Photovoltaic power generation prediction method, device and storage medium

Abstract

The present application relates to the field of photovoltaic power generation prediction technology, and in particular to a photovoltaic power generation prediction method, device and storage medium. The photovoltaic power generation prediction method includes: inputting historical real weather characteristics and historical real power generation into a trained teacher model to obtain a teacher model feature vector and teacher model predicted power generation; inputting historical predicted weather characteristics and historical real power generation into a student model to be trained to obtain a student model feature vector and student model predicted power generation; training the student model according to the teacher model feature vector, teacher model predicted power generation, student model feature vector, student model predicted power generation and historical real power generation to obtain a trained student model; and predicting photovoltaic power generation according to the trained student model. The method of the present application adds a real weather-assisted learning method when learning the relationship between predicted weather characteristics and power generation, thereby improving the robustness and accuracy of the model.

Description

Technical Field

[0001] The present application relates to the technical field of photovoltaic power generation prediction, and in particular to a photovoltaic power generation prediction method, device, and storage medium. Background Art

[0002] As a clean, renewable energy source, solar energy is gaining increasing attention and recognition worldwide. With the continuous development and application of photovoltaic technology, photovoltaic power generation has become a crucial component of the new energy sector. Accurately predicting photovoltaic power generation is crucial to maintaining a stable balance between electricity consumption and stored energy. The accuracy of photovoltaic power generation predictions directly depends on the accuracy of weather forecasts (radiation, temperature, humidity, wind speed, cloud cover, etc.). Due to inherent errors in weather forecasts, predictions of photovoltaic power generation based on these weather characteristics are also inaccurate, resulting in high errors.

[0003] With the development of artificial intelligence, traditional machine learning and deep learning have become common methods for photovoltaic power generation forecasting. In recent years, the mainstream approach has been to base model design and optimization on historical photovoltaic power generation data and relatively accurate weather data corrected by weather platforms (without paying attention to forecasted weather). However, in actual implementation, modeling and forecasting are based on forecasted weather data. This approach fails to account for the significant discrepancies between forecasted and actual weather conditions, resulting in large forecast errors at the time of implementation. Summary of the Invention

[0004] The embodiments of the present application aim to provide a photovoltaic power generation prediction method, device and storage medium to solve the problem in the prior art that the photovoltaic power generation prediction model does not pay attention to the large difference between the predicted weather and the actual weather, resulting in errors in the prediction results.

[0005] To solve the above technical problems, the embodiments of the present application provide the following technical solutions:

[0006] According to a first aspect of the present application, a photovoltaic power generation prediction method is provided, comprising:

[0007] Obtain historical forecast weather characteristics, historical actual weather characteristics and historical actual power generation;

[0008] Inputting the historical real weather characteristics and the historical real power generation into the trained teacher model to obtain the teacher model feature vector and the teacher model predicted power generation;

[0009] Inputting the historical predicted weather characteristics and the historical actual power generation into the student model to be trained to obtain a student model feature vector and a student model predicted power generation;

[0010] The student model is trained by calculating the multivariate total loss based on the teacher model feature vector, the teacher model predicted power generation, the student model feature vector, the student model predicted power generation and the historical actual power generation to obtain a trained student model;

[0011] Photovoltaic power generation is predicted based on the trained student model.

[0012] Optionally, the teacher model is trained based on real historical weather characteristics.

[0013] Optionally, the calculating of the multivariate total loss according to the teacher model feature vector, the teacher model predicted power generation, the student model feature vector, the student model predicted power generation, and the historical actual power generation includes:

[0014] Calculating a distillation loss based on the teacher model feature vector and the student model feature vector;

[0015] Calculating a soft label loss based on the power generation predicted by the student model and the power generation predicted by the teacher model;

[0016] Calculating the true label loss based on the power generation predicted by the student model and the historical true power generation;

[0017] The multivariate total loss is calculated based on the distillation loss, the soft label loss, and the true label loss.

[0018] Optionally, the calculation formula for the multivariate total loss is:

[0019]

[0020] in, is the total multivariate loss, is the distillation loss, is the soft label loss, is the true label loss, 、 and is the weight coefficient, is the probability distribution of the teacher model feature vector, is the probability distribution of the student model feature vector, is the sample size, It is The true value of the sample, For the teacher model The predicted value of the sample, For students to model The predicted value of the sample.

