve prediction speed. At the same time, it can prevent the prediction error from gradually increasing over time, thereby jumping out of the local optimum and improving the accuracy of the prediction algorithm.

[0057] Step 2: Construct a degradation feature prediction model.

[0058] 1. Selection of prediction model.

[0059] In this embodiment, an LSTM neural network is used as a degradation feature prediction model. The input of the model is the combined degradation feature, and the output is the predicted degradation rate of the photovoltaic module. The hyperparameters and hidden layer parameters to be optimized are selected, and the value ranges of each hyperparameter and hidden layer parameter are determined. The structure of the degradation feature prediction model in this embodiment is as follows: Figure 2 The model construction process is as follows Figure 3 shown.

[0060] Select the main degradation characteristic parameter J with a time length T before the prediction moment T (J T =[j 0 ,j 0 ,......,j t ]) is used as the input variable of LSTM, and the output is the degradation rate Y of the photovoltaic module with the predicted time. T+1 . Time T+1 is the time to be predicted.

[0061] 2. Use the reverse learning algorithm and chaotic perturbation to obtain the mixed population.

[0062] The present invention first selects individuals with higher fitness from the starting population as elite individuals, uses a reverse learning strategy for each selected elite individual to obtain the reverse individuals corresponding to each elite individual, compares each elite individual with its corresponding reverse individual solution, selects individuals with better fitness and retains them in the reverse population; obtains the initial population based on the reverse population. The starting population is a population obtained by randomly setting multiple individuals within the range of hyperparameter values.

[0063] In this embodiment, a second set number of individuals with higher fitness are selected from the starting population as elite individuals, and the elite solutions are processed using the reverse learning algorithm and the chaotic perturbation respectively to obtain a reverse population and a chaotic population, and a first set number of solutions with higher fitness are selected from the reverse population and the chaotic population to form a mixed population. The first set number is less than or equal to twice the second set number.

[0064] The basic idea of ​​Opposition-based-Learning (OBL) is to find the corresponding reverse solution based on the current solution through the reverse learning mechanism, and then evaluate and compare to save the better solution.

[0065] The introduction of reverse solutions can expand the search area of ​​the algorithm, but reverse learning has a certain degree of blindness, and the search space where the reverse solution is located may not be more conducive to the search space of the current solution. For example, for those individuals whose fitness values ​​of the original solution are greater than the fitness values ​​of the reverse solution, reverse search is a waste of search time, and their original field search should be strengthened. For individuals whose fitness values ​​of the original solution are less than the fitness values ​​of the reverse solution, the value of reverse search is higher than the development value of their field. In view of this situation, an elite strategy is added, elite individuals are introduced, reverse learning is carried out through elite individuals, and effective information of elite individuals is fully utilized to generate elite reverse solutions, guide the search process to approach the optimal solution, and select excellent individuals from the current solution and elite reverse solutions as the next generation of population elite reverse learning.

[0066] In this embodiment, the second set number is 10% of the number of initial values ​​of the hyperparameters. The dynamic boundary of the elite solution is obtained, and the reverse solution is solved using the reverse learning strategy. The sparrows before and after the update are compared. If they are better, the previous sparrows are replaced, otherwise the previous elite sparrows are retained.

[0067] Let x i (t) is defined as a solution at the tth iteration, then the reverse solution is x i (t)′, f(x) is the objective function. i (t))≥f((x i(t)')), then x i (t) is the elite individual of the tth iteration, which is retained to the reverse population. i (t))<f((x i (t)')), then x i (t)′ is the elite individual of the tth iteration, which is retained to the reverse population. Let x ij For an ordinary individual x i The value in the jth dimension, then its reverse solution is:

[0068] x ij (t)′=m(l ij (t)+u ij (t))-x ij (t)

[0069] Where: m is a random number between 0 and 1, called the elite reverse coefficient. ij (t) and u ij (t) is the minimum and maximum value of x(t)' in the jth dimension. ij (t),u ij (t)] is the interval constructed by the elite group.

[0070] In this embodiment, the chaotic perturbation adopts Tent chaotic perturbation, and the population composed of elite solutions is processed using Tent mapping to obtain a chaotic population. The Tent mapping expression is shown in the following formula:

[0071]

[0072] Where: rand(0,1) is a random number between 0 and 1, N Γ is the number of particles in the chaotic sequence, z i represents the i-th value of the chaotic map, z i+1 It represents the i+1th value of the chaotic map, and mod is the remainder operator.

