A method, apparatus, device and medium for predicting output parameters of a photovoltaic module

Abstract

This application discloses a method, apparatus, device, and medium for predicting the output parameters of photovoltaic modules, belonging to the field of photovoltaic module technology. The method includes: predicting the output parameters of photovoltaic modules using a target model; wherein the creation process of the target model includes: acquiring historical operating data of the photovoltaic modules and selecting training data from it; training the training data using a residual convolutional neural network, a residual gated recurrent unit network, and a fully connected layer; and obtaining the target model when the root mean square error between the output data of the fully connected layer and the preset label data reaches its minimum value; the residual convolutional neural network is a neural network obtained by optimizing a convolutional neural network using a residual network, and the residual gated recurrent unit network is a neural network obtained by optimizing a gated recurrent unit network using a residual network. This method can not only more accurately predict the output parameters of photovoltaic modules but also improve the generalization ability of the model.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic module technology, and in particular to a method, apparatus, equipment and medium for predicting the output parameters of a photovoltaic module. Background Technology

[0002] In existing technologies, most methods employ backpropagation (BP) neural networks, multilayer perceptrons, support vector machines, and decision trees to predict the output parameters of photovoltaic (PV) modules. However, these methods suffer from low prediction accuracy and poor generalization ability. Currently, there is no effective solution to this technical problem. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for predicting the output parameters of photovoltaic modules, so as to improve the accuracy of photovoltaic module prediction results and the generalization ability of the model. The specific solution is as follows:

[0004] A method for predicting the output parameters of a photovoltaic module, comprising:

[0005] The output parameters of a photovoltaic module are predicted using a pre-trained target model; wherein the creation process of the target model includes:

[0006] Obtain historical operating data of the photovoltaic module, and select target training data from the historical operating data;

[0007] The target training data is trained using a residual convolutional neural network, a residual gated recurrent unit network, and a fully connected layer. The target model is obtained when the root mean square error between the output data of the fully connected layer and the preset label data reaches its minimum value. The input and output data of the residual gated recurrent unit network are the output data of the residual convolutional neural network and the input data of the fully connected layer, respectively. Furthermore, the residual convolutional neural network is a neural network optimized from a convolutional neural network using a residual network, and the residual gated recurrent unit network is a neural network optimized from a gated recurrent unit network using the residual network.

[0008] Preferably, the residual convolutional neural network includes: one first residual convolutional module and i second residual convolutional modules; wherein, i≥1; the first residual convolutional module is obtained by inputting the result of adding two convolutional kernels into a first activation function; when i=1, the first second residual convolutional module is obtained by inputting the output result of the first residual convolutional module, the result of adding two convolutional kernels and a shortcut into a second activation function; when i≥2, the i-th second residual convolutional module is obtained by inputting the output result of the (i-1)-th second residual convolutional module, the result of adding two convolutional kernels and a shortcut into a second activation function.

[0009] Preferably, the two convolution kernels in the first residual convolution module are a 3*1 convolution kernel and a 1*1 convolution kernel, respectively, and the two convolution kernels in the second residual convolution module are a 3*1 convolution kernel and a 1*1 convolution kernel, respectively.

[0010] Preferably, both the first activation function and the second activation function are ReLU activation functions.

[0011] Preferably, the residual gated recurrent unit network is obtained by adding the branch network and the first gated recurrent unit; wherein, the branch network is obtained by normalizing and processing the second gated recurrent unit with the ReLU activation function, inputting the corresponding processing result into the third gated recurrent unit, and normalizing the third recurrent gated unit.

[0012] Preferably, the process of acquiring the historical operating data of the photovoltaic module includes:

[0013] The voltage, irradiance, temperature, and current data of the photovoltaic module at historical operating times are obtained.

[0014] Preferred options also include:

[0015] The optimal model parameters of the target model are obtained using K-fold cross-validation.

[0016] Accordingly, the present invention also discloses a device for predicting the output parameters of a photovoltaic module, comprising:

[0017] The parameter prediction module is used to predict the output parameters of photovoltaic modules using a pre-trained target model; wherein, the creation process of the target model includes:

[0018] A data selection unit is used to acquire historical operating data of the photovoltaic module and select target training data from the historical operating data;

[0019] The model training unit is used to train the target training data using a residual convolutional neural network, a residual gated recurrent unit network, and a fully connected layer. The target model is obtained when the root mean square error between the output data of the fully connected layer and the preset label data reaches its minimum value. The input and output data of the residual gated recurrent unit network are the output data of the residual convolutional neural network and the input data of the fully connected layer, respectively. Furthermore, the residual convolutional neural network is a neural network optimized from a convolutional neural network using a residual network, and the residual gated recurrent unit network is a neural network optimized from a gated recurrent unit network using the residual network.