[0021] Optionally, the teacher model and the student model have the same network structure.

[0022] Optionally, the network structure includes a global feature module, a feature pyramid module, a time series module, a multi-head attention module and a high-level feature expression module connected in sequence.

[0023] Optionally, the global feature module uses dilated convolution and large convolution kernel to extract features.

[0024] Optionally, the feature pyramid module inputs the features extracted in the previous stage into multiple branches, each branch uses a dilated convolution with a different expansion rate to extract features, and then splices and superimposes the features of multiple branches by channel.

[0025] According to the second aspect of the present application, an electronic device is provided, comprising at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the photovoltaic power generation prediction method described above.

[0026] According to a third aspect of the present application, a computer storage medium is provided, wherein the computer storage medium stores instructions or programs, and when the instructions or programs are executed by at least one processor, the at least one processor executes the photovoltaic power generation prediction method described above.

[0027] The beneficial effects of the embodiments of the present application are as follows: Different from the prior art, the embodiments of the present application provide a photovoltaic power generation prediction method, which first obtains historical predicted weather characteristics, historical real weather characteristics and historical real power generation; then inputs the historical real weather characteristics and historical real power generation into a trained teacher model to obtain the teacher model feature vector and the teacher model predicted power generation, inputs the historical predicted weather characteristics and historical real power generation into a student model to be trained to obtain the student model feature vector and the student model predicted power generation; then, the multivariate total loss is calculated based on the teacher model feature vector, the teacher model predicted power generation, the student model feature vector and the student model predicted power generation to train the student model, and a trained student model is obtained; finally, photovoltaic power generation is predicted based on the trained student model. The method of the present application adds a real weather-assisted learning method when learning the relationship between predicted weather characteristics and power generation, improves the robustness of the model, and reduces the problem of cross-domain accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] One or more embodiments are exemplarily illustrated by pictures in the corresponding drawings. These exemplifications do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings are represented as similar elements. Unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0029] Figure 1 An exemplary diagram of a training framework of a teacher model is shown;

[0030] Figure 2 An exemplary diagram of a training framework of a student model is shown;

[0031] Figure 3 The module diagram of the feature extraction network structure is shown as an example;

[0032] Figure 4 The structural diagram of the global feature module is shown as an example;

[0033] Figure 5 The structural diagram of the feature pyramid module is shown as an example;

[0034] Figure 6 A schematic diagram of a photovoltaic power generation prediction method is shown as an example;

[0035] Figure 7 The diagram exemplarily shows a structural diagram of an electronic device. DETAILED DESCRIPTION

[0036] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0037] In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0038] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0039] The key terms in this application are explained as follows:

[0040] Historical weather characteristics: The weather characteristics that were finally recorded at a certain time in the past. For example, if it is day T, and the weather characteristics that actually occurred on day T-2 have been recorded by various meteorological sensors, then these actual weather records on day T-2 are called "historical" weather.

[0041] Historical weather forecasts are those that have been predicted for the future. For example, suppose it's day T, and on day T-3, a weather forecast was made for day T-2. Since both day T-3 and day T-2 have already occurred, the weather forecast for day T-2 on day T-3 is called "historical forecast" weather.

[0042] Prelu (Parametric rectified linear unit) activation function: It is an improved ReLU activation function that automatically adjusts the slope of the negative interval by introducing learnable parameters, solving the problem of the traditional ReLU having a gradient of 0 in the negative interval (the "neuron death" problem).

[0043] The embodiment of the present application proposes a photovoltaic power generation prediction method based on a knowledge transfer model, and the knowledge transfer model includes a student model and a teacher model. Among them, the student model and the teacher model have the same network structure and different input features. By transferring the pre-trained teacher model knowledge to the student model, the student model can obtain more knowledge from the teacher model, thereby improving the robustness and accuracy of photovoltaic power generation prediction. Specifically, the teacher model is trained in real weather features of relatively high quality, and the model has high accuracy and robustness. The student model is trained based on the predicted weather features, and the feature vectors and prediction results output by the teacher model are used as the student's training targets, so that the student model can pay attention to the differences between the predicted weather features and the real weather features, obtain more knowledge, and thus have higher robustness in future power generation forecasts.

