Multi-scale hybrid photovoltaic power generation power prediction method and related equipment

Abstract

The invention provides a multi-scale hybrid photovoltaic power generation power prediction method and related equipment, and relates to the technical field of time sequence prediction. The method comprises the steps of 1, acquiring power data of a target photovoltaic power station and at least one other photovoltaic power station in an area where the target photovoltaic power station is located in the same historical time period; and step 2, inputting the obtained power data of the plurality of photovoltaic power stations in the same historical time period into a preset power prediction model to obtain the power data of the target photovoltaic power station in the future time period. The power prediction model in the invention can fully model the cross-latitude dependency relationship between the target photovoltaic power station and other photovoltaic power stations in the area range where the target photovoltaic power station is located.

Description

Technical Field

[0001] The present invention relates to the technical field of time series prediction, and in particular to a multi-scale hybrid photovoltaic power generation prediction method and related equipment. Background Art

[0002] In recent years, the proportion of photovoltaic power generation using solar energy in the energy structure has increased year by year, becoming one of the main ways to alleviate climate warming. However, the process from the emission of sunlight to its reaching the photovoltaic panel is susceptible to many factors, such as weather conditions and geographical location. Therefore, the photovoltaic power generation process is volatile and difficult to accurately predict, which in turn poses a challenge to the safe and stable operation of the power grid. To solve this problem, various photovoltaic power generation prediction models have emerged. The prediction methods can be divided into physical, statistical, artificial intelligence (including deep neural networks), integrated and hybrid categories. With the growth of computer computing power and the increase in the amount of available data, photovoltaic power generation prediction based on deep learning has become the mainstream prediction method. Among them, the deep learning-based methods can be divided into three categories according to the prediction model: recurrent neural network-based, convolutional neural network-based and attention mechanism-based.

[0003] Methods based on recurrent neural network models have the problems of gradient vanishing and gradient exploding; convolutional neural networks were originally used to extract image features, and have limitations in capturing the dependencies of one-dimensional data such as time series. Attention mechanism has become a popular method in the field of long sequence prediction. Transformer abandons recurrent connections and relies entirely on attention mechanism to capture the characteristics of long time series. Transformer model variants emerge in an endless stream and are also used to solve the problem of photovoltaic power generation prediction. It is worth noting that the above models have not made targeted improvements to the characteristics of the photovoltaic power generation process, nor have they designed modules to deeply explore the high correlation between exogenous variables and photovoltaic power generation series. At the same time, since the output power of photovoltaic power generation is closely related to weather conditions, with the development of weather forecasting technology, weather forecast data for the next few days can usually be obtained when predicting photovoltaic power generation. However, these models do not introduce weather forecast data, resulting in the lack of timeliness of the prediction model. Summary of the invention

[0004] In view of the problem that the model of the existing deep learning-based photovoltaic power generation prediction method cannot fully model the cross-dimensional dependency relationship between the target photovoltaic power station and other photovoltaic power stations in the area where it is located, the present invention provides a multi-scale hybrid photovoltaic power generation prediction method and related equipment. The constructed power prediction model can fully exploit the high correlation between multiple photovoltaic power stations at the same time step, and can take into account both coarse-grained global feature modeling and fine-grained local information extraction; in addition, an auxiliary information processing module is introduced into the model, and future meteorological forecast data is introduced and combined with real-time measurement data, which can improve the model prediction accuracy, timeliness and reliability.

[0005] In a first aspect, the present invention provides a multi-scale hybrid photovoltaic power generation prediction method, comprising:

[0006] Step 1: Obtain power data of the target photovoltaic power station and at least one other photovoltaic power station in its area in the same historical time period;

[0007] Step 2: Input the acquired power data of multiple photovoltaic power stations in the same historical time period into a preset power prediction model to obtain the power data of the target photovoltaic power station in the future time period.

