A method, apparatus, and equipment for spatiotemporal prediction of distributed photovoltaic power based on scene-conditional graph attention.

Abstract

This invention relates to the field of photovoltaic power prediction technology, and discloses a method, apparatus, and device for spatiotemporal prediction of distributed photovoltaic power based on scene-conditional graph attention. The method includes acquiring historical power sequences and meteorological time series of distributed photovoltaic power plants; constructing an uncertainty feature system; clustering historical data to obtain similar uncertainty scenarios; constructing a spatiotemporal graph network including a graph attention module, a convolutional feature extraction module, and a time modeling module; injecting similar uncertainty scenarios into the graph attention module to extract spatial features of the distributed photovoltaic power plants, obtaining scene-aware spatial features; inputting meteorological time series into the convolutional feature extraction module for feature extraction, obtaining local meteorological features; fusing the scene-aware spatial features and local meteorological features to generate a fused feature sequence; capturing spatiotemporal evolution patterns through the time modeling module; and outputting the predicted power value of the target photovoltaic power plant. This invention improves the accuracy of photovoltaic power prediction.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power prediction technology, and more specifically, to a method, apparatus, and device for spatiotemporal prediction of distributed photovoltaic power based on scene-conditional graph attention. Background Technology

[0002] Against the backdrop of the global push for carbon neutrality, the penetration rate of distributed photovoltaic (PV) power in the power system has increased significantly. However, due to the significant influence of factors such as irradiance, cloud cover, and local meteorological conditions, distributed PV power exhibits strong fluctuations, abrupt changes, and disorder. In addition, there is a dynamic spatial correlation between geographically proximate distributed PV power plants, which poses a severe challenge to the safe and stable operation and economic dispatch of the power grid.

[0003] Currently, there is considerable research on photovoltaic power prediction, mainly including modeling methods based on single-site time series and modeling methods based on graph structures. Among them, the modeling method based on single-site time series ignores the spatial correlation between power stations; while the modeling method using static graph structures, although introducing spatial information, is difficult to adapt to uncertainties across scenarios, resulting in the model's inability to effectively take into account the spatiotemporal dependence and evolution laws under complex scenarios, and thus suffers from low prediction accuracy. Summary of the Invention

[0004] To overcome the shortcomings of low prediction accuracy in existing photovoltaic power prediction technologies, this invention proposes the following technical solution: In the first aspect, this invention proposes a spatiotemporal prediction method for distributed photovoltaic power based on scene-conditional graph attention, comprising: Obtain historical power sequences of multiple distributed photovoltaic power stations and corresponding meteorological time series of the region; An uncertainty feature system is constructed based on the historical power sequence; Based on the aforementioned uncertainty feature system, historical data are clustered to obtain similar uncertainty scenarios that represent different uncertainty patterns; Construct a spatiotemporal graph network that includes a graph attention module, a convolutional feature extraction module, and a temporal modeling module; The similar uncertainty scenarios are injected into the graph attention module as conditional information. The attention weight calculation of the graph attention module is adjusted using the similar uncertainty scenarios. Based on the adjusted attention weights, spatial features are extracted from multiple distributed photovoltaic power stations to obtain the spatial features of scene perception. The meteorological time series is input into the convolutional feature extraction module for feature extraction to obtain local meteorological features; The spatial features of the scene perception and the local meteorological features are fused to generate a fused feature sequence; The fused feature sequence is input into the time modeling module, which captures the spatiotemporal evolution pattern and outputs the power prediction value of the target photovoltaic power station.

[0005] As a preferred technical solution, the uncertainty feature system includes a discrete intensity index and a discrete rate index constructed from the volatility dimension; The discrete intensity index is calculated according to the following formula:

[0006] in, Sampling points The corresponding average value, The standard deviation of the sampled data; The dispersion index is calculated using the following formula:

[0007] in, Sampling points The power value, The sampling time interval, Sampling points The power value.

[0008] As a preferred technical solution, the uncertainty feature system further includes constructing a mutation intensity index and a mutation frequency index from the mutation dimension; The absolute power difference of the mutation intensity index is calculated using the following formula:

[0009] The mutation frequency index is calculated using the following formula:

[0010] in, Indicates the number of valid mutation events. This indicates the total number of rating statistics.