[0044] Please refer to Figure 1 , shows a training framework diagram of a teacher model. Figure 1As shown in Figure 1, the training process of the teacher model is as follows: (1) the real weather features are input into the teacher model; (2) the real weather features are extracted by the feature extraction network structure and then input into the high-level feature expression module for compression processing to obtain the teacher model feature vector; (3) the teacher model prediction results (i.e., the teacher model predicted power generation) are obtained based on the teacher model feature vector; (4) the loss is calculated based on the teacher model prediction results and the true label (i.e., the historical true power generation corresponding to the real weather features), and the teacher model is trained based on the loss to obtain a trained teacher model. Among them, the real weather features are relatively accurate weather features that have been corrected by history. The teacher model feature vector is a feature vector with a preset fixed length, and its main purpose is to achieve knowledge transfer between the teacher model and the student model.

[0045] Please refer to Figure 2 , shows a training framework diagram of a student model. Figure 2 As shown, the training process of the student model is: (1) Load Figure 1 The trained teacher model is input into the student model; (2) the predicted weather features are input into the student model, and the real weather features corresponding to the predicted weather features are input into the teacher model; (3) the teacher model obtains the teacher model feature vector and the teacher model prediction results based on the real weather features, and the student model obtains the student model feature vector and the student model prediction results based on the predicted weather features; (4) the multivariate total loss is calculated based on the distillation loss between the teacher model feature vector and the student model feature vector, the soft label loss between the teacher model prediction results and the student model prediction results, and the real label loss between the student model prediction results and the real label, and the student model is trained based on the multivariate total loss to obtain a trained student model. It should be noted that when training the student model, the teacher model only performs inference and does not perform training. Using the multivariate loss function in the student model training process can enable the student model to learn knowledge from different fields and improve the robustness of the model. It is understandable that the weather features used to train the teacher model can be in different time periods from the weather features used to train the student model. For example, assuming it is September now, the weather features used to train the teacher model can be the historical real weather features from January to June, and the weather features used to train the student model can be the historical predicted weather features and historical real weather features from May to August.

[0046] Please refer to Figure 3 , which shows the module diagram of the feature extraction network structure. Figure 3As shown in the figure, the feature extraction network architecture mainly includes a global feature module, a feature pyramid module, a time series module, and a multi-head attention module. The global feature module uses dilated convolutions and large convolution kernels to extract features, increasing the receptive field and capturing multi-scale contextual information. The feature pyramid module processes the input feature map using dilated convolution layers with different dilation rates. This enables the network to capture combinations of weather features at different scales, effectively increasing the model's receptive field without significantly increasing computational cost.

[0047] Please refer to Figure 4 , shows the structure diagram of the global feature module. Figure 4 As shown in Figure 2, the processing process of the global feature module is as follows: (1) Input weather features, i.e., real weather features or predicted weather features. (2) Perform dilated convolution with a kernel size of 15 and a dilation rate of 1. (3) Perform batch normalization layer processing. (4) Perform nonlinear mapping through the Prelu activation function. (5) Perform dilated convolution again with a kernel size of 9 and a dilation rate of 1. (6) Perform batch normalization layer processing again. (4) Perform nonlinear mapping through the Prelu activation function again.

[0048] Please refer to Figure 5 , shows the structure diagram of the feature pyramid module. Figure 5 As shown in Figure 2, the processing process of the feature pyramid module is as follows: (1) input the features extracted in the previous stage; (2) input them to multiple branches in parallel, each of which contains a hole convolution layer, a batch normalization layer and a Prelu activation function, for example, Figure 5 The features extracted in the previous stage are input into four branches. The size of the dilated convolution kernel of the first branch is 1, and the dilation rate is 1; the size of the dilated convolution kernel of the second branch is 3, and the dilation rate is 12; the size of the dilated convolution kernel of the third branch is 3, and the dilation rate is 24; the size of the dilated convolution kernel of the fourth branch is 3, and the dilation rate is 36; (3) The features of the four branches are spliced and superimposed according to the channel.

