A photovoltaic power output typical scene acquisition method, device, equipment and storage medium

Abstract

The application discloses a kind of photovoltaic output typical scene acquisition method, device, equipment and storage medium.The method includes obtaining the generative adversarial network based on the principle of mutual information maximization of pre-training, and obtains the meteorological characteristic data at the photovoltaic region to be accessed;The meteorological characteristic data is input into the generator of the generative adversarial network based on the principle of mutual information maximization, and obtains candidate photovoltaic output scene set;Using convolutional autoencoder and embedded clustering are combined, the candidate photovoltaic output scene set is reduced, and the photovoltaic output typical scene matched with the photovoltaic region to be accessed is obtained.The technical scheme of the application provides a kind of photovoltaic output typical scene acquisition method, improves the security and stability of traditional power grid operation after accessing photovoltaic power station.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution network, and particularly relates to a photovoltaic output typical scene acquisition method, device, equipment and storage medium. BACKGROUND

[0002] The rapid development of photovoltaic power generation technology makes solar energy resources fully play its advantages of sufficient reserves and green and clean, and is widely used in the field of distributed energy utilization. Under such industry background, whether the traditional power grid connected with photovoltaic power station can guarantee the safety and stability of its operation is a problem that power system planners pay close attention to. In this regard, it is necessary to pre-identify problems and evaluate the carrying capacity of the power distribution network with photovoltaic access. An important prerequisite for completing the above power distribution network operation and planning work is to need a certain number of photovoltaic output scenes with certain rationality. However, the traditional power distribution network often has no or lacks complete photovoltaic output data, which brings challenges to the operation and planning of the power distribution network with high penetration rate of photovoltaic.

[0003] There is a natural strong correlation between the photovoltaic and other renewable energy power generation output and the geographical environment and solar radiation. Under the same geographical environment, different numerical values of solar radiation can correspond to photovoltaic output scenes with specific characteristics. Therefore, how to fully utilize these solar radiation meteorological factors to effectively generate photovoltaic output scenes with rationality has important value and practical significance for the operation and planning of the power distribution network. SUMMARY

[0004] The present application provides a photovoltaic output typical scene acquisition method, device, equipment and storage medium, to provide a photovoltaic output typical scene acquisition method, and improve the safety and stability of the operation of the traditional power grid connected with photovoltaic power station.

[0005] According to an aspect of the present application, a photovoltaic output typical scene acquisition method is provided, which comprises:

[0006] acquiring a pre-trained target generative adversarial network based on the principle of mutual information maximization, and acquiring meteorological feature data at a photovoltaic region to be accessed;

[0007] inputting the meteorological feature data into a generator of the target generative adversarial network based on the principle of mutual information maximization, to acquire a candidate photovoltaic output scene set;

[0008] combining a convolutional autoencoder and embedded clustering to reduce the candidate photovoltaic output scene set, to acquire a photovoltaic output typical scene matched with the photovoltaic region to be accessed.

[0009] According to another aspect of the present application, a photovoltaic output typical scene acquisition device is provided, which comprises:

[0010] a network and feature data obtaining module, configured to obtain a pre-trained target generative adversarial network based on a principle of mutual information maximization, and obtain meteorological feature data at a photovoltaic region to be accessed;

[0011] a candidate photovoltaic output scene set obtaining module, configured to input the meteorological feature data into a generator of the target generative adversarial network based on the principle of mutual information maximization, and obtain a candidate photovoltaic output scene set;

[0012] a photovoltaic output typical scene obtaining module, configured to reduce the candidate photovoltaic output scene set by using a convolutional autoencoder combined with embedded clustering, and obtain a photovoltaic output typical scene matched with the photovoltaic region to be accessed.

[0013] According to another aspect of the present application, an electronic device is provided, the electronic device comprising:

[0014] at least one processor; and

[0015] a memory connected to the at least one processor in communication; wherein,

[0016] the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the photovoltaic output typical scene obtaining method according to any one of the embodiments of the present application.

[0017] According to another aspect of the present application, a computer readable storage medium is provided, the computer readable storage medium stores computer instructions for enabling a processor to implement the photovoltaic output typical scene obtaining method according to any one of the embodiments of the present application when executed by the processor.

[0018] The technical scheme of the embodiments of the present application, by obtaining a pre-trained target generative adversarial network based on a principle of mutual information maximization, and obtaining meteorological feature data at a photovoltaic region to be accessed; inputting the meteorological feature data into a generator of the target generative adversarial network based on the principle of mutual information maximization, and obtaining a candidate photovoltaic output scene set; reducing the candidate photovoltaic output scene set by using a convolutional autoencoder combined with embedded clustering, and obtaining a photovoltaic output typical scene matched with the photovoltaic region to be accessed, provides a photovoltaic output typical scene obtaining method, and improves the safety and stability of the operation of a traditional power grid after a photovoltaic power station is accessed.

[0019] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0021] Figure 1 A flow chart of a photovoltaic output typical scene acquisition method provided for the first embodiment of the present application;

[0022] Figure 2a A flow chart of another photovoltaic output typical scene acquisition method provided for the second embodiment of the present application;

[0023] Figure 2b A comparison diagram between a generated scene and a real scene in the example of applying the photovoltaic output typical scene acquisition method provided for the second embodiment of the present application when iteration is performed 100 times;

[0024] Figure 2c A comparison diagram between a generated scene and a real scene in the example of applying the photovoltaic output typical scene acquisition method provided for the second embodiment of the present application when iteration is performed 200 times;

[0025] Figure 2d A comparison between a generated scene and a real scene in the example of applying the photovoltaic output typical scene acquisition method provided for the second embodiment of the present application when iteration is performed 300 times;

[0026] Figure 2e A comparison between a generated scene and a real scene in the example of applying the photovoltaic output typical scene acquisition method provided for the second embodiment of the present application when iteration is performed 400 times;

[0027] Figure 2f A comparison diagram between a generated scene and a real scene in the example of applying the photovoltaic output typical scene acquisition method provided for the second embodiment of the present application when iteration is performed 500 times;

[0028] Figure 2g A comparison diagram between a generated photovoltaic output scene X' and a real scene X in the example of applying the photovoltaic output typical scene acquisition method provided for the second embodiment of the present application when iteration is performed 500 times; test test A comparison diagram between a generated photovoltaic output scene X' and a real scene X in the example of applying the photovoltaic output typical scene acquisition method provided for the second embodiment of the present application when iteration is performed 500 times;

[0029] Figure 2h A generated photovoltaic output scene in a training batch in the example of applying the photovoltaic output typical scene acquisition method provided for the second embodiment of the present application;

[0030] Figure 2i ​A typical photovoltaic output scene diagram after cutting the photovoltaic output scene set in the example of the application of the photovoltaic output typical scene acquisition method provided for the second embodiment of the present application;

[0031] Figure 3 A structural schematic diagram of a photovoltaic output typical scene acquisition device provided for the third embodiment of the present application;

[0032] Figure 4 A structural schematic diagram of an electronic device for implementing the photovoltaic output typical scene acquisition method of the embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the present application, 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 only a 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 efforts should fall within the scope of the present application.