[0011] As a preferred technical solution, the uncertainty feature system further includes constructing a disorder intensity index and a disorder frequency index from the disorder dimension; The disorder intensity index is calculated using the following formula:

[0012] in, Indicates power falling into Possibilities within the range The number of power intervals to be divided; The disorder frequency index is calculated using the following formula:

[0013] in, express The proportion of total power.

[0014] As a preferred technical solution, the clustering of historical data based on the uncertainty feature system includes: Construct a feature attention module, where the dimension for any given day is... D Uncertainty eigenvectors The interaction between features is learned through nonlinear transformation, and a normalized attention weight vector is generated, which is calculated as follows:

[0015] in, and This is the weight matrix. and It is the bias vector; The similar uncertainty scene embedding is obtained by weighted summation of the original feature vector using the attention weight vector, and its calculation formula is as follows:

[0016] in, For the first Attention weights corresponding to each uncertain feature; For the first The values ​​of the original uncertainty features are used; clustering algorithms are used to cluster the embeddings of the similar uncertainty scenarios to generate the similar uncertainty scenarios.

[0017] As a preferred technical solution, the graph attention module is a graph attention mechanism, which represents multiple photovoltaic power stations as a graph, where each power station is a node, and the initial feature of the node is its historical power sequence; the geographical distance between power stations is calculated based on the geographical coordinates of each distributed photovoltaic power station, and the closest node is selected for each power station based on the geographical distance. Each power station is used as a neighbor node to construct a neighbor set. The step of injecting the similar uncertainty scenario as conditional information into the graph attention module includes: For the central node and its neighboring nodes in the graph structure, the attention coefficient is calculated through a linear transformation incorporating scene information. The formula is as follows:

[0018] in, Attention coefficient and These are the characteristics of the central node and its neighboring nodes, respectively. The scene vector is obtained by embedding the similar uncertainty scenarios. , and For learnable parameters, This indicates vector concatenation; The attention coefficients of all neighbors of a node are normalized to obtain the final attention weight, which is calculated as follows:

[0019] in, The set of neighbors of the central node; The transformed features of neighboring nodes are weighted and summed according to the attention weights, and then nonlinearly activated to obtain the spatial features for scene perception. The calculation formula is as follows:

[0020] In the formula, This represents the next layer of features after transforming neighboring nodes.

[0021] As a preferred technical solution, the feature fusion of the spatial features of the scene perception and the local meteorological features includes: A one-dimensional convolutional neural network is used as the convolutional feature extraction module to process the meteorological time series. By applying a set of convolutional kernels, the local meteorological features are extracted, and their calculation formula is as follows:

[0022] in, Indicates the first k Each convolutional kernel at time step t The output characteristics, Indicates at time step Meteorological data input, Indicates the first k Each convolutional kernel is located at... The weight, Indicates the first k The bias term of each convolution kernel, For the first Each convolutional kernel length, K This represents the total number of convolutional kernels. Indicates local meteorological characteristics. This represents a vector concatenation operation. Use the Sigmoid activation function; The spatial features of scene perception output by the graph attention module are concatenated with the local meteorological features as vectors, and then linear mapping and normalization are applied to obtain a fused feature sequence, the calculation formula of which is:

[0023] in, This is a feature of intermediate splicing. This is the final fused feature sequence. The output features of the graph attention module, This is a local meteorological feature. This represents the normalization after a linear mapping. The weights and biases are for the linear mapping.

[0024] As a preferred technical solution, the step of inputting the fused feature sequence into the time modeling module includes: A Long Short-Term Memory (LSTM) network is used as the time modeling module to perform gated recursive computation on the fused feature sequence. The gating and output satisfy the following:

[0025]

[0026] in, , and These are the forget gate, input gate, and output gate, respectively. Indicates the state of candidate cells. Indicates the current cell state. Indicates the current hidden state. This indicates the hidden state at the previous moment; These are the corresponding weight matrices; The hidden state of the previous moment The corresponding cyclic weight matrix; These are the corresponding bias vectors; The hyperbolic tangent activation function is used. For Hadamah accumulation; Based on the hidden state, the predicted power value of the target photovoltaic power station is output, and its calculation formula is:

[0027] in, To predict power, and These are the corresponding weights and biases, respectively.