[0049] The time series module extracts temporal features from historical weather patterns using a long short-term memory (LSTM) network, capturing the temporal relationship between PV power generation and historical weather characteristics. This module allows the model to better focus on the following features: 1) patterns in PV power generation over time; 2) cyclical characteristics of PV power generation, such as seasonal variations and long-term trends; 3) the temporal continuity of PV power generation; and 4) mitigates issues such as gradient shrinkage and gradient explosion during computation. In one embodiment, the time series module employs a two-layer LSTM architecture.

[0050] The multi-head attention module is used to focus on different parts of the input sequence from multiple different angles or representation spaces. The features and models learned by each attention head may be different, which helps the model to fully understand the data. By using multiple independent attention heads, the model can learn a more diverse combination of features, each focusing on different information, and ultimately merging this information, thereby enhancing the overall expressiveness of the model. Since the attention mechanism can directly connect any two positions in the sequence, it is particularly suitable for modeling long-distance dependencies. This is very useful for capturing long-term patterns such as seasonality and trends in time series. In one embodiment, the multi-head attention module uses 10 attention heads.

[0051] In an embodiment of the present application, introducing a combination of a time series module and a multi-head attention module into the feature extraction network structure can enable the model to better focus on temporal and spatial features, thereby improving the robustness of the model.

[0052] The advanced feature representation module uses a fully connected layer to convert the input high-dimensional feature vector into a low-dimensional feature vector of fixed length through a series of linear transformations. Because the fully connected layer linearly combines all features extracted by the feature extraction network structure, it can perform feature compression, condensing complex feature information into a relatively low-dimensional space, thereby enabling knowledge transfer between the teacher model and the student model.

[0053] In one embodiment, a point convolution module is included between the feature pyramid module and the time series module. This module performs point convolution on the feature map output by the feature pyramid module to reduce the number of channels in the feature map, thereby reducing the computational complexity of subsequent layers. 1x1 convolution allows for linear combinations between different channels, achieving cross-channel information fusion and enhancing the expressiveness of features.

[0054] The following uses the historical weather characteristics of one day as an example to illustrate how the teacher model or student model processes historical weather characteristics. The details are as follows:

[0055] Step 1: Input historical weather features with dimensions of (1, 48, 200).

[0056] Here, 1 represents the input time unit day, 48 represents 12 hours of daytime, each hour is calculated as 15 minutes, and 200 represents the weather characteristics.

[0057] Step 2: Extract features through the feature extraction network structure, including:

[0058] a) Extract features through the global feature module and output a feature map with a dimension of (1, 192, 50);

[0059] b) Extract features through the feature pyramid module, and each branch outputs a feature map with a dimension of (1, 192, 50);

[0060] c) Merge the four branches by channel and output a feature map of dimension (1, 768, 50);

[0061] d) Perform dimensionality reduction through point convolution, outputting a feature map with dimensions of (1, 192, 50);

[0062] e) Extract features through LSTM and output a feature map with a dimension of (1, 192, 100);

[0063] f) Extract features through multi-head attention and output a feature map with a dimension of (1, 100);

[0064] g) Extract features through the advanced feature expression module and output a feature map with a dimension of (1, 96).

[0065] Step 3: Output the prediction result, a feature map with a dimension of (1, 48).

[0066] Please refer to Figure 6 , Figure 6 A schematic flow chart of a photovoltaic power generation prediction method is shown, the method comprising:

[0067] S601, obtaining historical forecast weather characteristics, historical actual weather characteristics and historical actual power generation.

[0068] Specifically, the historical forecast meteorological data of the photovoltaic power station, the historical real meteorological data after the historical forecast meteorological data is corrected, the historical power generation data, the relevant time data, etc. are collected to obtain the dimension A B C's historical forecast weather characteristics, dimension A B The historical real weather characteristics and dimensions of C are A The historical actual power generation of B. Where A is the number of days, B is the number of sunshine intervals per day, and C is the number of weather characteristics that affect power generation. Weather characteristics that affect power generation may include temperature, radiation, air pressure, wind speed, etc. For example, the historical forecast weather characteristics are: {Date: October 20, 2024,

[0069] Time interval: 1, temperature: 28℃, humidity: 65%, air pressure: 1008 Pa, ...,

[0070] Time interval: 2, temperature: 28℃, humidity: 65%, air pressure: 1008 Pa, ...,

[0071] …

[0072] Time interval: 48, temperature: 25℃, humidity: 55%, air pressure: 1008 Pa, ...,}.

