A two-stage photovoltaic power output identification method, device and storage medium

Abstract

The present application relates to a kind of two-stage photovoltaic output identification method, equipment, storage medium, comprising the following steps: based on the historical load sample obtained before distributed photovoltaic access, WGAN model is iteratively trained;Based on the WGAN model after training, the net load sample after distributed photovoltaic access and context loss function, noise vector is iteratively optimized, generates actual load sample, completes rough identification;Based on actual load sample, select the user of similar night net load, calculate photovoltaic output difference, based on the relationship of photovoltaic output ratio and photovoltaic capacity same optimization target is constructed, noise vector is iteratively optimized again, obtain the final actual load sample including photovoltaic output information, complete fine identification.Compared with prior art, the present application does not need to rely on external environment data when identifying, also can reflect the actual power consumption behavior of user and the power consumption characteristics caused by environmental factors, improve the convenience of practical application.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of photovoltaic output prediction, in particular to a two-stage photovoltaic output identification method, device and storage medium. BACKGROUND

[0002] For a comprehensive energy microgrid system, due to the limited self-power load and the strong intermittency and randomness of photovoltaic power generation, the amount of power generation has a great influence on the safe and stable operation of the system, therefore, reliable photovoltaic power prediction plays an important role in the microgrid power system.

[0003] Chinese patent application publication No. CN115222138A discloses a microgrid photovoltaic short-term power interval prediction method based on EEMD-LSTM, including the following steps: S1, data input: input photovoltaic power generation related data, including meteorological factors, power generation system parameters, photovoltaic output and other information; S2, data preprocessing: integrating multiple source data, mainly multiple files or database data, for integration and summary processing to achieve analysis of data characteristics at different levels, and using KNN algorithm to detect missing data.

[0004] The above-mentioned application needs to consider the influence of environmental factors such as light and temperature, and environmental data is not easy to obtain, so its application scene is limited. Therefore, there is currently a lack of a method that can identify photovoltaic output without relying on external environmental data. SUMMARY

[0005] The purpose of the present application is to overcome the defects of the prior art and provide a two-stage photovoltaic output identification method, device and storage medium, which simplifies the implementation of photovoltaic output identification by reducing the required data.

[0006] The purpose of the present application can be achieved by the following technical solutions:

[0007] In one aspect of the present application, a two-stage photovoltaic output identification method is provided, including the following steps:

[0008] S1, based on the obtained historical load sample before the access of distributed photovoltaic, the WGAN model is iteratively trained;

[0009] S2, based on the trained WGAN model, the net load sample after the access of distributed photovoltaic and the context loss function, the noise vector is iteratively optimized to generate the actual load sample, and the rough identification is completed;

[0010] S3, based on the actual load sample, selecting users with similar night net load, calculating the photovoltaic output difference value, constructing an optimization target based on the relationship between the photovoltaic output ratio and the photovoltaic capacity, and iteratively optimizing the noise vector again to obtain the final actual load sample including photovoltaic output information, and completing fine identification.

[0011] As a preferred technical solution, the process of iteratively training the WGAN model specifically includes the following steps:

[0012] The noise vector is used as the input feature of the generator of the WGAN model, the historical load before the distributed photovoltaic access is used as the output label, and the optimization target is set based on the photovoltaic output and the actual load to optimize until the generator and the discriminator of the WGAN model reach balance.

[0013] As a preferred technical solution, the context loss function is constructed based on the difference between the net load sample after the distributed photovoltaic access and the sample generated by the generator of the WGAN model.

[0014] As a preferred technical solution, in S2, the context loss function is:

[0015]

[0016] wherein, is the context loss, represents the multiplication operation of the corresponding elements of the matrix, and respectively represent the net load sample of the generated sample and the test set, represents a binary mask matrix consistent with the dimension of the sample data, represents the modulus operation.

[0017] As a preferred technical solution, in S2 or S3, the process of iteratively optimizing the noise vector includes the following steps:

[0018] The noise vector is input into the generator of the trained WGAN model to obtain the generated sample;

[0019] The generated sample and the net load sample after the distributed photovoltaic access are input into the discriminator of the trained WGAN model to determine whether it meets the preset condition, if not, the noise vector is updated, if yes, the iteration is ended, the noise vector after the iteration is input into the generator, and the actual load sample is output.