[0020] Accordingly, the present invention also discloses a device for predicting the output parameters of a photovoltaic module, comprising:

[0021] Memory, used to store computer programs;

[0022] A processor is configured to execute the computer program to implement the steps of a method for predicting output parameters of a photovoltaic module as disclosed above.

[0023] Accordingly, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a method for predicting output parameters of a photovoltaic module as disclosed above.

[0024] As can be seen, in this invention, when creating a target model to predict the output parameters of a photovoltaic module, the first step is to acquire historical operating data of the photovoltaic module and select target training data from this historical operating data. Then, the target training data is trained using a residual convolutional neural network, a residual gated recurrent unit network, and a fully connected layer. The target model is obtained when the root mean square error between the output data of the fully connected layer and the preset label data reaches its minimum value. The output data of the residual convolutional neural network serves as the input data of the residual gated recurrent unit network, and the output data of the residual gated recurrent unit network serves as the input data of the fully connected layer. Furthermore, the residual convolutional neural network is a neural network obtained by optimizing a convolutional neural network using a residual network, and the residual gated recurrent unit network is a neural network obtained by optimizing a gated recurrent unit network using a residual network. Compared to existing technologies, because residual convolutional neural networks (RCNs) and residual gated recurrent unit networks (RRNNs) are neural networks optimized from convolutional neural networks using residual networks, they can have a greater number of hidden layers than ordinary neural network models. Therefore, the target model created using this method can not only more accurately predict the output parameters of photovoltaic modules, but also improve the model's generalization ability. Accordingly, the photovoltaic module output parameter prediction device, equipment, and medium provided by this invention also have the above-mentioned beneficial effects. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0026] Figure 1 A flowchart illustrating a method for predicting output parameters of a photovoltaic module provided in an embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of the structure of a residual convolutional neural network provided in an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of the structure of a residual gated cyclic unit network provided in an embodiment of the present invention;

[0029] Figure 4 This is an overall schematic diagram of training target training data using residual convolutional neural networks, residual gated recurrent unit networks, and fully connected layers.

[0030] Figure 5 A structural diagram of a photovoltaic module output parameter prediction device provided in an embodiment of the present invention;

[0031] Figure 6 This is a structural diagram of a photovoltaic module output parameter prediction device provided in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figure 1 , Figure 1 A flowchart of a method for predicting output parameters of a photovoltaic module provided in an embodiment of the present invention is shown. The method includes:

[0034] The output parameters of photovoltaic modules are predicted using a pre-trained target model; the creation process of the target model includes:

[0035] Step S11: Obtain historical operating data of photovoltaic modules and select target training data from the historical operating data;

[0036] Step S12: Train the target training data using a residual convolutional neural network, a residual gated recurrent unit network, and a fully connected layer. When the root mean square error between the output data of the fully connected layer and the preset label data reaches its minimum value, the target model is obtained. The input and output data of the residual gated recurrent unit network are the output data of the residual convolutional neural network and the input data of the fully connected layer, respectively. Furthermore, the residual convolutional neural network is a neural network obtained by optimizing a convolutional neural network using a residual network, and the residual gated recurrent unit network is a neural network obtained by optimizing a gated recurrent unit network using a residual network.

[0037] In this embodiment, a method for predicting the output parameters of a photovoltaic module is provided. When using the target model provided in this embodiment to predict the output parameters of a photovoltaic module, not only can the accuracy of the prediction results of the photovoltaic module output parameters be improved, but also the generalization ability of the photovoltaic module model can be improved.

[0038] When creating a target model to predict the output parameters of a photovoltaic (PV) module, the first step is to acquire historical operating data of the PV module and then select target training data from this historical data. This historical operating data includes all data generated by the PV module during operation, such as ambient temperature, radiation intensity, sunlight, current, voltage, and humidity. The purpose of selecting target training data from this historical operating data is to eliminate erroneous and invalid data, thereby making the resulting target model for the PV module more accurate and reliable.