[0028] Secondly, the present invention also proposes a distributed photovoltaic power spatiotemporal prediction device based on scene-conditional graph attention, applied to the distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention as described in any of the embodiments of the first aspect, comprising: The acquisition module is used to acquire historical power sequences of multiple distributed photovoltaic power stations and corresponding meteorological time sequences of the region; The first construction module is used to construct an uncertainty feature system based on the historical power sequence; The clustering module is used to cluster historical data according to the uncertainty feature system to obtain similar uncertainty scenarios that represent different uncertainty patterns; The second building module is used to construct a spatiotemporal graph network that includes a graph attention module, a convolutional feature extraction module, and a temporal modeling module; The first extraction module is used to inject the similar uncertainty scenario as conditional information into the graph attention module, adjust the attention weight calculation of the graph attention module using the similar uncertainty scenario, and extract spatial features of multiple distributed photovoltaic power stations based on the adjusted attention weight to obtain the spatial features of scene perception. The second extraction module is used to input the meteorological time series into the convolutional feature extraction module for feature extraction to obtain local meteorological features; The fusion module is used to fuse the spatial features of the scene perception with the local meteorological features to generate a fused feature sequence. The output module is used to input the fused feature sequence into the time modeling module, capture the spatiotemporal evolution law through the time modeling module, and output the power prediction value of the target photovoltaic power station.

[0029] Thirdly, the present invention also proposes an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the operations performed by the distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention as described in any of the embodiments of the first aspect.

[0030] The beneficial effects of the present invention include at least the following: This invention constructs an uncertainty feature system based on historical power sequences and clusters historical data to uncover similar uncertainty scenarios representing different uncertainty patterns. These similar uncertainty scenarios are then injected as conditional information into the graph attention module, which adjusts the calculation of attention weights. This allows the model to adaptively adjust the focus on extracting spatial features between distributed photovoltaic power plants based on the current uncertainty scenario, thereby obtaining scene-aware spatial features. Combined with the extraction of local meteorological features and the capture of spatiotemporal evolution patterns by the time modeling module, this method effectively overcomes the limitations of traditional static graph structure models in adapting to the strong fluctuations and uncertainties of distributed photovoltaic power. It achieves dynamic and accurate modeling of spatial dependencies under different uncertainty patterns, thus significantly improving the prediction accuracy of distributed photovoltaic power. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the spatiotemporal prediction method for distributed photovoltaic power based on scene-conditional graph attention provided in an embodiment of the present invention.

[0032] Figure 2 This is a comparison chart of spatiotemporal prediction results of distributed photovoltaic power based on scene-conditional graph attention provided in an embodiment of the present invention.

[0033] Figure 3 This is an architectural diagram of a distributed photovoltaic power spatiotemporal prediction device based on scene-conditional graph attention, provided in an embodiment of the present invention.

[0034] Figure 4 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0035] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred technical solutions. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred technical solutions are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0036] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0037] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0038] Example 1 This embodiment proposes a distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention, such as... Figure 1 As shown, Figure 1 This is a flowchart illustrating a distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention provided in this embodiment. The method includes the following steps: S1: Obtain historical power sequences of multiple distributed photovoltaic power stations and corresponding meteorological time series of the region; S2: Construct an uncertainty feature system based on the historical power sequence; S3: Cluster historical data according to the uncertainty feature system to obtain similar uncertainty scenarios that represent different uncertainty patterns; S4: Construct a spatiotemporal graph network that includes a graph attention module, a convolutional feature extraction module, and a temporal modeling module; S5: The similar uncertainty scenario is injected into the graph attention module as conditional information. The attention weight calculation of the graph attention module is adjusted using the similar uncertainty scenario. Based on the adjusted attention weight, spatial features are extracted from multiple distributed photovoltaic power stations to obtain the spatial features of scene perception. S6: Input the meteorological time series into the convolutional feature extraction module for feature extraction to obtain local meteorological features; S7: Perform feature fusion on the spatial features of the scene perception and the local meteorological features to generate a fused feature sequence; S8: Input the fused feature sequence into the time modeling module, capture the spatiotemporal evolution law through the time modeling module, and output the power prediction value of the target photovoltaic power station.