[0034] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product or device.

[0035] Embodiment one

[0036] Figure 1 A flowchart of a photovoltaic output typical scene acquisition method provided for the first embodiment of the present application. The present embodiment can be applicable to the case of constructing a photovoltaic output typical scene for a photovoltaic region not connected. The method can be executed by a photovoltaic output typical scene acquisition device, which can be realized in the form of hardware and / or software, and can be configured in a server or a server cluster. As shown in the figure, the method comprises: Figure 1

[0037] S110, acquiring a pre-trained target generative adversarial network based on the principle of mutual information maximization, and acquiring meteorological feature data at a photovoltaic region to be connected. ​

[0038] The target generative adversarial network (GAN) based on the principle of maximizing mutual information can be trained using photovoltaic (PV) output data and corresponding meteorological data collected from PV-connected areas. The areas to be connected to PV are those not yet connected. Meteorological data could include, for example, solar radiation data.

[0039] In this embodiment, it is necessary to pre-train a generative adversarial network based on the mutual information maximization principle of the target, and then use the trained generative adversarial network based on the mutual information maximization principle of the target to perform the subsequent S120 and S130 operations for obtaining typical photovoltaic power output scenarios.

[0040] In one optional implementation, obtaining the pre-trained target generative adversarial network based on the mutual information maximization principle may include: acquiring raw photovoltaic power output data and raw meteorological data from the connected photovoltaic area; filtering out missing values ​​in the raw photovoltaic power output data and raw meteorological data to obtain photovoltaic power output data and meteorological data after the first processing; normalizing the photovoltaic power output data and meteorological data after the first processing using the maximum value normalization principle to obtain photovoltaic power output sample data and meteorological sample data; and training the original generative adversarial network based on the mutual information maximization principle for at least one round using the photovoltaic power output sample data and meteorological sample data to obtain the target generative adversarial network.

[0041] For example, the following provides a specific method for obtaining photovoltaic power output sample data and meteorological sample data.

[0042] Step 1: Data Acquisition: Obtain raw photovoltaic (PV) output data and corresponding meteorological data from PV power plants in the connected PV area. Both the PV output data and the meteorological data are time series in yearly units. The raw PV output data collected in one day can be denoted as x. i =[x1,x2,…,x n The raw meteorological data can be denoted as y. ji =[y j1 ,y j2 ,…,y jn ], where n represents the length of the time series, which is 144 here (i.e., one sampling point every 10 minutes, and 144 points are collected every day); j represents the number of meteorological features, which may include solar radiation, weather temperature, precipitation and wind speed, etc.

[0043] Step 2: Data cleaning and processing: Remove missing values ​​from the original photovoltaic power output data and meteorological data on a daily basis.

[0044] Step 3, data normalization processing: according to the maximum value normalization principle, the photovoltaic output original data and the meteorological original data are normalized to reduce the calculation error caused by the data dimension, so as to obtain the photovoltaic output sample data X o and the meteorological sample data Y o . Wherein: In the formula: x max is the maximum value of the photovoltaic output original data of the year, y max is the maximum value of the meteorological original data of the year.

[0045] Based on the above optional implementation, using the photovoltaic output sample data and the meteorological sample data, the original generative adversarial network based on the principle of mutual information maximization is trained for at least one round to obtain the target generative adversarial network based on the principle of mutual information maximization, which can include:

[0046] S1101, the photovoltaic output sample data is divided into photovoltaic output training set data and photovoltaic output validation set data, and the meteorological sample data is divided into meteorological training set data and meteorological validation set data;

[0047] S1102, using the photovoltaic output training set data and the meteorological training set data, the generator, the discriminator and the classifier of the original generative adversarial network based on the principle of mutual information maximization are trained until the first loss function corresponding to the generator, the second loss function corresponding to the discriminator and the target function corresponding to the original generative adversarial network based on the principle of mutual information maximization converge, obtaining the standby target generative adversarial network based on the principle of mutual information maximization; wherein the generator and the discriminator are trained simultaneously;

[0048] S1103, using the photovoltaic output validation set data and the meteorological validation set data, the standby target generative adversarial network based on the principle of mutual information maximization is verified to obtain the target generative adversarial network based on the principle of mutual information maximization.

[0049] Illustratively, the division process of the photovoltaic output sample data and the meteorological sample data in S1101 can be explained by the following process.

[0050] For the photovoltaic output sample data X o , the annual sequence is converted into daily sequence to obtain m 144 m daily sequences, and further converted into a 12 12 square matrix, denoted as X. For the meteorological feature sample data Y o , the annual sequence is also converted into daily sequence to obtain m 144 m daily sequences, and further the daily average of the m daily sequences is obtained to obtain m meteorological features corresponding to the photovoltaic output sample, denoted as Y. Randomly divide 80% of the photovoltaic output sample data X o and the meteorological feature sample data Yo For the training set data, respectively denoted as X train and Y train ; the remaining 20% of the photovoltaic output sample data X o and meteorological feature sample data Y o are the validation set data, respectively denoted as X test and Y test .

[0051] Further, the "training the generator, discriminator and classifier of the original generative adversarial network based on the principle of maximizing mutual information, using photovoltaic output training set data and meteorological training set data" in S1102 can be refined as follows: input the preset random noise variable and the meteorological training set data into the generator, and output the photovoltaic output generated sample set data; input the photovoltaic output training set data and the photovoltaic output generated sample set data into the discriminator, and output the identification results corresponding to the photovoltaic output training set data and the photovoltaic output generated sample set data respectively; input the photovoltaic output training set data, the photovoltaic output generated sample set data and the meteorological training set data into the classifier, maximize the mutual information between the photovoltaic output generated sample set data and the meteorological training set data, and output the value of the mutual information.