[0039] Next, the residual network is used to optimize the convolutional neural network, resulting in a residual convolutional neural network. Then, the residual network is used to optimize the gated recurrent unit network, resulting in a residual gated recurrent unit network. After creating the residual convolutional neural network and the residual gated recurrent unit network, they are used together with fully connected layers to train the target training data.

[0040] When training on the target training data, the target training data is first input into the residual convolutional neural network (RCN). The output data of the RNN serves as the input data for the residual gated recurrent unit (RNU) network, which in turn serves as the input data for the fully connected layer. The purpose of connecting the fully connected layer to the backend of the RNN and RNU is to map the feature data output by these networks to the feature space containing the pre-defined label data. This allows the learnable parameters of each network layer to be updated iteratively from the last layer backwards using the chain rule, ensuring the effectiveness of the model parameters. When the root mean square error between the output data of the fully connected layer and the pre-defined label data reaches its minimum, it indicates that the model parameters in the RNN, RNU, and fully connected layers have been successfully trained, resulting in the trained target model.

[0041] It is understandable that, since residual convolutional neural networks are neural networks obtained by optimizing convolutional neural networks using residual networks, and residual gated recurrent unit networks are neural networks obtained by optimizing gated recurrent unit networks using residual networks, residual convolutional neural networks and residual gated recurrent unit networks can have a greater number of hidden layers than ordinary neural networks. Therefore, the target model created through this embodiment can improve the accuracy of predicting the output parameters of photovoltaic modules, and can also improve the generalization ability of the model.

[0042] It should be noted that in this embodiment, the output parameters of the photovoltaic module can be any data related to the operation of the photovoltaic module. For example, the output parameters of the photovoltaic module can be the output current, the voltage, or the temperature. In this case, it is only necessary to adaptively adjust the data type of the target training data according to people's actual application needs, which will not be elaborated here.

[0043] As can be seen, in this embodiment, when creating a target model to predict the output parameters of a photovoltaic module, the first step is to acquire historical operating data of the photovoltaic module and select target training data from this data. Then, the target training data is trained using a residual convolutional neural network, a residual gated recurrent unit network, and a fully connected layer. The target model is obtained when the root mean square error between the output data of the fully connected layer and the preset label data reaches its minimum value. The output data of the residual convolutional neural network is the input data of the residual gated recurrent unit network, and the output data of the residual gated recurrent unit network is the input data of the fully connected layer. Furthermore, the residual convolutional neural network is a neural network obtained by optimizing a convolutional neural network using a residual network, and the residual gated recurrent unit network is a neural network obtained by optimizing a gated recurrent unit network using a residual network. Compared to existing technologies, because residual convolutional neural networks (RCNs) are neural networks optimized from convolutional neural networks using residual networks, and residual gated recurrent unit networks (GRUNNs) are neural networks optimized from gated recurrent unit networks using residual networks, these residual RNNs and GRUNNs can have a greater number of hidden layers than ordinary neural network models. Therefore, the target model created by this method can not only more accurately predict the output parameters of photovoltaic modules, but also improve the generalization ability of the model.

[0044] Based on the above embodiments, this embodiment further explains and optimizes the technical solution. As a preferred implementation, the residual convolutional neural network includes: one first residual convolutional module and i second residual convolutional modules; wherein, i≥1; the first residual convolutional module is obtained by inputting the result of adding two convolutional kernels into a first activation function; when i=1, the first second residual convolutional module is obtained by inputting the output result of the first residual convolutional module, the result of adding two convolutional kernels and a shortcut into a second activation function; when i≥2, the i-th second residual convolutional module is obtained by inputting the output result of the (i-1)-th second residual convolutional module, the result of adding two convolutional kernels and a shortcut into a second activation function.

[0045] In this embodiment, when setting up the residual convolutional neural network, one first residual convolutional module and i second residual convolutional modules can be set up in the residual convolutional neural network. That is, one first residual convolutional module and at least one second residual convolutional module can be set up in the residual convolutional neural network. Among them, the first residual convolutional module is a traditional residual convolutional module, and the second residual convolutional module is an improved and optimized residual convolutional module.