[0039] Understandably, by constructing an uncertainty feature system and clustering historical data, similar uncertainty scenarios representing different uncertainty patterns can be obtained. This overcomes the limitations of traditional static graph structures in adapting to the strong fluctuations in photovoltaic power, accurately adapting to photovoltaic scenarios under different meteorological conditions, and solving the problem of poor adaptability of prediction models in complex scenarios. At the same time, by injecting similar uncertainty scenarios as conditional information into the graph attention module to adjust the attention weight, on the one hand, adaptive perception of the dynamic spatial dependencies between power stations is achieved while ensuring spatial topological connectivity, effectively avoiding spatial feature extraction distortion caused by ignoring scenario differences. On the other hand, by combining the synergistic effect of convolutional feature extraction and temporal modeling modules, the efficiency and accuracy of capturing spatiotemporal evolution patterns are greatly improved. The scene-aware spatiotemporal prediction mechanism has natural robustness and can significantly improve the power prediction accuracy of distributed photovoltaics under strong uncertainty environments.

[0040] Example 2 This embodiment improves upon the spatiotemporal prediction method for distributed photovoltaic power based on scene-conditional graph attention proposed in Embodiment 1.

[0041] In this embodiment, the uncertainty feature system includes a discrete intensity index and a discrete rate index constructed from the volatility dimension; The discrete intensity index is calculated according to the following formula:

[0042] in, Sampling points The corresponding average value, The standard deviation of the sampled data; The dispersion index is calculated using the following formula:

[0043] in, Sampling points The power value, The sampling time interval, Sampling points The power value.

[0044] In this embodiment, the uncertainty feature system further includes constructing a mutation intensity index and a mutation frequency index from the dimension of mutation. The absolute power difference of the mutation intensity index is calculated using the following formula:

[0045] The mutation frequency index is calculated using the following formula:

[0046] in, Indicates the number of valid mutation events. This indicates the total number of rating statistics.

[0047] In this embodiment, the uncertainty feature system further includes constructing a disorder intensity index and a disorder frequency index from the disorder dimension; The disorder intensity index is calculated using the following formula:

[0048] in, Indicates power falling into Possibilities within the range The number of power intervals to be divided; The disorder frequency index is calculated using the following formula:

[0049] in, express The proportion of total power.

[0050] In this embodiment, the clustering of historical data based on the uncertainty feature system includes: Construct a feature attention module, where the dimension for any given day is... D Uncertainty eigenvectors The interaction between features is learned through nonlinear transformation, and a normalized attention weight vector is generated, which is calculated as follows:

[0051] in, and This is the weight matrix. and It is the bias vector; The similar uncertainty scene embedding is obtained by weighted summation of the original feature vector using the attention weight vector, and its calculation formula is as follows:

[0052] in, For the first Attention weights corresponding to each uncertain feature; For the first The value of the original uncertain feature; Clustering algorithms are used to cluster the embeddings of the similar uncertainty scenarios to generate the similar uncertainty scenarios.

[0053] In this embodiment, the graph attention module is a graph attention mechanism, which represents multiple photovoltaic power stations as a graph, where each power station is a node, and the initial feature of the node is its historical power sequence. The geographical distance between power stations is calculated based on the geographical coordinates of each distributed photovoltaic power station, and the closest node is selected for each power station based on the geographical distance. Each power station is used as a neighbor node to construct a neighbor set. The step of injecting the similar uncertainty scenario as conditional information into the graph attention module includes: For the central node and its neighboring nodes in the graph structure, the attention coefficient is calculated through a linear transformation incorporating scene information. The formula is as follows:

[0054] in, Attention coefficient and These are the characteristics of the central node and its neighboring nodes, respectively. The scene vector is obtained by embedding the similar uncertainty scenarios. , and For learnable parameters, This indicates vector concatenation; The attention coefficients of all neighbors of a node are normalized to obtain the final attention weight, which is calculated as follows:

[0055] in, The set of neighbors of the central node; The transformed features of neighboring nodes are weighted and summed according to the attention weights, and then nonlinearly activated to obtain the spatial features for scene perception. The calculation formula is as follows:

[0056] In the formula, This represents the next layer of features after transforming neighboring nodes.