[0008] Further, the preset power prediction model includes an embedding layer, a multi-scale hybrid encoder / decoder and a prediction module;

[0009] The embedding layer is used to generate an embedding representation for the time series of each photovoltaic power station, and add the position code of the photovoltaic power station to the embedding representation to generate an input vector; wherein the time series of each photovoltaic power station is obtained by arranging the power data of the photovoltaic power station in a preset time period in chronological order;

[0010] The multi-scale hybrid encoder / decoder, including an L-layer encoder and an L-layer decoder, is used to extract the global / local time pattern of the input sequence and capture the correlation between multiple photovoltaic power plants in both time and space dimensions; wherein the input sequence is constructed based on the input vectors of all photovoltaic power plants;

[0011] The prediction module is used to generate a prediction result based on the output of the multi-scale hybrid decoder.

[0012] Furthermore, the preset power prediction model also includes an auxiliary information processing module;

[0013] The auxiliary information processing module is used to generate an embedded representation of the input future weather forecast information and model the correlation between the future weather forecast information and the prediction target.

[0014] Furthermore, the input of the l-th layer encoder in the multi-scale hybrid encoder is for:

[0015]

[0016] Among them, X 0 represents the input sequence, Represents the result of downsampling the output of the l-1 layer encoder;

[0017] The output of the l-th layer encoder in the multi-scale hybrid encoder is calculated according to the following formula

[0018] X time =LayerNorm(X :,c +MLP(X :,c ))

[0019]

[0020] Among them, LayerNorm(·) represents layer normalization, X :,c A vector representing all time steps of the c-th PV plant; represents the vector of all photovoltaic power plants at the i-th time point; Dropout(·) represents the random dropout layer, X time and X dim They represent the information extraction results of the time dimension and the spatial dimension respectively, MSA represents the multi-head self-attention mechanism, and MLP represents the multi-layer perceptron.

[0021] Furthermore, the input of the l-th layer decoder in the multi-scale hybrid decoder is for:

[0022]

[0023] in, Represents the output of the lth layer of the auxiliary information processing module; represents the result of upsampling the output of the l+1th layer decoder;

[0024] The j-th output of the l-th layer decoder in the multi-scale hybrid decoder is calculated according to the following formula:

[0025]

[0026] in, represents the j-1th output of the l+1th layer, The result after upsampling; MSMD(·) represents multi-scale hybrid decoder, X outputStores the output of the first layer encoder in the multi-scale hybrid encoder.

[0027] Furthermore, the auxiliary information processing module models the correlation between future weather forecast information and the forecast target, specifically including:

[0028] Z embed =Linear(Z input )+E position

[0029]

[0030] Among them, Z embed represents the embedded representation of future weather forecast information, Z input represents the future weather forecast information sequence, Linear(·) represents the linear mapping; E position represents the position encoding of the future weather forecast information sequence; Dropout(·) represents the random dropout layer, represents the output of the encoder at layer l, Represents the output of the lth layer of the auxiliary information processing module.

[0031] Furthermore, the prediction module generates a prediction result according to the following formula:

[0032] X comb =Linear(X output )

[0033] X flatten =Flatten(X comb )

[0034] Y pred =Dropout(Linear(X flatten ))

[0035] Among them, X output represents the output of the multi-scale hybrid decoder, Linear(·) represents the linear mapping, Flatten(·) represents the flattening layer, Dropout(·) represents the random dropout layer, and Y pred Indicates the final prediction result.

[0036] In a second aspect, the present invention provides a multi-scale hybrid photovoltaic power generation prediction system, comprising:

[0037] A power data acquisition module, used to acquire power data of a target photovoltaic power station and at least one other photovoltaic power station in its area in the same historical time period;

[0038] The prediction module is used to input the acquired power data of multiple photovoltaic power stations in the same historical time period into a preset power prediction model to obtain the power data of the target photovoltaic power station in the future time period.

[0039] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0040] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0041] Beneficial effects of the present invention:

[0042] (1) The present invention obtains the power data of the target photovoltaic power station and the exogenous photovoltaic power stations in the area where it is located within a certain time period, and uses the training data to train a preset power prediction model, so that the model can establish a cross-dimensional dependency relationship between the target photovoltaic power station and the exogenous photovoltaic power station, and deeply explore the high correlation between the photovoltaic power generation power sequences of the exogenous photovoltaic power station and the target photovoltaic power station, thereby improving the prediction accuracy and reliability of the model.