[0046] Specifically, the first residual convolutional module is obtained by inputting the result of adding two convolutional kernels into the first activation function; when i=1, one second residual convolutional module is set in the residual convolutional neural network. In this case, the first second residual convolutional module inputs the output of the first residual convolutional module, the result of adding two convolutional kernels and a shortcut (direct connection channel) into the second activation function; when i≥2, at least two second residual convolutional modules are set in the residual convolutional neural network. In this case, the structure of the first second residual convolutional module is consistent with the structure of the first second residual convolutional module mentioned above, and will not be repeated here; and the i-th second residual convolutional module is obtained by inputting the output of the (i-1)-th second residual convolutional module, the result of adding two convolutional kernels and a shortcut into the second activation function.

[0047] As a preferred implementation, in practical applications, one first residual convolutional module and four second residual convolutional modules can be set up in the residual convolutional neural network. See also... Figure 2 , Figure 2 This is a schematic diagram of the structure of a residual convolutional neural network provided in an embodiment of the present invention. Figure 2 The residual convolutional neural network shown includes: one first residual convolutional module and four second residual convolutional modules; wherein, the first residual convolutional module is obtained by adding the result of two convolutional kernels and inputting it into a first activation function; the first second residual convolutional module is obtained by adding the output of the first residual convolutional module, the result of two convolutional kernels and a shortcut, and inputting it into a second activation function; the second second residual convolutional module is obtained by adding the output of the first second residual convolutional module, the result of two convolutional kernels and a shortcut, and inputting it into a second activation function; the third second residual convolutional module is obtained by adding the output of the second second residual convolutional module, the result of two convolutional kernels and a shortcut, and inputting it into a second activation function; the fourth second residual convolutional module is obtained by adding the output of the third second residual convolutional module, the result of two convolutional kernels and a shortcut, and inputting it into a second activation function.

[0048] It should be noted that in practical applications, the parameters of the convolution kernel can be adjusted according to the size of the target training data and the accuracy requirements of the target model's prediction results; these will not be elaborated upon here. Furthermore, since the purpose of the first and second activation functions is to perform nonlinear transformations on the input data and increase the expressive power of the neural network, in practice, the first and second activation functions can be set to sigmoid, tanh, or softmax functions, etc.

[0049] In a preferred embodiment, the two convolutional kernels in the first residual convolutional module are set to 3*1 and 1*1 kernels, respectively, and the two convolutional kernels in the second residual convolutional module are also set to 3*1 and 1*1 kernels, respectively. This configuration ensures that the network model depth is increased while maintaining the same receptive field, thus further improving the prediction accuracy of the target model. Figure 2 The residual convolutional neural network shown is equivalent to a set of 2 1 ×3 4 One hidden layer.

[0050] Furthermore, in this embodiment, both the first and second activation functions are set to ReLU activation functions. This is because, compared to other types of activation functions, ReLU activation functions not only improve the training speed of neural networks but also reduce overfitting in network model creation.

[0051] Clearly, the technical solution provided in this embodiment can further improve the accuracy and reliability of the target model in predicting the output parameters of photovoltaic modules.

[0052] Based on the above embodiments, this embodiment further explains and optimizes the technical solution. Please refer to [link / reference]. Figure 3 , Figure 3 This is a schematic diagram of a residual gated recurrent unit network provided in an embodiment of the present invention. In a preferred embodiment, the residual gated recurrent unit network is obtained by adding a branch network and a first gated recurrent unit; wherein the branch network is obtained by normalizing and applying the ReLU activation function to a second gated recurrent unit, inputting the corresponding processing result to a third gated recurrent unit, and then normalizing the third recurrent gated unit.

[0053] In this embodiment, when creating the residual gated recurrent unit network, two network branches are set up within the residual gated recurrent unit network. The first network branch consists of a first gated recurrent unit, and the second network branch consists of a branch network. When constructing the branch network, the second gated recurrent unit needs to be normalized and processed with the ReLU activation function first. Then, the processing result of the normalized and ReLU-processed second gated recurrent unit is input into a third gated recurrent unit, and the third recurrent gated unit is normalized as well. Figure 3 In this context, the first GRU represents the first gated recurrent unit, the second GRU represents the second gated recurrent unit, the third GRU represents the third gated recurrent unit, BN represents normalization, and ReLU represents the ReLU activation function. Please refer to [link to relevant documentation]. Figure 4 , Figure 4 This is an overall schematic diagram of training the target training data using residual convolutional neural networks, residual gated recurrent unit networks, and fully connected layers.