[0020] As a preferred technical solution, in S3, the calculation of the loss function value in the iterative optimization process includes the following steps:

[0021] Using the actual load as noise, the relationship between the photovoltaic output ratio and the photovoltaic capacity is established by using the maximum information coefficient method, and the noise of the actual load is reduced by using variational mode decomposition. The fitting coefficients are solved by grid search.

[0022] The actual photovoltaic output value is calculated based on the fitting coefficients, and the loss function value is calculated based on the actual photovoltaic output value and the photovoltaic output value generated by the generator according to the noise vector.

[0023] As a preferred technical solution, the optimization objective in S3 is:

[0024]

[0025] in, This represents the photovoltaic output value generated by the generator based on the noise vector. Represents the fitting coefficient. This represents the difference in photovoltaic output after variational mode decomposition. Indicates net load. This indicates the modulo operation.

[0026] As a preferred technical solution, the K-nearest neighbor algorithm is used to select users with similar net load at night.

[0027] In another aspect, an electronic device is provided, comprising: one or more processors and a memory, wherein the memory stores one or more programs, the one or more programs including instructions for executing the above-described two-stage photovoltaic power output identification method.

[0028] In another aspect, the present invention provides a computer-readable storage medium including one or more programs executable by one or more processors of an electronic device, the one or more programs including instructions for performing the above-described two-stage photovoltaic power output identification method.

[0029] Compared with the prior art, the present invention has the following advantages:

[0030] (1) Reduce the data required for identification: Unlike some existing identification methods that require the prior collection of environmental data to assist in identification, which presents difficulties in environmental data collection, this application only needs to obtain historical load samples before distributed photovoltaic access and net load samples after access. In the first stage, a trained improved generative adversarial network model based on gradient penalty optimization (Wasserstein GAN with Gradient Penalty, WGAN) is selected to reconstruct the missing measurement data by solving the context-consistent generation problem, and to perform rough identification, which simplifies the analysis process of the user's actual load.

[0031] (2) Improve the convenience of practical application: Unlike some existing identification methods that require known net load and identified actual load data for fine identification, the present application performs fine identification based on the condition that the night photovoltaic output ratio is the same as the photovoltaic capacity in the second stage, and finally completes photovoltaic output identification to overcome the problem that photovoltaic fluctuations are large in the coarse identification stage and cannot accurately identify photovoltaic output. The present application does not need to rely on external environmental data when identifying, and can also reflect the actual power consumption behavior of users and the power consumption characteristics caused by environmental factors, improving the convenience of practical application. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 Flowchart of the two-stage photovoltaic output identification method of the embodiment;

[0033] Figure 2 Schematic diagram of the application process of the identification method of the embodiment;

[0034] Figure 3 Schematic diagram of the generation of the actual load sample of the user in the embodiment;

[0035] Figure 4 Actual power generation of three different capacities of photovoltaic for three days and normalized curve in the embodiment;

[0036] Figure 5 Comparison of the average MAE of each hour in 2012-2013 of four photovoltaic output decomposition algorithms in the embodiment;

[0037] Figure 6 MAE of each day in 2012 under different decomposition methods in the embodiment;

[0038] Figure 7 Decomposition results of a user in July of a week under different decomposition methods in the embodiment. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0040] Embodiment 1

[0041] In view of the problems of the prior art described above, referring to Figure 1 The present embodiment provides a two-stage photovoltaic output identification method, comprising the following steps:

[0042] S1, training the WGAN model using historical load data before the installation of distributed photovoltaic.

[0043] Firstly, the timestamps of the daytime stage, nighttime stage and whole day stage are defined, which satisfy the following relationship:

[0044]

[0045]

[0046] In the formula, is the start time of the daytime stage, is the end time of the daytime stage, represents the total number of data in the whole day. The user's net load, actual load and photovoltaic output satisfy the following relationship:

[0047]

[0048] In the formula, , and are the sets composed of the net load , actual load and photovoltaic output in the whole day time period. And the net load P is known, the values of the user's actual load C and user's photovoltaic output G need to be estimated.