[0043] (2) A multi-scale hybrid encoder / decoder is designed in the power prediction model constructed by the present invention, so that the model can capture the temporal patterns of photovoltaic power time series at different scales, extract global information at a shallow level, and model the overall trend. As the level goes deeper, finer-grained local features are continuously captured, so that the model can more accurately describe the subtle changes and fluctuations in the time series; in the multi-scale encoder / decoder, a channel mixing module is designed and an attention mechanism is adopted to make the model pay more attention to the exogenous photovoltaic power stations that are more relevant to the prediction target, and more fully consider the supporting role of the exogenous photovoltaic power stations for the prediction target.

[0044] (3) An auxiliary information processing module is also designed in the power prediction model constructed by the present invention. In the process of predicting the photovoltaic power generation of the target photovoltaic power station, the high correlation between photovoltaic power and meteorological factors is taken into account, and future meteorological data is introduced to enhance the timeliness and accuracy of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A schematic flow chart of a multi-scale hybrid photovoltaic power generation prediction method provided by an embodiment of the present invention;

[0046] Figure 2 A network architecture diagram of a power prediction model provided by an embodiment of the present invention;

[0047] Figure 3 A diagram showing the prediction results of various prediction methods provided by the embodiments of the present invention on a sunny day;

[0048] Figure 4 A graph showing prediction results of various prediction methods provided by an embodiment of the present invention on a cloudy day;

[0049] Figure 5 This is a structural block diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] like Figure 1 As shown, an embodiment of the present invention provides a multi-scale hybrid photovoltaic power generation prediction method, comprising the following steps:

[0052] S101: Obtaining power data of a target photovoltaic power station and at least one other photovoltaic power station in its area in the same historical time period

[0053] S102: performing data preprocessing on the acquired power data of multiple photovoltaic power stations in the same historical time period;

[0054] Specifically, data preprocessing mainly includes data cleaning, data missing and outlier processing, data normalization, and data set segmentation.

[0055] S103: Input the pre-processed power data of multiple photovoltaic power stations in the same historical time period into a preset power prediction model (i.e. Figure 1 The multi-scale hybrid model in is used to obtain the power data of the target PV power station in the future time period.

[0056] Specifically, the preset power prediction model includes an embedding layer, a multi-scale hybrid encoder / decoder and a prediction module. The embedding layer is used to generate an embedding representation for the time series of each photovoltaic power station, and the position encoding of the photovoltaic power station is added to the embedding representation to generate an input vector to help the model better understand the sequential relationship between the power data at each time point in the time series; the time series of each photovoltaic power station is obtained by arranging the power data of the photovoltaic power station in a preset time period in chronological order; the multi-scale hybrid encoder / decoder includes an L-layer encoder and an L-layer decoder, which are used to extract the global / local time pattern of the input sequence and capture the correlation between multiple photovoltaic power stations in two dimensions of time and space; the input sequence is constructed based on the input vectors of all photovoltaic power stations; the prediction module is used to generate prediction results based on the output of the multi-scale hybrid decoder.

[0057] The photovoltaic power prediction method provided by the embodiment of the present invention obtains the power data of the target photovoltaic power station (referred to as the target site) and other photovoltaic power stations (referred to as exogenous sites) within the area where it is located within a certain time period, and trains a preset power prediction model through training data, so that the model can establish a cross-dimensional dependency relationship between the target site and the exogenous site, and deeply explore the high correlation between the photovoltaic power generation power sequences of the exogenous site and the target site, thereby improving the prediction accuracy and reliability of the model.

[0058] In one embodiment, the training process of the preset power prediction model is:

[0059] Constructing training data includes: selecting a reference time point and a target site, and obtaining C+Z time series of multiple sites including the target site in a period before and after the reference time point; wherein, in the first C time series, each series contains the actual observation values ​​of each site at the H time points before the reference time point; and in the last Z time series, each series contains the actual observation values ​​of each site at the F time points after the reference time point.

[0060] The above C+Z time series are input into the power prediction model, and the predicted value of the target site at F time points after the reference time point is output; the loss between the F predicted values ​​output by the model and the corresponding F observed values ​​is calculated, the gradient of the parameters of each layer in the model is calculated using back propagation, and the model parameters are adjusted using the gradient descent algorithm to minimize the loss, and finally a trained model is obtained.