[0073] The reverse population and chaotic population are obtained in the above manner, and the numbers of the reverse population and the chaotic population are both the second set number, so the number of the generated mixed population (the first set number) should be less than or equal to the sum of the numbers of the reverse population and the chaotic population, that is, the first set number is less than or equal to twice the second set number. In the present embodiment, the second set number is equal to the first set number. As other implementation methods, when selecting elite solutions from the starting population, and selecting those with higher fitness from the reverse population and the chaotic population and retaining them in the initial population, it is also possible not to select according to the set number, for example, all individuals with fitness higher than a threshold are selected.

[0074] 3. Use the mixed population as the initial population of the swarm intelligence algorithm to optimize the hyperparameters and hidden layer parameters.

[0075] The swarm intelligence algorithm can be a particle swarm algorithm, a wolf colony algorithm, an ant colony algorithm, or other swarm intelligence algorithms. In this embodiment, the swarm intelligence algorithm specifically selects the chaotic sparrow algorithm (CSSA). Select the degradation rate T i The mean absolute error function is the fitness function. i is the prediction result of the degradation rate of photovoltaic modules at time i. During the iteration process, the chaotic sparrow algorithm calculates the fitness value f of each sparrow. i And the average fitness value of the population f avg , when f i <f avg When Gaussian mutation is used to update the sparrow position, f i ≥f avg When the chaotic disturbance is added, the position of the sparrow is updated, and all the sparrows after mutation or disturbance are regarded as the contemporary population, and the steps of the traditional sparrow algorithm are continued. When the algorithm end condition is met, the optimization result of the hyperparameter is output. In this embodiment, the chaotic disturbance is Tent chaotic disturbance.

[0076] Tent chaotic perturbation has been introduced above, and Gaussian mutation is introduced below.

[0077] Gaussian mutation comes from Gaussian distribution. It calculates the new position by adding a random number that meets the standard normal distribution conditions to the original position, thus overcoming the disadvantage of the algorithm falling into the local optimal area and improving the robustness of the algorithm. Its formula is shown below:

[0078] mutation(χ)=χ(1+N(0,1))

[0079] Where: χ is the original position; N(0,1) is a standard normal distribution; mutation(χ) is the position after Gaussian mutation.

[0080] The chaotic sparrow algorithm is used to iterate and optimize until the algorithm termination condition is met, and the optimization result of the hyperparameter is output. In this embodiment, the algorithm termination condition is that the maximum number of iterations is reached.

[0081] Use the optimization results of hyperparameters and hidden layer parameters to update the hyperparameters and hidden layer parameters in the degradation feature prediction model. Use the obtained training set to train it and update the weights. After training, use the test set for testing.

[0082] Step 3: Input the degradation characteristics of the set time period before the time to be predicted obtained in step 1 into the trained degradation characteristic prediction model to obtain the degradation rate of the photovoltaic module at the time to be predicted.

[0083] Through the above steps, the photovoltaic module degradation rate prediction method based on reverse learning of the present invention uses the reverse learning strategy to process the elite individuals to obtain the reverse population, and obtains the initial population of the swarm intelligence algorithm based on the reverse population. Different from the method of reverse learning in the prior art for reverse solution optimization of poor individuals, the present invention selects the best from the elite individuals through reverse learning to obtain a better reverse population. The reverse population not only does not have the poor individuals in the entire population, but is also better than the selected elite individuals, greatly improving the quality of the initial population, and can obtain a better solution faster, improving the effect of hyperparameter and hidden layer parameter optimization, thereby improving the accuracy of photovoltaic module degradation rate prediction and the optimization efficiency of the model. Further, the present invention uses chaotic perturbation and directional learning strategies at the same time to obtain the initial population of the swarm intelligence algorithm, expands the solvable space of the initial population, and can effectively prevent the swarm intelligence algorithm from falling into the local optimal solution. The prediction accuracy and optimization efficiency of the photovoltaic module degradation rate prediction model are further improved.

[0084] Computer equipment example:

[0085] A computer device of the present invention includes a processor, wherein the processor implements the steps of the method described in the method embodiment when executing a computer program. The specific process, principle and beneficial effects of the method have been described in detail in the method embodiment, and will not be repeated in this embodiment.