[0049] In this embodiment, the improved generative adversarial network model based on gradient penalty optimization (Wasserstein GAN with Gradient Penalty, WGAN) is selected to reconstruct the missing data by solving the context consistent generation problem, so as to obtain the distribution rule of complex data and generate new data satisfying the distribution, such as the user's actual load C and user's photovoltaic output G in this method. WGAN uses the Wasserstein distance to measure the objective function, which can alleviate the problem of gradient disappearance in the training process compared with the traditional JS distance, and improve the training stability. The definition of the Wasserstein distance is as follows:

[0050]

[0051] In the formula, is the set of joint probability distribution and as the marginal distribution; is the expected lower bound, which means that needs to move to the distance.

[0052] ​​​​Since directly calculating the Wasserstein distance between arbitrary distributions is difficult, its Kantorovich-Rubinstein dual form is used:

[0053]

[0054]

[0055] In the formula, Indicate function It satisfies K-Lipschitz continuity, and the absolute value of its derivative is bounded above. The original objective function... writing , writing The optimization objective under the Wasserstein distance can then be obtained.

[0056] Feature inputs were selected, and the historical load before photovoltaic (PV) installation and the net load after PV installation were used as the training and testing sets, respectively. The input feature of the model generator was a noise vector, and the output label was the historical load data. The error between the PV output data identified by the model and the PV output data in the testing set was used as the evaluation index for the accuracy of PV output identification.

[0057] After training the model and building the WGAN model, train the model on the training set until the generator and discriminator reach equilibrium (i.e., reach the Nash equilibrium point). At the end of training, save the model parameters and then perform a rough identification on the actual load, fixing the network parameters of the generator and discriminator of the model.

[0058] S2. Based on the WGAN model, a rough identification result of photovoltaic output is obtained.

[0059] See Figure 2 The WGAN model is trained using historical load data prior to distributed photovoltaic installations. After training, optimization is needed among the generated samples that conform to certain patterns. This requires constructing the generated samples... Contextual constraints within a time period, and using these contextual constraints as a loss function on the noise vector. During training, the context constraints are defined as follows:

[0060] (1)

[0061] In the formula, This refers to the element-wise multiplication operation of a matrix. and respectively represent the sample generated by the generator and the net load sample of the test set, M is a binary mask matrix consistent with the dimension of the sample data, where the element is 1, indicating that the corresponding data needs to be extracted, otherwise the element is 0. The noise vector z is trained based on the context loss, and the iterative optimization process is as follows Figure 3 .

[0062] After the iteration, the z vector optimized by the generator is input, the actual sample is obtained, and then it is combined with the test set sample to obtain the rough identification result of the actual load:

[0063] (2)

[0064] According to the relationship between the net load, the actual load and the photovoltaic output, the rough identification result of the photovoltaic output can be obtained as follows:

[0065] (3)

[0066] After obtaining the result, the rough identification result is evaluated, and it is judged whether the requirement is met (i.e. whether the identification result of the actual load meets the corresponding true actual load distribution rule and characteristics), if it is met, the next step is entered, if it is not met, the learning rate and parameters of the algorithm are adjusted, and the previous steps are repeated.

[0067] S3, find users with similar electricity behavior by using K nearest neighbor algorithm (KNN), and subtract the actual demand of the users to obtain the photovoltaic output difference.

[0068] Find users with similar electricity behavior but different photovoltaic capacity by KNN algorithm, and subtract these users to eliminate the influence of actual load and obtain the photovoltaic output difference value of the user, i.e. photovoltaic factor.

[0069] When the users with similar actual load demand are obtained by KNN algorithm, in order to avoid the problem that the actual demand of each user is different, the actual load of the user needs to be scaled to the same order of magnitude. The feature scaling of the net load can be performed by dividing the average value of the night net load, as follows:

[0070] (4)

[0071] In the formula, the denominator is the average value of the night net load, represents the total number of time points at night.

[0072] In order to reduce the influence of the demand of each user, after the feature scaling of the same value order, the data between the scaled users is subtracted, and the influence of the actual load of the user is reduced, as follows:

[0073] (5)

[0074] can be simply expressed as:

[0075] (6)

[0076] S4, combined with the rough recognition results, using the method of maximum information coefficient (MIC), search for the best coefficient of the difference between the target user photovoltaic output and the inter-house photovoltaic output to solve the corrected photovoltaic output recognition result.