[0061] In one embodiment, the power prediction model also includes an auxiliary information processing module (AIPM); the auxiliary information processing module is used to generate an embedded representation of the input future weather forecast information and model the correlation between the future weather forecast information and the prediction target.

[0062] It is understandable that weather has a significant impact on the power of photovoltaic power stations. Therefore, the accuracy and reliability of model predictions can be further improved by introducing future weather forecast information.

[0063] Based on the above embodiments, the input of the first layer encoder in the multi-scale hybrid encoder is for:

[0064]

[0065] Among them, X 0 Represents the input sequence, which can be expressed as:

[0066] X 0 =Embed value (X input )+Embed postion (2)

[0067] Among them, X input Represents the original time series data of the input, Embed value (·) represents a linear value embedding layer; postion represents the corresponding position code, H is the original sequence length of the input sequence, dm is the dimension of the latent space (i.e., the dimension in which the time series data is embedded), and C refers to the number of input features; It represents the result of downsampling the output of the encoder at the l-1th layer; it can be understood that downsampling is used to merge the data at consecutive time points in the window according to the preset window size, and the total number of elements in the output feature sequence will be reduced after downsampling. For the convenience of description, S l is the number of output data of the lth downsampling layer, then

[0068] In the embodiment of the present invention, for input The downsampling of The process is expressed by formula (3):

[0069]

[0070] where l∈[1,L), is the output of the l-th layer encoder in the multi-scale hybrid encoder; Avgpool1d(·) is a one-dimensional pooling layer.

[0071] The output of the l-th layer encoder in the multi-scale hybrid encoder is calculated according to the following formula

[0072] X time =LayerNorm(X :,c +MLP(X :,c )) (4)

[0073]

[0074] Among them, LayerNorm(·) represents layer normalization, X :,c represents the vector of all time steps of the c-th PV plant, represents the vector of all photovoltaic power stations at the i-th time point, Dropout(·) represents a random dropout layer (which reduces overfitting by randomly dropping some neurons during training); X time Represents the extraction result of time dimension information, Taking the spatial location of the photovoltaic power station as a variable, then X dim represents the spatial dimension information extraction result, which is used to represent the correlation between photovoltaic power stations at different spatial locations; X time and X dim All are intermediate vectors; MSA stands for multi-head self-attention mechanism, and MLP stands for multi-layer perceptron.

[0075] From the above formulas (4) to (6), it can be seen that the multi-scale hybrid encoder mainly uses MLP to mix the input sequence in the time dimension first, and then uses MSA and MLP to mix it in the channel dimension.

[0076] The input of the lth layer decoder in the multi-scale hybrid decoder for:

[0077]

[0078] Among them, Z o l ut represents the output of the lth layer of the auxiliary information processing module, represents the result of upsampling the output of the l+1th layer decoder, It can be understood that upsampling is used to improve the resolution of the time series and to fuse the output information of different decoder layers in the multi-scale hybrid decoder.

[0079] In the embodiment of the present invention, for input and The upsampling process is described by the following formula (8):

[0080]

[0081] where l∈[1,L);m∈[1,L-l+1], Represents the mth of the L-l+1 output results of the l+1th layer decoder.

[0082] The j-th output of the l-th layer decoder in the multi-scale hybrid decoder is calculated according to the following formula:

[0083]

[0084] in, Represents the j-1th output of the l+1th layer The result after upsampling; MSMD(·) represents a multi-scale hybrid decoder, whose feature extraction process is the same as the calculation process of the time and channel hybrid units in the above encoder. Refer to formula (4) to formula (6), that is, the input of each layer of the decoder is regarded as X in formula (4), and the output result obtained after sequentially undergoing formula (4) to formula (6) is X output Stores the output of the first layer encoder in the multi-scale hybrid encoder,

[0085] The auxiliary information processing module models the correlation between future weather forecast information and the forecast target, specifically including:

[0086] Z embed =Linear(Z input )+E position (11)

[0087]

[0088] Among them, Z embed represents the embedded representation of future weather forecast information, Z input represents the future weather forecast information sequence, Z refers to the number of input weather forecast variables, F is the length of the weather forecast information sequence, and also the length of the power sequence to be predicted; Linear(·) represents linear mapping; E position Represents the position code of the future weather forecast information sequence, Dropout(·) represents a random dropout layer. represents the output of the encoder at layer l, Represents the output of the lth layer of the auxiliary information processing module.