[0073] The historical actual power generation corresponding to the historical forecast weather feature is: {2.5, ..., 10}, a total of 48 values. The historical actual weather feature corresponds to the historical forecast weather feature and is a relatively accurate weather feature corrected based on the historical forecast weather feature using real meteorological data. The historical actual power generation is the actual power generation for the date and time interval corresponding to the historical forecast weather feature / historical actual weather feature. In other words, a training sample for the student model includes the historical actual weather feature, the historical forecast weather feature, and the historical actual power generation (i.e., the true label).

[0074] S602: Input historical real weather characteristics and historical real power generation into the trained teacher model to obtain the teacher model feature vector and the teacher model predicted power generation.

[0075] Specifically, the teacher model includes Figures 3 to 5 The feature extraction network structure and high-level feature expression module shown in the figure are shown in the figure. The trained teacher model is trained based on historical real weather features. The loss function used in training is the MAE function (Mean Absolute Error), and the formula is:

[0076]

[0077] in, is the sample size, It is The true value of the sample, For the The predicted value of the sample.

[0078] During the student model training process, the teacher model is only used to make predictions based on historical weather patterns and is not trained. After the historical weather patterns are fed into the trained teacher model, the output is the teacher model feature vector and the teacher model's predicted power generation.

[0079] S603: Input historical forecast weather characteristics and historical actual power generation into the student model to be trained to obtain a student model feature vector and student model forecast power generation.

[0080] Specifically, the student model also includes Figures 3 to 5 The feature extraction network structure and high-level feature expression module shown in the figure. The historical forecast weather features are extracted by the feature extraction network structure and high-level feature expression module to obtain the student model feature vector, and the student model predicted power generation is obtained based on the student model feature vector.

[0081] Step S604 , calculating the multivariate total loss based on the teacher model feature vector, the teacher model predicted power generation, the student model feature vector, the student model predicted power generation and the historical actual power generation to train the student model and obtain a trained student model.

[0082] In one embodiment, a multivariate loss function is constructed to train the student model, so that the student model can learn knowledge from different fields. The multivariate loss function consists of a distillation loss function, a soft label loss function, and a true label loss function. The distillation loss uses the KL divergence (Kullback-Leibler Divergence, also known as relative entropy) loss function, which calculates the distillation loss by calculating the distribution difference between the teacher model feature vector and the student model feature vector. The formula is as follows:

[0083]

[0084] in, is the probability distribution of the teacher model feature vector, is the probability distribution of the student model feature vector. Specifically, The teacher model feature vector or the student model feature vector is Softmax normalized to the teacher model feature vector or the student model feature vector, that is, the feature vector is converted into a probability distribution. The Softmax formula is:

[0085]

[0086] in, =[ ] is the eigenvector, For the The probability of each category. After Softmax normalization, the probability of all categories is 1, that is, =1.

[0087] The soft label loss uses the MAE function, with the power generation predicted by the student model as the predicted value and the power generation predicted by the teacher model as the label to calculate the soft label loss. The formula is as follows:

[0088]

[0089] in, For the teacher model The predicted value of the sample, For students to model The predicted value of the sample.

[0090] The true label loss also uses the MAE function, with the power generation predicted by the student model as the predicted value and the historical true power generation as the label to calculate the true label loss. The formula is as follows:

[0091]

[0092] in, It is The true value of the sample, For students to model The predicted value of the sample.

[0093] In one embodiment, the multivariate total loss The calculation formula is:

[0094]

[0095] in, 、 and is the weight coefficient, and + + =1. Generally, > > .

[0096] Step S605 , predicting photovoltaic power generation based on the trained student model.