[0077] After obtaining the photovoltaic output difference, combined with the rough recognition result obtained, the method of maximum information coefficient is used for fine recognition, and is regarded as noise. If the user areas in the photovoltaic power supply area are similar, it can be considered that the photovoltaic installation method and power generation mode of the users in this area are also similar, so the photovoltaic power generation capacity and photovoltaic power generation capacity of the photovoltaic power generation capacity are strongly linearly related, such as Figure 4 The photovoltaic output curves of three adjacent users in the same power supply area in Australia for three consecutive days are shown in the figure, it can be seen that although the capacities of the users are different, the power generation modes are close to the same, so it can be concluded that the photovoltaic output ratio of users with similar geographical positions and the photovoltaic capacity ratio are basically the same.

[0078] The photovoltaic power generation capacity and capacity have a proportional relationship, and the linear relationship between the photovoltaic power generation capacity and capacity of user i and user j is expressed as:

[0079] (7)

[0080] In the formula, and are the photovoltaic capacities of user and user .

[0081] Therefore, also has a strong linear relationship with and , as shown in the formula:

[0082]

[0083] In order to further reduce the influence of , the variational mode decomposition (VMD) is applied to the denoising . Further reduce the constraint variational problem expression:

[0084] (8)

[0085] In the formula, is the number of modes that need to be decomposed, correspond to the first modal components and central frequencies, is the Dirac function, denotes the convolution operation.

[0086] The net load data of users is relatively low frequency and high frequency combination, so it is easy to extract from the net load. The extracted is a more refined photovoltaic factor. The photovoltaic factor obtained after VMD is denoted as . The photovoltaic power generation of each user is equal to the coefficient multiplied by the VMD , and the photovoltaic power generation of user is equal to:

[0087] (9)

[0088] In the formula, is solved by the grid search algorithm, and the metric solved by the grid search algorithm is:

[0089] (10)

[0090] By manually setting the lower limit , the upper limit and the discretization step , a detailed search is performed on the manually specified subset . If the correlation of and reaches the maximum value, it can be considered that the photovoltaic power generation obtained has a great correlation with the photovoltaic power generation estimated in the first stage, and if the correlation is determined, a detailed search is performed on the manually specified subset and then the photovoltaic power generation of the user is obtained according to (9) .

[0091] In the first stage of rough identification, only the night net load data is used to establish the context loss function, and in the second stage, the photovoltaic factor found is used to estimate the daytime output of the user's photovoltaic power station by using the proportional relationship between the photovoltaic output and the photovoltaic capacity, and on this basis, the actual load similarity loss in the daytime is obtained, and then it is superimposed with the context loss in the night, so as to realize more refined actual load identification. Therefore, in the second stage, the final optimization target of reconstructing the missing data of the actual load is defined as:

[0092] (11)

[0093] The subsequent steps are basically similar to the coarse identification steps. The noise vector is trained based on Equation (11), and then the optimized z is input into the generator to obtain the actual load sample. Then, the fine identification result of photovoltaic output is obtained based on the relationship between net load, actual load and photovoltaic output. (Including photovoltaic output information of multiple users in multiple time periods obtained from the identification)

[0094] To verify the effectiveness of this technical solution, the following embodiment is presented:

[0095] The data in this example comes from Ausgrid, the Australian power provider. To evaluate the performance and effectiveness of photovoltaic power output identification, the root mean square error was selected. Mean absolute error and power decomposition accuracy As evaluation indicators, each indicator is defined as follows:

[0096] (12)

[0097] (13)

[0098] (14)

[0099] In the formula, for The measured value of photovoltaic output at any given time. for The photovoltaic output identification value at any given time. This represents the total number of time series points. This reflects the error between the photovoltaic output identification results and the measured values. This reflects the accuracy of photovoltaic power output identification, therefore when and The smaller, The larger the value, the closer the photovoltaic output identification result is to the measured value, meaning the higher the accuracy of photovoltaic output identification.

[0100] This experiment compares the accuracy of distributed photovoltaic power output decomposition using two algorithms: unsupervised photovoltaic decomposition algorithm based on multi-constraint optimization (decomposition method 1) and unsupervised photovoltaic decomposition algorithm based on difference (decomposition method 2). The net load data of a user group with similar geographical locations is used in the data. Figure 5 The box plot compares the hourly MAE averages of four photovoltaic power output decomposition algorithms from 2012 to 2013. Overall, compared with decomposition methods 1 and 2, the present invention significantly improves the decomposition accuracy, and the decomposition accuracy of the coarse photovoltaic identification method is superior to that of decomposition method 1 to a certain extent. This verifies the feasibility of the proposed method indirectly identifying photovoltaic power output from the perspective of reconstructing missing actual daytime load data.