[0089] The prediction module generates prediction results according to the following formula:

[0090] X comb =Linear(X output ) (14)

[0091] X flatten =Flatten(X comb ) (15)

[0092] Y pred =Dropout(Linear(X flatten )) (16)

[0093] Among them, X output represents the output of the multi-scale hybrid decoder, Linear(·) represents the linear mapping, Flatten(·) represents the flattening layer, Dropout(·) represents the random dropout layer, Indicates the final prediction result.

[0094] Corresponding to the above method, an embodiment of the present invention further provides a multi-scale hybrid photovoltaic power generation prediction system, including: a power data acquisition module and a prediction module.

[0095] Among them, the power data acquisition module is used to obtain the power data of the target photovoltaic power station and at least one other photovoltaic power station in the area where it is located in the same historical time period; the prediction module is used to input the acquired power data of multiple photovoltaic power stations in the same historical time period into a preset power prediction model to obtain the power data of the target photovoltaic power station in the future time period.

[0096] It should be noted that the photovoltaic power generation prediction system provided in the embodiment of the present invention is to implement the above method. The specific functions thereof can be referred to the above method embodiments, which will not be described in detail here.

[0097] In order to verify the effectiveness of the solution of the present invention, the present invention also provides the following experimental data.

[0098] The experiment was conducted in a single GPU (NVIDIA GeForce RTX 3080) and PyTorch environment. The power data of Yongle Photovoltaic Power Station in Pan'an County, Jinhua City, Zhejiang Province, China was used for the experiment. The actual power data of the photovoltaic power generation system for one year from January 1, 2022 to January 1, 2023 was selected. 24 sample points were selected every day, and the data time resolution was 1h. The comparison of prediction errors with other methods is shown in Table 1.

[0099] Table 1 Comparison of prediction results of each model under different validation indicators

[0100]

[0101] In Table 1, the optimal prediction results are bolded and the suboptimal results are underlined. Invention-1 represents a prediction model that uses only measured photovoltaic power data, which is superior to the existing model in the evaluation of the three indicators. Invention-2 is based on Invention-1 and introduces an auxiliary information processing module to effectively process meteorological forecast data. The difference between the prediction results and the actual values ​​is minimal, which verifies the effectiveness of the invention. Figure 3 and Figure 4 These are the prediction results of various prediction methods on sunny and cloudy days respectively.

[0102] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 501, a communication interface 502, a memory 503 and a communication bus 504, wherein the processor 501, the communication interface 502 and the memory 503 communicate with each other through the communication bus 504. The processor 501 may call the logic instructions in the memory 503 to execute a multi-scale hybrid photovoltaic power generation power prediction method, the method comprising: step 1: obtaining power data of a target photovoltaic power station and at least one other photovoltaic power station in its region in the same historical time period; step 2: inputting the obtained power data of multiple photovoltaic power stations in the same historical time period into a preset power prediction model to obtain the power data of the target photovoltaic power station in the future time period.

[0103] In addition, when the logic instructions in the above-mentioned memory 503 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0104] An embodiment of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments, for example, including: Step 1: Obtain power data of a target photovoltaic power station and at least one other photovoltaic power station in its area in the same historical time period; Step 2: Input the acquired power data of multiple photovoltaic power stations in the same historical time period into a preset power prediction model to obtain the power data of the target photovoltaic power station in the future time period.

[0105] An embodiment of the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the methods provided by the above-mentioned method embodiments are implemented, for example, including: Step 1: Obtaining power data of a target photovoltaic power station and at least one other photovoltaic power station in its area in the same historical time period; Step 2: Inputting the acquired power data of multiple photovoltaic power stations in the same historical time period into a preset power prediction model to obtain the power data of the target photovoltaic power station in the future time period.

[0106] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-scale hybrid photovoltaic power generation prediction method, characterized in that: include: Step 1: Obtain power data of the target photovoltaic power station and at least one other photovoltaic power station in its area in the same historical time period; Step 2: Input the acquired power data of multiple photovoltaic power stations in the same historical time period into a preset power prediction model to obtain the power data of the target photovoltaic power station in the future time period.