[0052] S1103 can be refined as follows: input the preset random noise variable and the meteorological validation set data into the generator, and output the photovoltaic output generated validation set data; compare the first curve corresponding to the photovoltaic output generated validation set data with the second curve corresponding to the photovoltaic output validation set data, and obtain the comparison result; when the comparison result reaches the training end condition, obtain the target generative adversarial network based on the principle of maximizing mutual information; when the comparison result does not reach the training end condition, return to execute the operation of training the generator, discriminator and classifier of the original generative adversarial network based on the principle of maximizing mutual information, using photovoltaic output training set data and meteorological training set data, until the target generative adversarial network based on the principle of maximizing mutual information is obtained.

[0053] In this embodiment, the target generative adversarial network based on the principle of maximizing mutual information can be a generative adversarial network based on maximum mutual information.

[0054] For example, the following provides a specific process of training the target generative adversarial network based on the principle of maximizing mutual information based on X train , Y train , X test and Y test .

[0055] Step (1), the generator G inputs the preset random noise variable Z and the meteorological training set data Y trainThe input data, Z, follows a Gaussian distribution. The generator G consists of one fully connected layer and three deconvolutional neural networks. The relevant parameters of each layer of the generator G are shown in the second column of Table 1. In the input layer, 100 represents the dimension of Z, and 1 represents the dimension of Y. train The dimension of the fully connected layer is 1024, and the number of the three deconvolutional layers are 512, 256 and 128 respectively, with a stride of 2×2 for each layer.

[0056] The generator G first uses a fully connected layer for upsampling, then uses three deconvolutional neural networks to extract data features, finally obtaining the photovoltaic output generation sample set data X' of the corresponding training set. train =G(Z|Y train The output layer of generator G uses the tanh function as the activation function, while the remaining layers use ReLU activation. Batch Normalization (BN) is applied before ReLU activation, and the loss function is as follows: Loss G =-E Z～P(Z) [log(1-D(G(X′|Y train In the formula, E represents the expectation; G(·) represents the output of the generator; D(·) represents the output of the discriminator; and P(Z) represents the Gaussian distribution that Z follows.

[0057] Step (2): Discriminator D determines whether the input data belongs to a generated sample or a real sample.

[0058] The discriminator D and the generator G are trained simultaneously to generate samples X'. train (i.e., photovoltaic power generation sample set data) or real sample X train (i.e., photovoltaic power output training data) is used as the input data for discriminator D. The discriminator D structure consists of 3 convolutional neural network layers and 2 fully connected layers. The relevant parameters of each network layer are shown in the third column of Table 1. The parameter of the fully connected layer is 1024, and the number of the three convolutional neural network layers are 128, 256 and 512 respectively, with a stride of 2×2 for each layer.

[0059] Discriminator D first undergoes a series of downsampling operations through three convolutional neural networks, then passes through two fully connected layers, and finally outputs a judgment value of 1 (representing a real sample) or 0 (representing a generated sample). The output layer of discriminator D uses the Sigmoid function as its activation function, while the remaining layers use LeakyReLU activation, with batch normalization applied before LeakyReLU activation. The training objective of discriminator D is to obtain a discriminator that can accurately determine whether input data belongs to a generated sample or a real sample. Its loss function is as follows: In the formula, P(X) train ) represents the real sample Xtrain The distribution from which the sample X' train is generated. train The distribution from which the sample X'

[0060] Thus, the objective function of the target generative adversarial network based on the mutual information maximization principle can be obtained as a minimax game between the generator G and the discriminator D, as shown in equation (1).

[0061]

[0062] Step (3), constructing a classifier Q to maximize the mutual information between the photovoltaic output generated sample set data X' train =G(Z|Y train ) and the meteorological training set data Y train .

[0063] The structure of the classifier Q consists of 3 convolutional neural network layers and 1 fully connected layer. The related parameters of each layer network are shown in Table 1, column 4.

[0064] In order to enable Y train to effectively control the generation mode of the scene, there should be a high mutual information between Y train and X' train =G(Z|Y train ), and the mutual information between the two should be maximized. The mutual information maximization principle refers to quantifying the potential "information amount" between two random variables and maximizing it with certain mathematical principles. Here, we can define the difference between the entropy terms of X' train =G(Z|Y train ) and the meteorological Y train , as shown in equation (2).

[0065] I(Y train ; X′ train ) = H(Y train ) - H(Y train | X′ train ) Equation (2)

[0066] In the formula, H(Y train ) represents the Shannon entropy; H(Y train | X' train ) is the conditional entropy, which represents the remaining uncertainty in Y train when X' train is introduced; I(Y train ; X' train ) is the mutual information when X' train is observed, Y trainThe amount of reduction in uncertainty.

[0067] If Y train and X' train are independent, then I(Y train ; X' train ) = 0; if Y train and X' train are linked by a deterministic relationship, then the mutual information should be maximized. In this case, the objective function of the GAN can be improved as a min-max game with the maximum mutual information and the hyperparameter λ, as shown in equation (3).

[0068]

[0069] However, in practical applications, the mutual information term I(Y train ; X' train ) is difficult to maximize directly because the posterior distribution P(Y train | X train ) needs to be accessed. Therefore, a classifier Q(Y train | X train ) that is close enough to P(Y train | X train ) is needed to obtain a lower bound of I(Y train ; X' train ). The structure of the classifier Q(Y train | X train ) is shown in Table 1, which is consistent with the discriminator D, only a fully connected layer (FC) is added at the end to output the parameters. At this time, equation (2) can be improved as:

[0070]

[0071] where H(Y train ) is regarded as a constant; D KL (·) is the KL (Kullback-Leibler) divergence, whose value is greater than or equal to 0; Y' train represents the output of the classifier; L I (G, Q) is referred to as the variational lower bound.

[0072] Therefore, the objective function formula (3) of the GAN whose objective is based on the mutual information maximization principle can be improved as the following formula (5):

[0073]

[0074] Table 1

[0075]

[0076]

[0077] Step (4): Combine the random noise variable Z and the meteorological validation set data Y test Input the trained generator G to generate the corresponding photovoltaic power generation validation set data X' test =G(Z|Y test ), and X' test The output value curve (equivalent to the first curve) and X test By comparing the cumulative distribution function curve (equivalent to the second curve), the cumulative distribution function can reflect whether the photovoltaic output curve has a reasonable edge distribution, thereby verifying the rationality and effectiveness of the model.