[0101] Figure 6 The MAE comparison results of the present application and decomposition method 1 and decomposition method 2 on each day in 2012 are shown. From the whole year of 2012, the present application has the best decomposition effect and error, and the MAE is basically maintained below 5%, while the present application has poor decomposition performance when the photovoltaic feature compensation is not considered in the first stage (rough identification), and the MAE exceeds 10% on a large number of dates.

[0102] Table 1 shows the annual daily RMSE, MAE and PDA of different decomposition methods

[0103]

[0104] Table 1 shows the average daily , and of different methods in 2012. It can be seen from the table that the present application has obvious improvement compared with method 1 and method 2, and greatly improves the accuracy of photovoltaic identification.

[0105] Figure 7 The photovoltaic output and actual load decomposition results of different decomposition methods for a user in a week in July are shown. From Figure 7 (a), it can be seen that the actual load curve generated by the present application basically overlaps with the true value, and the decomposition result is the closest. And the photovoltaic identification result of the present application is obtained according to the identified actual load result, so from Figure 7 (b), it can be seen that the photovoltaic output identification result of the present application is also the closest to the true value compared with other methods.

[0106] In summary, the present technical solution proposes a two-stage photovoltaic output identification method. First, the historical load data before photovoltaic access and the net load data after photovoltaic access are used as training set and test set respectively, and the WGAN model is constructed to be trained, and the context loss of the generated actual load data in the night period is constructed according to the test set net load data, and the noise vector is optimized through the context loss function After inputting the model, the generated actual load data is obtained, and then combined with the net load data to obtain a rough actual load identification result, and according to the relationship of P, C and G, the rough photovoltaic output identification result can be obtained. Further, the KNN algorithm is used to find users with similar power consumption behaviors but different capacities, and the net load between users is subtracted to obtain the photovoltaic output difference value, and then the maximum information coefficient method is used to search for the best coefficient of the photovoltaic output of the user and the photovoltaic output difference value between users. Finally, on this basis, the similarity loss of the generated sample is constructed, and the iterative optimization step of the rough photovoltaic output identification is repeated to obtain the fine photovoltaic output identification result based on the corrected rough identification result. The simulation results show that the two-stage photovoltaic output identification method constructed by the present application has obvious improvement in accuracy compared with the existing photovoltaic output identification means, and has certain engineering use value.

[0107] Compared with the prior art, the present application has the following advantages:

[0108] (1) The two-stage identification method is used for the photovoltaic output of the user in the present application. In the first stage (rough identification), the actual daytime load of the user is estimated and filled according to the nighttime net load data according to the characteristics of the user without photovoltaic output at night, which can better reflect the actual power consumption behavior characteristics of the user and simplify the analysis process of the actual load of the user.

[0109] (2) In the second stage (fine identification), the rough identification result is taken as the basis, and the photovoltaic output information implied in the net load is considered, which solves the problem of high-precision identification of photovoltaic output when the photovoltaic fluctuation is large in the rough identification stage. Compared with the existing photovoltaic power generation identification method, the influence of environmental factors such as light and temperature needs to be considered, and the environmental data is not easy to obtain. The present application does not need to rely on external environmental data, and can also reflect the actual power consumption behavior of the user and the power consumption characteristics caused by environmental factors, improving the convenience of practical application.

[0110] (3) In the second stage, compared with the partial scheme, the measured data of the photovoltaic reference station near the user is used to correct the rough identification result in the first stage. This step requires that the photovoltaic reference station needs to be in the same area as the target household photovoltaic, and relies on the measured data of the photovoltaic reference station. If the power generation mode of the target photovoltaic and the photovoltaic reference station is different, the identification effect will be obviously decreased. In view of this problem, the present application considers the photovoltaic feature information implied in the net load in the second stage fine identification, and obtains the photovoltaic output difference value between users by matching the actual power consumption behavior characteristics similar but the photovoltaic output different user net load sample difference. Then, the correlation between the photovoltaic output difference value and the photovoltaic estimated output value is used to correct the photovoltaic output identification result. Without additional acquisition of the measured data of the photovoltaic reference station, only the historical load data and the net load data of the user are needed, and the cost of acquiring external data is avoided. In addition, the household photovoltaic power generation modes of the users installed in the same area are similar, and the identification error caused by the difference in power generation mode can be ignored.