[0054] Understandably, residual-gated recurrent unit networks (RNNs) are able to learn more data sequence information compared to traditional convolutional neural networks. This is because the residual-gated recurrent unit network essentially has two layers of convolutional neural networks (CNNs) internally configured to... 1 Therefore, by using a residual gated recurrent unit network, the accuracy of the target model in predicting the output parameters of photovoltaic modules can be further improved.

[0055] Based on the above embodiments, this embodiment further explains and optimizes the technical solution. As a preferred implementation, the above step: the process of obtaining historical operating data of photovoltaic modules, includes:

[0056] Obtain voltage, irradiance, temperature, and current data of photovoltaic modules at historical operating times.

[0057] Understandably, in practical applications, the most important function of photovoltaic (PV) modules is power generation. Therefore, in this embodiment, a target model can be used to predict the output current of the PV module. In this case, it is necessary to obtain the operational data related to the output current of the PV module in advance. Since the output current of the PV module is most strongly correlated with the voltage, irradiance, and temperature data of the PV module, this embodiment uses the voltage, irradiance, temperature, and current data of the PV module at historical operating times to create the target model. This avoids using a large amount of invalid data when creating the target model, thereby significantly improving the speed of target model creation.

[0058] Specifically, in this embodiment, Simulink was used to acquire a total of 61,656 sets of voltage data, irradiance data, temperature data, and current data of the photovoltaic module at historical operating times. Then, 60% of these data were extracted as the training set, 20% as the validation set, and 20% as the test set.

[0059] Clearly, the technical solution provided in this embodiment can further improve the creation speed when creating a target model.

[0060] Based on the above embodiments, this embodiment further explains and optimizes the technical solution. As a preferred implementation, the above prediction method further includes:

[0061] K-fold cross-validation is used to obtain the optimal model parameters for the target model.

[0062] In this embodiment, K-fold cross-validation can also be used to obtain the optimal model parameters for the target model. Specifically, when using K-fold cross-validation, the target training data is divided into K parts, each subset being disjoint and of equal size. One subset is selected sequentially from each of the K subsets as the validation set for the target model, and the remaining K-1 subsets are used as the training set for the target model. This process is repeated K times for individual model training and validation. Finally, the average of the K validation results is used as the validation error of the target model. Therefore, K-fold cross-validation not only obtains as much effective data as possible from a limited dataset but also reduces overfitting to some extent.

[0063] Obviously, the technical solution provided in this embodiment can obtain the optimal model parameters of the target model. In this way, when using the target model to predict the output parameters of photovoltaic modules, the accuracy and reliability of the prediction results can be further improved.

[0064] Please see Figure 5 , Figure 5 This is a structural diagram of a photovoltaic module output parameter prediction device provided in an embodiment of the present invention. The device includes:

[0065] The parameter prediction module is used to predict the output parameters of photovoltaic modules using a pre-trained target model; the creation process of the target model includes:

[0066] The data selection unit 21 is used to acquire historical operating data of photovoltaic modules and select target training data from the historical operating data;

[0067] The model training unit 22 is used to train the target training data using a residual convolutional neural network, a residual gated recurrent unit network, and a fully connected layer. When the root mean square error between the output data of the fully connected layer and the preset label data reaches the minimum value, the target model is obtained. The input data and output data of the residual gated recurrent unit network are the output data of the residual convolutional neural network and the input data of the fully connected layer, respectively. Furthermore, the residual convolutional neural network is a neural network obtained by optimizing a convolutional neural network using a residual network, and the residual gated recurrent unit network is a neural network obtained by optimizing a gated recurrent unit network using a residual network.

[0068] The photovoltaic module output parameter prediction device provided in this embodiment of the invention has the beneficial effects of the photovoltaic module output parameter prediction method disclosed above.

[0069] Please see Figure 6 , Figure 6 This is a structural diagram of a photovoltaic module output parameter prediction device provided in an embodiment of the present invention. The device includes:

[0070] Memory 31 is used to store computer programs;

[0071] The processor 32 is configured to execute a computer program to implement the steps of a method for predicting the output parameters of a photovoltaic module as disclosed above.

[0072] The photovoltaic module output parameter prediction device provided in this embodiment of the invention has the beneficial effects of the photovoltaic module output parameter prediction method disclosed above.

[0073] Accordingly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a method for predicting output parameters of a photovoltaic module as disclosed above.

[0074] The computer-readable storage medium provided in this embodiment of the invention has the beneficial effects of the aforementioned method for predicting the output parameters of a photovoltaic module.