[0078] S120. Input meteorological feature data into the generator of the target generative adversarial network based on the principle of maximizing mutual information to obtain a set of candidate photovoltaic power output scenarios.

[0079] In this embodiment, meteorological characteristic data collected from the area to be connected to the photovoltaic system is input into the aforementioned "X-based" process. train Y train X test and Y test The training objective is to train a generative adversarial network based on the principle of maximizing mutual information. The trained target is to generate a generator G based on the principle of maximizing mutual information, and obtain a set of candidate photovoltaic power output scenarios that match the photovoltaic area to be connected.

[0080] S130. By combining convolutional autoencoders with embedded clustering, the candidate photovoltaic power output scenario set is reduced to obtain typical photovoltaic power output scenarios that match the photovoltaic area to be connected.

[0081] In this embodiment, the candidate photovoltaic power output scenario set may include multiple candidate photovoltaic power output scenarios, and the candidate photovoltaic power output scenario set may be further reduced to obtain typical photovoltaic power output scenarios.

[0082] The technical solution of this invention involves obtaining a pre-trained generative adversarial network (GAN) based on the principle of maximizing mutual information, and acquiring meteorological feature data of the photovoltaic (PV) region to be connected. The meteorological feature data is then input into the generator of the GAN based on the principle of maximizing mutual information to obtain a set of candidate PV output scenarios. A combination of convolutional autoencoder and embedded clustering is used to reduce the set of candidate PV output scenarios, thereby obtaining typical PV output scenarios that match the PV region to be connected. This provides a method for obtaining typical PV output scenarios, improving the safety and stability of the traditional power grid operation after the PV power station is connected.

[0083] Example 2

[0084] Figure 2aThe flowchart of another photovoltaic power output typical scene acquisition method provided for the second embodiment of the present application refines the operation of reducing the candidate photovoltaic power output scene set and acquiring the photovoltaic power output typical scene matching the photovoltaic region to be accessed by combining the convolutional autoencoder with the embedded clustering, as shown in Figure 2a The method comprises the following steps.

[0085] In S210, a pre-trained target generative adversarial network based on the principle of mutual information maximization is acquired, and meteorological feature data at the photovoltaic region to be accessed is acquired.

[0086] In the embodiment, to acquire the photovoltaic power output typical scene at the photovoltaic region to be accessed, in addition to the pre-trained target generative adversarial network based on the principle of mutual information maximization, meteorological feature data needs to be collected from the photovoltaic region to be accessed.

[0087] In S220, the meteorological feature data is input into the generator of the target generative adversarial network based on the principle of mutual information maximization, and a candidate photovoltaic power output scene set is acquired.

[0088] For example, the meteorological feature data Y of the photovoltaic region to be accessed is acquired. unkown A preset random noise variable Z and Y unkown are input into the trained generator G, M candidate photovoltaic power output scenes are generated, denoted as X', and thus the photovoltaic power output scenes with a known distribution are obtained, thereby providing data and scene basis for subsequent safe operation evaluation and carrying capacity evaluation of the power grid.

[0089] In S230, each candidate photovoltaic power output scene of the candidate photovoltaic power output scene set is input into the encoder of the convolutional autoencoder for feature extraction, and a latent feature vector of each candidate photovoltaic power output scene is obtained.

[0090] In S240, each latent feature vector is input into the decoder of the convolutional autoencoder for reconstruction, and each reconstructed photovoltaic power output scene matching each candidate photovoltaic power output scene is obtained, and the encoder and the decoder are iterated by using a preset reconstruction loss function until the average variance of the preset reconstruction loss function is minimized.

[0091] In S250, each latent feature vector is input into the clustering layer of the convolutional autoencoder, and each latent feature vector is clustered according to a preset clustering number until the target loss function corresponding to the convolutional autoencoder converges, and a preset number of photovoltaic power output typical scenes are obtained; wherein the target loss function is determined by the preset reconstruction loss function and the clustering loss corresponding to the clustering layer.

[0092] In the embodiment, the scene reduction method based on the convolutional autoencoder and the embedded clustering is used to reduce the candidate photovoltaic power output scene set generated in S220, and representative photovoltaic power output typical scenes are obtained.

[0093] A convolutional autoencoder can include an encoder and a decoder, and the network structure of a convolutional autoencoder can be shown in Table 2. The time series representation of the M candidate photovoltaic power output scenarios generated in S220 is defined as input x'=[x′1,x′2,…x′…]. i ,…x' n ],in n x′ represents the time series length, here taken as 144, meaning one sampling point every 10 minutes; i The photovoltaic output value at any given time.

[0094] During the encoding phase of the encoder, the encoder reduces the dimensionality and extracts the latent features of each candidate photovoltaic power output scenario. The input is transformed through a hidden layer, as follows: z latent = p(h1), where: z represents the output after convolution. latent Representing latent features, whose dimension is much smaller than x'; σ f W represents the activation function. k Let b represent the convolution kernel of the k-th hidden layer. k This represents the deviation of the k-th hidden layer.

[0095] The encoder first sets x' = [x'1, x'2, ... x'] i , ...x' n The image is reshaped into a 12x12 square matrix, then passed through two convolutional neural network layers, one flattening layer, and one fully connected layer. The final output layer outputs the latent feature z. latent The feature vector is 12*1. The relevant parameters of each network layer are shown in Table 2. The parameters of the flattened layer are 1152, the parameters of the fully connected layer are 12, the number of the two convolutional neural network layers are 64 and 128 respectively, the stride is 2×2, and the activation function is σ. f Choose the LeakyReLU activation function.

[0096] In the decoding stage of the decoder, the latent features z obtained in the encoding stage are utilized. latent The reconstructed photovoltaic power output curve can be expressed as: h2=S(z latent ), y′=σ e (h2*U k +c2), where y' is the reconstructed data of input x'; σ e U represents the activation function of the decoder. k Let represent the convolution kernel of the k-th hidden layer, and c represent the bias of the k-th hidden layer.