[0111] Embodiment 2

[0112] The embodiment provides an electronic device, including one or more processors and a memory, the memory has one or more programs stored therein, and the one or more programs include instructions for performing the two-stage photovoltaic output identification method as described in Embodiment 1.

[0113] Embodiment 3

[0114] The embodiment provides a computer readable storage medium, including one or more programs for an electronic device to execute, and the one or more programs include instructions for performing the two-stage photovoltaic output identification method as described in Embodiment 1.

[0115] The above merely provides a specific implementation of the present application, but the protection scope of the present application is not limited to this. Any skilled person in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A two-stage photovoltaic power output identification method, characterized in that, Includes the following steps: S1. Based on the historical load samples obtained before the distributed photovoltaic grid connection, the WGAN model is iteratively trained. S2, based on the trained WGAN model, the net load sample after distributed photovoltaic access and the context loss function, iteratively optimizes the noise vector to generate actual load samples and completes rough identification; S3. Based on the actual load sample, users with similar net nighttime loads are selected, and the photovoltaic output difference is calculated. An optimization objective is constructed based on the relationship between the photovoltaic output ratio and photovoltaic capacity. The noise vector is iteratively optimized again to obtain the final actual load sample including photovoltaic output information, thus completing the fine identification. In S2 or S3, the iterative optimization process for the noise vector includes the following steps: The noise vector is input into the generator of the trained WGAN model to obtain the generated samples; The generated samples and the net load samples after distributed photovoltaic grid connection are input into the discriminator of the trained WGAN model to determine whether they meet the preset conditions. If not, the noise vector is updated. If yes, the iteration ends, and the noise vector after iteration is input into the generator to output the actual load samples.

2. The two-stage photovoltaic power output identification method according to claim 1, characterized in that, The iterative training process of the WGAN model includes the following steps: The noise vector is used as the input feature of the generator of the WGAN model, and the historical load before distributed photovoltaic access is used as the output label. The optimization objective is set based on the photovoltaic output and the actual load to optimize the model until the generator and discriminator of the WGAN model reach a balance.

3. The two-stage photovoltaic power output identification method according to claim 1, characterized in that, The context loss function is constructed based on the difference between the net load sample after distributed photovoltaic access and the sample generated by the generator of the WGAN model.

4. The two-stage photovoltaic power output identification method according to claim 1, characterized in that, In S2, the context loss function is: in, For context loss, This represents the element-wise multiplication operation of a matrix. and These represent the samples generated by the generator and the net payload samples of the test set, respectively. This represents a binary mask matrix with the same dimensions as the sample data. This indicates the modulo operation.

5. The two-stage photovoltaic power output identification method according to claim 1, characterized in that, In S3, the calculation of the loss function value during the iterative optimization process includes the following steps: Using the actual load as noise, the relationship between the photovoltaic output ratio and the photovoltaic capacity is established by using the maximum information coefficient method, and the noise of the actual load is reduced by using variational mode decomposition. The fitting coefficients are solved by grid search. The actual photovoltaic output value is calculated based on the fitting coefficients, and the loss function value is calculated based on the actual photovoltaic output value and the photovoltaic output value generated by the generator according to the noise vector.

6. The two-stage photovoltaic power output identification method according to claim 1, characterized in that, In S3, the optimization objective is: in, This represents the photovoltaic output value generated by the generator based on the noise vector. Represents the fitting coefficient. This represents the difference in photovoltaic output after variational mode decomposition. Indicates net load. This indicates the modulo operation.

7. The two-stage photovoltaic power output identification method according to claim 1, characterized in that, The K-nearest neighbor algorithm is used to select users with similar net load at night.

8. An electronic device, characterized in that, include: One or more processors and a memory, wherein the memory stores one or more programs, the one or more programs including instructions for executing the two-stage photovoltaic power output identification method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, It includes one or more programs that are executed by one or more processors of an electronic device, the one or more programs including instructions for performing the two-stage photovoltaic output identification method as described in any one of claims 1-7.

Images