[0075] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0076] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0077] The above provides a detailed description of the method, apparatus, equipment, and medium for predicting the output parameters of a photovoltaic module provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method of predicting an output parameter of a photovoltaic assembly, characterized in that, include: The output parameters of a photovoltaic module are predicted using a pre-trained target model; wherein the creation process of the target model includes: Simulink is used to obtain the voltage, irradiance, temperature and current data of the photovoltaic module at historical operating times as historical operating data, and target training data is selected from the historical operating data. The target training data is trained using a residual convolutional neural network, a residual gated recurrent unit network, and a fully connected layer. The target model is obtained when the root mean square error between the output data of the fully connected layer and the preset label data reaches its minimum value. The input and output data of the residual gated recurrent unit network are the output data of the residual convolutional neural network and the input data of the fully connected layer, respectively. Furthermore, the residual convolutional neural network is a neural network optimized from a convolutional neural network using a residual network, and the residual gated recurrent unit network is a neural network optimized from a gated recurrent unit network using the residual network. The residual convolutional neural network includes: one first residual convolutional module and... A second residual convolutional module; wherein... The first residual convolution module is obtained by inputting the result of adding two convolution kernels into the first activation function; when When, the first second residual convolutional module is obtained by adding the output of the first residual convolutional module, two convolutional kernels, and a shortcut to the second activation function; when At that time, the first The second residual convolution module is to... The output of the second residual convolution module and the result of adding the two convolution kernels and a shortcut are input into the second activation function; wherein, the two convolution kernels in the first residual convolution module are a 3*1 convolution kernel and a 1*1 convolution kernel, and the two convolution kernels in the second residual convolution module are a 3*1 convolution kernel and a 1*1 convolution kernel. The residual gated recurrent unit network is obtained by adding the branch network and the first gated recurrent unit; wherein, the branch network is obtained by normalizing the second gated recurrent unit and processing it with the ReLU activation function, inputting the corresponding processing result into the third gated recurrent unit, and normalizing the third gated recurrent unit.

2. The prediction method according to claim 1, characterized in that, Both the first activation function and the second activation function are ReLU activation functions.

3. The prediction method according to claim 1 or 2, characterized in that, Also includes: The optimal model parameters of the target model are obtained using K-fold cross-validation.

4. A device for predicting the output parameters of a photovoltaic module, characterized in that, include: The parameter prediction module is used to predict the output parameters of photovoltaic modules using a pre-trained target model; wherein, the creation process of the target model includes: The data selection unit is used to use Simulink to acquire voltage data, irradiance data, temperature data, and current data of the photovoltaic module at historical operating times as historical operating data, and to select target training data from the historical operating data. A model training unit is used to train the target training data using a residual convolutional neural network, a residual gated recurrent unit network, and a fully connected layer. The target model is obtained when the root mean square error between the output data of the fully connected layer and the preset label data reaches its minimum value. The input and output data of the residual gated recurrent unit network are the output data of the residual convolutional neural network and the input data of the fully connected layer, respectively. The residual convolutional neural network is a neural network optimized from a convolutional neural network using a residual network, and the residual gated recurrent unit network is a neural network optimized from a gated recurrent unit network using a residual network. The residual convolutional neural network includes: one first residual convolutional module and... A second residual convolutional module; wherein... The first residual convolution module is obtained by inputting the result of adding two convolution kernels into the first activation function; when When, the first second residual convolutional module is obtained by adding the output of the first residual convolutional module, two convolutional kernels, and a shortcut to the second activation function; when At that time, the first The second residual convolution module is to... The output of the second residual convolutional module, the sum of two convolutional kernels and a shortcut, are input into the second activation function. The two convolutional kernels in the first residual convolutional module are a 3*1 kernel and a 1*1 kernel, respectively, and the two convolutional kernels in the second residual convolutional module are a 3*1 kernel and a 1*1 kernel, respectively. The residual gated recurrent unit network is obtained by adding a branch network and a first gated recurrent unit. The branch network is obtained by normalizing and applying ReLU activation to the second gated recurrent unit, inputting the corresponding processing result into a third gated recurrent unit, and then normalizing the third gated recurrent unit.

5. A device for predicting the output parameters of a photovoltaic module, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of a method for predicting output parameters of a photovoltaic module as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of a method for predicting output parameters of a photovoltaic module as described in any one of claims 1 to 3.

Images