[0097] Specifically, the decoder will convert the latent feature z latentAfter 1 fully connected layer, 1 reshaping layer and 2 deconvolutional neural network layers, a 12*12 square matrix is obtained, and is reshaped into a photovoltaic power output time series y'. The relevant parameters of each layer network are shown in Table 2, the parameters of the fully connected layer are 1152, the parameters of the fully connected layer are 12, the number of the two deconvolutional neural network layers is 64 and 1 respectively, the step length is 2*2, and the activation function is σ e The ReLU activation function is selected.

[0098] Table 2

[0099] Encoder Decoder Input layer 144*1 12*1 Network layer 1 Reshape, 12*12 Fully connected layer (FC), 1152 Network layer 2 Convolutional layer, 64 Reshape, 128 Network layer 3 Convolutional layer, 128 Deconvolutional layer, 64 Network layer 4 Flatten layer, 1152 Deconvolutional layer, 1 Output layer Fully connected layer, 12 Reshape, 144*1

[0100] The goal of the convolutional autoencoder is to make the reconstruction y' close to x', that is, to minimize the average variance of the loss function. We choose to use the gradient descent method (SDG) to solve the optimization problem. The reconstruction loss function can be defined as:

[0101] In order to extract high-quality typical photovoltaic power output scenarios, it is necessary to cluster a large set of photovoltaic power output scenarios to obtain typical scenarios. Therefore, a clustering layer is added to the embedding layer of the convolutional autoencoder, and the latent features of the photovoltaic power output can be optimized through the clustering process.

[0102] The embedded clustering takes the initial K-means clustering center as the initial setting of the network weight parameter and as the trainable weight. Then, the latent features z latent are mapped to soft labels using a student t-distribution.

[0103] Therefore, the loss function expression of the scenario reduction method based on the convolutional autoencoder and the embedded clustering is: L = L r + γL c , wherein γ represents the degree of controlling the distortion of the embedding layer, generally 0.1, L c represents the clustering loss. The above clustering loss L c and the reconstruction loss L r are combined for joint optimization to obtain the optimal photovoltaic power output typical scenario.

[0104] The input of the embedded clustering is the latent feature z latent output by the encoder, and the initial center of the K-means clustering is taken as the initial setting of the network weight parameter and the trainable weight. The latent feature z latent is mapped to soft labels using a student t-distribution. Specifically, first, the convolutional autoencoder without clustering loss is pre-trained to obtain the latent feature z latent after dimension reduction, and the latent feature z latentAnd the initial cluster center calculates the soft assignment degree; then, the target assignment is calculated; and then, the loss function is used to continue iteration, fine-tuning the network, updating the clustering result, comparing the changes of the two clustering assignments, if it is less than the given threshold, stop training, complete the final clustering.

[0105] The core idea of the K-means clustering algorithm is to divide the photovoltaic output scene set into K categories by iteratively optimizing the cluster center, each category represents a photovoltaic output typical scene, and each category of photovoltaic output typical scene is represented by a cluster center, and the sample is assigned to the category to which the nearest cluster center belongs. K-means clustering algorithm mainly includes the following steps:

[0106] 1) Select the number of clusters K: Before starting the algorithm, it is necessary to specify how many categories the data set will be divided into, here K is selected as 3.

[0107] 2) Initialize the cluster center: randomly select K data points from the data set as the initial cluster center.

[0108] 3) Assign samples to categories: for each sample, calculate the distance between it and each cluster center, and assign it to the nearest category.

[0109] 4) Update the cluster center: for each category, calculate the average of all samples in the category, and take the average as the new cluster center.

[0110] 5) Repeat steps 3) and 4) until the stopping condition is reached, here, the stopping condition is set to reach the error threshold 10 -4 .

[0111] 6) Output the clustering result: finally, K photovoltaic output typical scenes are obtained.

[0112] The clustering loss of embedded clustering is defined as the KL divergence between the distribution of soft labels and the target distribution, which is used to measure the matching degree between the two distributions, and can be expressed as:

[0113]

[0114] In the formula: q ij is defined as the probability of the i-th sample q i After passing through the encoding layer, z i belongs to the cluster center μ j ; p ij is defined as the auxiliary target distribution function.

[0115] In order to make those skilled in the art better understand the technical scheme of the embodiment, the following provides an example of applying the photovoltaic output typical scene acquisition method.

[0116] Step 1: Acquisition and preprocessing of raw data.

[0117] Using the photovoltaic system dataset obtained from the public platform, the photovoltaic output time series (kW) in one year in the dataset is selected as the photovoltaic raw data, and the solar horizontal radiation (w / m2xsr) meteorological data is selected as the controllable condition. The sampling frequency of the data time is 10 minutes, i.e. 144 sampling points per day. The above raw data is processed by data cleaning and maximum normalization to obtain photovoltaic output sample data and meteorological sample data.

[0118] Step 2: Photovoltaic output scene generation based on maximum mutual information of generative adversarial network.

[0119] After processing the photovoltaic output sample data and meteorological sample data and dividing the test set and the validation set, the Gaussian distribution random noise variable Z and the solar horizontal radiation control vector Y train are input into the generator, and the photovoltaic output generated sample X' train is obtained. train The photovoltaic output generated sample X' train and the real sample X test are input into the discriminator and the classifier.

[0120] The generative adversarial network based on maximum mutual information is trained, and the training parameters are set as follows: using Adam optimizer; the maximum number of iterations is 500; the data is trained in batches, and the number of a batch is 32; the training of the model is realized on the Pytorch platform.

[0121] The quality of the generated photovoltaic output scene is verified every 100 iterations, and a photovoltaic output sample is randomly extracted to compare the photovoltaic output generated scene X' test with the real scene X test As shown in Figure 2b - Figure 2f . It can be seen that as the number of iterations increases, the generated photovoltaic output scene is more and more close to the real photovoltaic output scene distribution. At the iteration of 500 times, the curves of the two basically overlap, and the model can effectively generate photovoltaic output scenes similar to the real scenes according to the meteorological characteristics. The cumulative distribution function of the photovoltaic output generated scene X' test and the real scene X test at the iteration of 500 times is shown in Figure 2g . As can be seen from the figure, the curves of the two almost overlap, which indicates that the model can generate photovoltaic output scenes with correct edge distribution.

[0122] Step 3: Generation of photovoltaic output scenes with unknown distribution.