[0086] The processor may be a microprocessor MCU, a programmable logic device FPGA or other processing device.

Claims

1. A photovoltaic module degradation rate prediction method based on reverse learning, characterized in that: include: Inputting the degradation characteristic parameter data of the photovoltaic module within a set time period before the time to be predicted into the trained degradation rate prediction model to predict the degradation rate of the photovoltaic module at the time to be predicted; The initial value of the hyperparameter in the degradation rate prediction model is obtained by optimizing the swarm intelligence algorithm, and the method for obtaining the initial population of the swarm intelligence algorithm includes: Select elite individuals from the starting population, use the reverse learning strategy for each selected elite individual to obtain the reverse individuals corresponding to each elite individual, compare each elite individual with its corresponding reverse individual, and select individuals with better fitness to be retained in the reverse population; The initial population is obtained according to the reverse population.

2. The photovoltaic module degradation rate prediction method based on reverse learning according to claim 1 is characterized in that: The process of obtaining the initial population based on the reverse population includes: using chaotic perturbation to perturb the elite individuals to obtain new individuals and retain them in the chaotic population; taking individuals with better fitness in the reverse population and the chaotic population to generate the initial population.

3. The photovoltaic module degradation rate prediction method based on reverse learning according to claim 1 is characterized in that: The degradation characteristic parameter is a degradation characteristic parameter selected from the initially selected degradation characteristic parameters and having a higher correlation with the degradation rate.

4. The photovoltaic module degradation rate prediction method based on reverse learning according to claim 3 is characterized in that: The correlations were calculated using the two-way Pearson correlation coefficient method.

5. The photovoltaic module degradation rate prediction method based on reverse learning according to claim 3 is characterized in that: The degradation characteristic parameters with a high correlation with the degradation rate include: the average temperature of the photovoltaic module, the maximum temperature difference, the relative humidity of the environment and the ultraviolet radiation. The maximum temperature difference is the temperature at the highest point of the photovoltaic module minus the temperature at the lowest point.

6. The photovoltaic module degradation rate prediction method based on reverse learning according to claim 1 is characterized in that: The swarm intelligence algorithm is an improved sparrow algorithm obtained by improving the traditional sparrow algorithm, and the improvement includes: After the position of the sparrow is updated, the fitness of each sparrow and the average fitness of all sparrows are calculated first: if the fitness of a sparrow is less than the average fitness of all sparrows, it is mutated using Gaussian mutation, otherwise chaotic perturbation is added to perturb it; All sparrows after mutation or disturbance were treated as contemporary populations for subsequent treatment.

7. The photovoltaic module degradation rate prediction method based on reverse learning according to claim 1 is characterized in that: The degradation rate prediction model is an LSTM neural network prediction model; the hidden layer parameters of the LSTM neural network prediction model are optimized using a swarm intelligence algorithm.

8. The photovoltaic module degradation rate prediction method based on reverse learning according to claim 1, characterized in that: The historical degradation rate in the training data of the degradation rate prediction model is calculated as follows: V=Λ(α1T+α2R H +α3E UV +α4ΔT)+Γ Where: V represents the degradation rate of photovoltaic modules, Λ represents the constant of the influence coefficient of different geographical regions, α i represents the weight of each influencing factor variable, where i = 1, 2, 3, 4; Γ represents the fitting error, T is the average temperature of the photovoltaic module, R H is the relative humidity of the environment where the photovoltaic module is located, E UV The amount of ultraviolet radiation in the environment where the photovoltaic module is located, △T is the maximum temperature difference of the photovoltaic module.

9. The photovoltaic module degradation rate prediction method based on reverse learning according to claim 2 is characterized in that: The process of taking individuals with better fitness from the reverse population and the chaotic population to generate the initial population includes: taking a first set number of individuals with better fitness from the reverse population and the chaotic population to generate the initial population; The process of selecting elite individuals from the initial population includes: selecting a second set number of individuals with higher fitness from the initial population as elite individuals, and the second set number is equal to the first set number.

10. A computer device comprising a processor, characterized in that: The processor implements the steps of the photovoltaic component degradation rate prediction method based on reverse learning as described in any one of claims 1 to 9 when executing the computer program.

Images