2. A multi-scale hybrid photovoltaic power generation prediction method according to claim 1, characterized in that: The preset power prediction model includes an embedding layer, a multi-scale hybrid encoder / decoder and a prediction module; The embedding layer is used to generate an embedding representation for the time series of each photovoltaic power station, and add the position code of the photovoltaic power station to the embedding representation to generate an input vector; wherein the time series of each photovoltaic power station is obtained by arranging the power data of the photovoltaic power station in a preset time period in chronological order; The multi-scale hybrid encoder / decoder, including an L-layer encoder and an L-layer decoder, is used to extract the global / local time pattern of the input sequence and capture the correlation between multiple photovoltaic power plants in both time and space dimensions; wherein the input sequence is constructed based on the input vectors of all photovoltaic power plants; The prediction module is used to generate a prediction result based on the output of the multi-scale hybrid decoder.

3. A multi-scale hybrid photovoltaic power generation prediction method according to claim 2, characterized in that: The preset power prediction model also includes an auxiliary information processing module; The auxiliary information processing module is used to generate an embedded representation of the input future weather forecast information and model the correlation between the future weather forecast information and the prediction target.

4. A multi-scale hybrid photovoltaic power generation prediction method according to claim 2 or 3, characterized in that: The input of the lth layer encoder in the multi-scale hybrid encoder is for: Among them, X 0 represents the input sequence, Represents the result of downsampling the output of the l-1 layer encoder; The output of the l-th layer encoder in the multi-scale hybrid encoder is calculated according to the following formula X time =LayerNorm(X :,c +MLP(X :,c )) Among them, LayerNorm(·) represents layer normalization, X :,c A vector representing all time steps of the c-th PV plant; represents the vector of all photovoltaic power plants at the i-th time point; Dropout(·) represents the random dropout layer, X time and X dim They represent the information extraction results of the time dimension and the spatial dimension respectively, MSA represents the multi-head self-attention mechanism, and MLP represents the multi-layer perceptron.

5. The multi-scale hybrid photovoltaic power generation prediction method according to claim 3, characterized in that: The input of the lth layer decoder in the multi-scale hybrid decoder for: in, Represents the output of the lth layer of the auxiliary information processing module; represents the result of upsampling the output of the l+1th layer decoder; The j-th output of the l-th layer decoder in the multi-scale hybrid decoder is calculated according to the following formula: in, represents the j-1th output of the l+1th layer, The result after upsampling; MSMD(·) represents multi-scale hybrid decoder, X output Stores the output of the first layer encoder in the multi-scale hybrid encoder.

6. A multi-scale hybrid photovoltaic power generation prediction method according to claim 3 or 5, characterized in that: The auxiliary information processing module models the correlation between future weather forecast information and the forecast target, specifically including: Z embed =Linear(Z input )+E position Among them, Z embed represents the embedded representation of future weather forecast information, Z input represents the future weather forecast information sequence, Linear(·) represents the linear mapping; E position represents the position encoding of the future weather forecast information sequence; Dropout(·) represents the random dropout layer, represents the output of the encoder at layer l, Represents the output of the lth layer of the auxiliary information processing module.

7. A multi-scale hybrid photovoltaic power generation prediction method according to claim 2, characterized in that: The prediction module generates prediction results according to the following formula: X comb =Linear(X output ) X flatten =Flatten(X comb ) Y pred =Dropout(Linear(X flatten )) Among them, X output represents the output of the multi-scale hybrid decoder, Linear(·) represents the linear mapping, Flatten(·) represents the flattening layer, Dropout(·) represents the random dropout layer, and Y pred Indicates the final prediction result.

8. A multi-scale hybrid photovoltaic power generation prediction system, characterized in that: include: A power data acquisition module, used to acquire power data of a target photovoltaic power station and at least one other photovoltaic power station in its area in the same historical time period; The prediction module is used to input the acquired power data of multiple photovoltaic power stations in the same historical time period into a preset power prediction model to obtain the power data of the target photovoltaic power station in the future time period.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images