[0123] The feeder of a certain regional distribution network is about to consider connecting photovoltaic, and the solar radiation data Y unknownThe Gaussian distributed random noise variable Z and the solar horizontal radiation control vector Y are... unknown Input the generator G trained in step 4 to obtain a set of generated scenarios containing 600 photovoltaic power output scenarios, thereby obtaining a known distribution of photovoltaic power output scenarios, providing data and scenario basis for subsequent safe operation assessment and carrying capacity assessment of the distribution network. Figure 2h The photovoltaic power generation scenarios of a training batch are shown in the generated scenario set.

[0124] Step 4: Obtaining typical photovoltaic power output scenarios.

[0125] Scene reduction based on convolutional autoencoder and embedded clustering reduces the photovoltaic output scene set obtained in step 3 to obtain representative typical photovoltaic output scenes, such as... Figure 2i As shown.

[0126] The technical solution of this invention involves acquiring a pre-trained generative adversarial network (GAN) based on the principle of maximizing mutual information, and obtaining meteorological feature data of the area to be connected to the photovoltaic power grid. The meteorological feature data is input into the generator of the GAN based on the principle of maximizing mutual information to obtain a set of candidate photovoltaic power output scenarios. Each candidate photovoltaic power output scenario in the set is input into the encoder of a convolutional autoencoder for feature extraction, resulting in a latent feature vector for each scenario. Each latent feature vector is input into the decoder of the autoencoder for reconstruction, resulting in reconstructed photovoltaic power output scenarios that match the candidate scenarios. The encoder and decoder are iterated using a preset reconstruction loss function until the average variance of the preset reconstruction loss function is minimized. Each latent feature vector is input into the clustering layer of the autoencoder, and clustered according to a preset number of clusters until the target loss function corresponding to the preset convolutional encoding converges, resulting in a preset number of typical photovoltaic power output scenarios. This provides a method for obtaining typical photovoltaic power output scenarios, improving the safety and stability of the traditional power grid operation after photovoltaic power plants are connected.

[0127] Example 3

[0128] Figure 3 This is a schematic diagram of a typical photovoltaic power output acquisition device provided in Embodiment 3 of the present invention. Figure 3 As shown, the device includes: a network and feature data acquisition module 310, a candidate photovoltaic power output scenario set acquisition module 320, and a typical photovoltaic power output scenario acquisition module 330. Wherein:

[0129] The network and feature data acquisition module 310 is used to acquire a pre-trained generative adversarial network based on the principle of maximizing mutual information of the target, and to acquire meteorological feature data of the photovoltaic area to be connected.

[0130] The candidate photovoltaic output scene set acquisition module 320 is configured to input the meteorological feature data into a generator of the target generative adversarial network based on the principle of mutual information maximization, and acquire a candidate photovoltaic output scene set.

[0131] The photovoltaic output typical scene acquisition module 330 is configured to reduce the candidate photovoltaic output scene set by using a convolutional autoencoder combined with embedded clustering, and acquire a photovoltaic output typical scene matched with the photovoltaic region to be accessed.

[0132] The technical scheme of the embodiment of the application comprises the following steps: acquiring a pre-trained target generative adversarial network based on the principle of mutual information maximization, and acquiring meteorological feature data at a photovoltaic region to be accessed; inputting the meteorological feature data into a generator of the target generative adversarial network based on the principle of mutual information maximization, and acquiring a candidate photovoltaic output scene set; reducing the candidate photovoltaic output scene set by using a convolutional autoencoder combined with embedded clustering, and acquiring a photovoltaic output typical scene matched with the photovoltaic region to be accessed. The embodiment of the application provides a photovoltaic output typical scene acquisition method, and improves the safety and stability of the operation of a traditional power grid after a photovoltaic power station is accessed.

[0133] Optionally, the network and feature data acquisition module 310 can comprise:

[0134] The raw data acquisition sub-module is configured to acquire photovoltaic output raw data and meteorological raw data from a photovoltaic region that has been accessed.

[0135] The once-processed data acquisition sub-module is configured to filter out missing values in the photovoltaic output raw data and the meteorological raw data, and obtain first-processed photovoltaic output data and first-processed meteorological data.

[0136] The sample data acquisition sub-module is configured to normalize the first-processed photovoltaic output data and the first-processed meteorological data by using a maximum value normalization principle, and obtain photovoltaic output sample data and meteorological sample data.

[0137] The network acquisition sub-module is configured to train an original generative adversarial network based on the principle of mutual information maximization at least once by using the photovoltaic output sample data and the meteorological sample data, and obtain the target generative adversarial network based on the principle of mutual information maximization.

[0138] Optionally, the network acquisition sub-module can comprise:

[0139] The sample data division unit is configured to divide the photovoltaic output sample data into photovoltaic output training set data and photovoltaic output verification set data, and divide the meteorological sample data into meteorological training set data and meteorological verification set data.

[0140] The backup target GAN based on the principle of mutual information maximization is obtained by inputting the preset random noise variable and the meteorological training set data into the generator, and outputting photovoltaic output generated sample set data.

[0141] The target GAN based on the principle of mutual information maximization is obtained by inputting the photovoltaic output verification set data and the meteorological verification set data into the backup target GAN based on the principle of mutual information maximization, and outputting the target GAN based on the principle of mutual information maximization.

[0142] Optionally, the backup target GAN based on the principle of mutual information maximization can be used for:

[0143] The preset random noise variable and the meteorological training set data are input into the generator, and photovoltaic output generated sample set data is output.

[0144] The photovoltaic output training set data and the photovoltaic output generated sample set data are input into the discriminator, and identification results corresponding to the photovoltaic output training set data and the photovoltaic output generated sample set data are output.

[0145] The photovoltaic output training set data, the photovoltaic output generated sample set data and the meteorological training set data are input into the classifier, mutual information between the photovoltaic output generated sample set data and the meteorological training set data is maximized, and a value of the mutual information is output.

[0146] The mutual information is represented as: I(Y train ; X' train ) = H(Y train ) - H(Y train | X' train ); wherein I(Y train ; X' train ) represents a reduction in uncertainty in Y train when X' train is observed, H(Y train ) represents a Shannon entropy, H(Y train | X' trai ) n represents a conditional entropy, and represents a remaining uncertainty in Y train when X' train is introduced.