[0097] The photovoltaic power generation prediction method provided by the embodiment of the present application first obtains historical predicted weather characteristics, historical real weather characteristics and historical real power generation; then the historical real weather characteristics and historical real power generation are input into the trained teacher model to obtain the teacher model feature vector and the teacher model predicted power generation; the historical predicted weather characteristics and historical real power generation are input into the student model to be trained to obtain the student model feature vector and the student model predicted power generation; then, the multivariate total loss is calculated based on the teacher model feature vector, the teacher model predicted power generation, the student model feature vector and the student model predicted power generation to train the student model to obtain a trained student model; finally, photovoltaic power generation is predicted based on the trained student model. The method of the present application adds a real weather-assisted learning method when learning the relationship between predicted weather characteristics and power generation, improves the robustness of the model, and reduces the problem of cross-domain accuracy.

[0098] According to an embodiment of the present application, an electronic device is provided, such as Figure 7 , which is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application, includes a processor 10, a communication interface 30, a memory 20, and a communication bus. The processor 10, communication interface 30, and memory 20 communicate with each other via the communication bus. The processor 10 can invoke logic instructions in the memory 20 to execute the aforementioned photovoltaic power generation prediction method.

[0099] In addition, the logic instructions in the above-mentioned memory 20 can be implemented in the form of a software functional unit and can be stored in several computer-readable storage media when sold or used as an independent product. Based on this understanding, the technical solution of the present application, or the part that contributes to the existing technology, or the part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the above-mentioned photovoltaic power generation prediction method of the present application. The above-mentioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0100] According to an embodiment of the present application, a computer-readable storage medium is provided, the type of which is as described above, and the computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor executes the steps of the photovoltaic power generation prediction method described above.

[0101] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the relevant technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or certain portions of the embodiments.

[0102] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of variations or substitutions within the technical scope disclosed in the present application. Therefore, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A photovoltaic power generation prediction method, characterized in that: The method comprises: Obtain historical forecast weather characteristics, historical actual weather characteristics and historical actual power generation; Inputting the historical real weather characteristics and the historical real power generation into the trained teacher model to obtain the teacher model feature vector and the teacher model predicted power generation; Inputting the historical predicted weather characteristics and the historical actual power generation into the student model to be trained to obtain a student model feature vector and a student model predicted power generation; The student model is trained by calculating the multivariate total loss based on the teacher model feature vector, the teacher model predicted power generation, the student model feature vector, the student model predicted power generation and the historical actual power generation to obtain a trained student model; Photovoltaic power generation is predicted based on the trained student model.

2. The method according to claim 1, characterized in that The teacher model is trained based on historical real weather characteristics.

3. The method according to claim 1, characterized in that The calculating of the multivariate total loss according to the teacher model feature vector, the teacher model predicted power generation, the student model feature vector, the student model predicted power generation and the historical actual power generation comprises: Calculating a distillation loss based on the teacher model feature vector and the student model feature vector; Calculating a soft label loss based on the power generation predicted by the student model and the power generation predicted by the teacher model; Calculating the true label loss based on the power generation predicted by the student model and the historical true power generation; The multivariate total loss is calculated based on the distillation loss, the soft label loss, and the true label loss.

4. The method according to claim 3, characterized in that The calculation formula of the multivariate total loss is: in, is the total multivariate loss, is the distillation loss, is the soft label loss, is the true label loss, 、 and is the weight coefficient, is the probability distribution of the teacher model feature vector, is the probability distribution of the student model feature vector, is the sample size, It is The true value of the sample, For the teacher model The predicted value of the sample, For students to model The predicted value of the sample.

5. The method according to any one of claims 1 to 4, characterized in that The teacher model and the student model have the same network structure.

6. The method according to claim 5, characterized in that The network structure includes a global feature module, a feature pyramid module, a time series module, a multi-head attention module and a high-level feature expression module connected in sequence.

7. The method according to claim 6, characterized in that The global feature module uses dilated convolution and large convolution kernel to extract features.

8. The method according to claim 6, characterized in that The feature pyramid module inputs the features extracted in the previous stage into multiple branches. After each branch extracts features using a dilated convolution with a different expansion rate, the features of multiple branches are spliced and superimposed by channel.

9. An electronic device, characterized in that: The method comprises at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the photovoltaic power generation prediction method according to any one of claims 1 to 8.

10. A computer storage medium, characterized in that The computer storage medium stores instructions or programs, and when the instructions or programs are executed by at least one processor, the at least one processor is caused to execute the photovoltaic power generation prediction method according to any one of claims 1 to 8.

Images