[0147] Optionally, the first loss function is represented as: Loss G = -E Z～P(Z) [log(1-D(G(X'|Y train )))];wherein E represents expectation; Z represents the preset random noise variable, P(Z) represents a distribution to which the preset random noise variable Z conforms; G(·) represents an output of the generator, and D(·) represents an output of the discriminator;

[0148] The second loss function is represented as: wherein X train represents the photovoltaic output training set data, Y train represents the meteorological training set data, X' train represents the photovoltaic output generated sample set data, P(X train ) represents a distribution to which X train conforms, and P(X' train ) represents a distribution to which X' train conforms;

[0149] The target function is a maximum-minimum game based on the generator, the discriminator and the classifier; and the target function is represented as: wherein Q represents the classifier, λ represents a hyperparameter, L I (G,Q) represents a variational lower bound of I(Y train ;X' train ).

[0150] Optionally, the target GAN based on the mutual information maximization principle acquisition unit can be specifically used for:

[0151] inputting the preset random noise variable and the meteorological verification set data into the generator to output photovoltaic output generated verification set data;

[0152] comparing a first curve corresponding to the photovoltaic output generated verification set data with a second curve corresponding to the photovoltaic output verification set data to obtain a comparison result;

[0153] when the comparison result reaches a training end condition, obtaining the target GAN based on the mutual information maximization principle;

[0154] when the comparison result does not reach the training end condition, returning to perform an operation of training the generator, the discriminator and the classifier of the original GAN based on the mutual information maximization principle by using the photovoltaic output training set data and the meteorological training set data until the target GAN based on the mutual information maximization principle is obtained.

[0155] Optionally, the photovoltaic power output typical scene acquisition module 330 can be specifically used for:

[0156] input each candidate photovoltaic power output scene of the candidate photovoltaic power output scene set into an encoder of the convolutional autoencoder for feature extraction, to obtain a latent feature vector of each candidate photovoltaic power output scene;

[0157] input each latent feature vector into a decoder of the convolutional autoencoder for reconstruction, to obtain each reconstructed photovoltaic power output scene matched with the candidate photovoltaic power output scene, and iteratively update the encoder and the decoder by using a preset reconstruction loss function until the average variance of the preset reconstruction loss function is minimized;

[0158] input each latent feature vector into a clustering layer of the convolutional autoencoder, and perform clustering processing on each latent feature vector according to a preset clustering number until a target loss function corresponding to the preset convolutional encoding converges, to obtain a preset clustering number of photovoltaic power output typical scenes; wherein the target loss function is determined by the preset reconstruction loss function and a clustering loss corresponding to the clustering layer.

[0159] The photovoltaic power output typical scene acquisition device provided in the embodiments of the present application can perform the photovoltaic power output typical scene acquisition method provided in any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0160] Embodiment four

[0161] Figure 4 A structural schematic diagram of an electronic device 400 that can be used to implement embodiments of the present application is shown. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described herein and / or claimed.

[0162] As shown in Figure 4 The electronic device 400 includes at least one processor 401 and a memory, such as a read-only memory (ROM) 402, a random access memory (RAM) 403, etc., which are communicatively connected to the at least one processor 401, wherein the memory stores a computer program that can be executed by the at least one processor. The processor 401 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 402 or the computer program loaded from the storage unit 408 into the random access memory (RAM) 403. In the RAM 403, various programs and data required for the operation of the electronic device 400 can also be stored. The processor 401, the ROM 402, and the RAM 403 are connected to each other through a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0163] A plurality of components in the electronic device 400 are connected to the I / O interface 405, including: an input unit 406, such as a keyboard, a mouse, etc.; an output unit 407, such as various types of displays, speakers, etc.; a storage unit 408, such as a magnetic disk, an optical disk, etc.; and a communication unit 409, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 409 allows the electronic device 400 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0164] The processor 401 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the processor 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 401 performs various methods and processes described above, such as the photovoltaic power output typical scenario acquisition method.

[0165] In some embodiments, the photovoltaic power output typical scenario acquisition method can be implemented as a computer program tangibly embodied in a computer readable storage medium, such as the storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 400 via the ROM 402 and / or the communication unit 409. When the computer program is loaded onto the RAM 403 and executed by the processor 401, one or more steps of the photovoltaic power output typical scenario acquisition method described above can be performed. Alternatively, in other embodiments, the processor 401 can be configured to perform the photovoltaic power output typical scenario acquisition method by any other appropriate means, such as by means of firmware.

[0166] The various embodiments of the systems and techniques described above can be implemented in digital electronic circuitry, integrated circuitry, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0167] Computer programs for implementing the methods of the present application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer program, when executed, can cause instructions defined in the flow charts and / or block diagrams to be implemented. The computer program can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, and partially on a remote machine or entirely on a remote machine or server.

[0168] In the context of the present application, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer-readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of a machine-readable storage medium will include one or more lines of a program of instructions in a transitory signal, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0169] To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0170] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0171] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0172] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0173] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for obtaining typical photovoltaic power output scenarios, characterized in that, include: The pre-trained target generative adversarial network is used to obtain the meteorological feature data of the photovoltaic area to be connected; The meteorological feature data is input into the generator of the target based on the principle of mutual information maximization of the generative adversarial network to obtain a set of candidate photovoltaic power output scenarios; By combining convolutional autoencoder with embedded clustering, the candidate photovoltaic power output scenario set is reduced to obtain typical photovoltaic power output scenarios that match the photovoltaic area to be connected; The target generative adversarial network based on the mutual information maximization principle is obtained by training the original generative adversarial network based on the mutual information maximization principle for at least one round; the original generative adversarial network based on the mutual information maximization principle includes a generator, a discriminator, and a classifier; the first loss function corresponding to the generator during training is expressed as: ;in, Expressing expectations; This represents a preset random noise variable. Represents the preset random noise variable The distribution it follows; This represents the output of the generator. This represents the output of the discriminator; the second loss function corresponding to the discriminator during training is expressed as: ;in, This refers to the photovoltaic power output training set data. This refers to the meteorological training set data. This indicates that the photovoltaic output generates a sample set of data. express The distribution that it follows express The distribution follows; the objective function of the original generative adversarial network based on the principle of maximizing mutual information is a minimax game based on the generator, the discriminator, and the classifier, and the objective function is expressed as: ,in, This refers to the classifier. Indicates hyperparameters, express The variational lower bound, Indicates when observed hour, The amount of uncertainty reduced.

2. The method according to claim 1, characterized in that, Obtaining pre-trained target generative adversarial networks based on the mutual information maximization principle includes: Obtain raw photovoltaic power output data and raw meteorological data from areas where photovoltaic systems have been connected; Missing values ​​in the original photovoltaic power output data and the original meteorological data are filtered out to obtain the photovoltaic power output data and meteorological data after the first processing. The photovoltaic power output data and the meteorological data after the first processing are normalized using the principle of maximum value normalization to obtain photovoltaic power output sample data and meteorological sample data. Using the photovoltaic power output sample data and the meteorological sample data, the original generative adversarial network based on the mutual information maximization principle is trained for at least one round to obtain the target generative adversarial network based on the mutual information maximization principle.

3. The method according to claim 2, characterized in that, Using the photovoltaic power output sample data and the meteorological sample data, the original generative adversarial network based on the mutual information maximization principle is trained for at least one round to obtain the target generative adversarial network based on the mutual information maximization principle, including: The photovoltaic power output sample data is divided into photovoltaic power output training set data and photovoltaic power output verification set data, and the meteorological sample data is divided into meteorological training set data and meteorological verification set data; Using the photovoltaic power output training set data and the meteorological training set data, the generator, the discriminator, and the classifier are trained until the first loss function, the second loss function, and the objective function converge, resulting in a generative adversarial network for the backup target based on the principle of maximizing mutual information; wherein, the generator and the discriminator are trained simultaneously. The photovoltaic power output verification set data and the meteorological verification set data are used to verify the generative adversarial network of the backup target based on the mutual information maximization principle, thereby obtaining the generative adversarial network of the target based on the mutual information maximization principle.

4. The method according to claim 3, characterized in that, The generator, discriminator, and classifier are trained using the photovoltaic power output training set data and the meteorological training set data, including: The generator is input with a preset random noise variable and the meteorological training set data, and outputs a photovoltaic power generation sample set data. The photovoltaic power output training set data and the photovoltaic power output generation sample set data are input into the discriminator, and the discriminator outputs the recognition results corresponding to the photovoltaic power output training set data and the photovoltaic power output generation sample set data, respectively. The photovoltaic power output training set data, the photovoltaic power output generation sample set data, and the meteorological training set data are input into the classifier. The mutual information between the photovoltaic power output generation sample set data and the meteorological training set data is maximized, and the value of the mutual information is output. The mutual information is represented as follows: ;in, Represents Shannon entropy; Represents conditional entropy, and represents the introduction of... hour The remaining uncertainties.

5. The method according to claim 3, characterized in that, The photovoltaic power output verification set data and the meteorological verification set data are used to verify the generative adversarial network of the backup target based on the mutual information maximization principle, resulting in the generative adversarial network of the target based on the mutual information maximization principle, including: The preset random noise variable and the meteorological verification set data are input into the generator, and the photovoltaic output generation verification set data is output. The first curve corresponding to the photovoltaic power generation verification set data is compared with the second curve corresponding to the photovoltaic power generation verification set data to obtain the comparison result; When the comparison results reach the training termination condition, the target generative adversarial network based on the mutual information maximization principle is obtained; If the comparison result does not meet the training termination condition, return to the operation of training the generator, discriminator and classifier of the original generative adversarial network based on the mutual information maximization principle using the photovoltaic power output training set data and the meteorological training set data, until the target generative adversarial network based on the mutual information maximization principle is obtained.

6. The method according to claim 1, characterized in that, By combining convolutional autoencoders with embedded clustering, the candidate photovoltaic output scenario set is reduced to obtain typical photovoltaic output scenarios that match the photovoltaic region to be connected, including: Each candidate photovoltaic output scenario in the candidate photovoltaic output scenario set is input into the encoder of the convolutional autoencoder for feature extraction to obtain the potential feature vector of each candidate photovoltaic output scenario. Each potential feature vector is input into the decoder of the convolutional autoencoder for reconstruction, resulting in each reconstructed photovoltaic output scenario that matches each candidate photovoltaic output scenario. The encoder and decoder are iterated using a preset reconstruction loss function until the average variance of the preset reconstruction loss function is minimized. Each potential feature vector is input into the clustering layer of the convolutional autoencoder, and each potential feature vector is clustered according to a preset number of clusters until the target loss function corresponding to the convolutional autoencoder converges, thereby obtaining a preset number of typical photovoltaic power output scenarios; wherein the target loss function is jointly determined by the preset reconstruction loss function and the clustering loss corresponding to the clustering layer.

7. A device for acquiring typical photovoltaic power output scenarios, characterized in that, include: The network and feature data acquisition module is used to acquire a pre-trained generative adversarial network based on the mutual information maximization principle for the target, and to acquire meteorological feature data of the photovoltaic area to be connected. The candidate photovoltaic power output scenario set acquisition module is used to input the meteorological feature data into the generator of the target based on the principle of mutual information maximization of the generative adversarial network to acquire the candidate photovoltaic power output scenario set; The photovoltaic output typical scenario acquisition module is used to reduce the candidate photovoltaic output scenario set by combining convolutional autoencoder and embedded clustering, and obtain the photovoltaic output typical scenario that matches the photovoltaic area to be connected; The target generative adversarial network based on the mutual information maximization principle is obtained by training the original generative adversarial network based on the mutual information maximization principle for at least one round; the original generative adversarial network based on the mutual information maximization principle includes a generator, a discriminator, and a classifier; the first loss function corresponding to the generator during training is expressed as: ;in, Expressing expectations; This represents a preset random noise variable. Represents the preset random noise variable The distribution it follows; This represents the output of the generator. This represents the output of the discriminator; the second loss function corresponding to the discriminator during training is expressed as: ;in, This refers to the photovoltaic power output training set data. This refers to the meteorological training set data. This indicates that the photovoltaic output generates a sample set of data. express The distribution that it follows express The distribution follows; the objective function of the original generative adversarial network based on the principle of maximizing mutual information is a minimax game based on the generator, the discriminator, and the classifier, and the objective function is expressed as: ,in, This refers to the classifier. Indicates hyperparameters, express The variational lower bound, Indicates when observed hour, The amount of uncertainty reduced.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the photovoltaic power output typical scenario acquisition method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that are used to cause a processor to execute the method for obtaining typical photovoltaic power output scenarios as described in any one of claims 1-6.

Images