[0057] In this embodiment, the feature fusion of the spatial features of the scene perception and the local meteorological features includes: A one-dimensional convolutional neural network is used as the convolutional feature extraction module to process the meteorological time series. Apply convolution kernel group The local meteorological features are extracted, and their calculation formula is as follows:

[0058] in, Indicates the first k Each convolutional kernel at time step t The output characteristics, Indicates at time step Meteorological data input, Indicates the first k Each convolutional kernel is located at... The weight, Indicates the first k The bias term of each convolution kernel, For the first Each convolutional kernel length, K This represents the total number of convolutional kernels. Indicates local meteorological characteristics. This represents a vector concatenation operation. Use the Sigmoid activation function; The spatial features of scene perception output by the graph attention module are concatenated with the local meteorological features as vectors, and then linear mapping and normalization are applied to obtain a fused feature sequence, the calculation formula of which is:

[0059] in, This is a feature of intermediate splicing. This is the final fused feature sequence. The output features of the graph attention module, This is a local meteorological feature. This represents the normalization after a linear mapping. The weights and biases are for the linear mapping.

[0060] In this embodiment, inputting the fused feature sequence into the time modeling module includes: A Long Short-Term Memory (LSTM) network is used as the time modeling module to perform gated recursive computation on the fused feature sequence. The gating and output satisfy the following:

[0061]

[0062] in, , and These are the forget gate, input gate, and output gate, respectively. Indicates the state of candidate cells. Indicates the current cell state. Indicates the current hidden state. This indicates the hidden state at the previous moment; These are the corresponding weight matrices; The hidden state of the previous moment The corresponding cyclic weight matrix; These are the corresponding bias vectors; The hyperbolic tangent activation function is used. For Hadamah accumulation; Based on the hidden state, the predicted power value of the target photovoltaic power station is output, and its calculation formula is:

[0063] in, To predict power, and These are the corresponding weights and biases, respectively.

[0064] like Figure 2 As shown, Figure 2 This is a comparison chart of the spatiotemporal prediction results of distributed photovoltaic power based on scene-conditional graph attention provided in an embodiment of the present invention. In the chart, the horizontal axis represents the time sampling point, and the vertical axis represents the active power (kW).

[0065] As shown in the figure, in scenarios with stable photovoltaic power changes, the predicted curve of the model in this invention highly coincides with the actual observed curve, demonstrating a good fit. However, in scenarios with drastic fluctuations and frequent abrupt changes in photovoltaic power, thanks to the conditional injection of similar uncertainties and the spatial feature adjustment of the graph attention mechanism, the model can still accurately capture the instantaneous trends and peak-trough details of power changes without significant lag or bias. In summary, regardless of whether the scenario is stable or highly volatile, the method of this invention maintains high prediction accuracy, verifying its effectiveness in handling complex spatiotemporal dependencies and coping with strong uncertainties.

[0066] Example 3 like Figure 3 As shown, this embodiment proposes a distributed photovoltaic power spatiotemporal prediction device based on scene-conditional graph attention, which is applied to the distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention as described in the above embodiment. It includes: an acquisition module 100, a first construction module 200, a clustering module 300, a second construction module 400, a first extraction module 500, a second extraction module 600, a fusion module 700, and an output module 800.

[0067] The acquisition module 100 is used to acquire historical power sequences of multiple distributed photovoltaic power stations and meteorological time sequences of corresponding regions; the first construction module 200 is used to construct an uncertainty feature system based on the historical power sequences; the clustering module 300 is used to cluster historical data according to the uncertainty feature system to obtain similar uncertainty scenarios representing different uncertainty patterns; the second construction module 400 is used to construct a spatiotemporal graph network including a graph attention module, a convolutional feature extraction module, and a time modeling module; and the first extraction module 500 is used to inject the similar uncertainty scenarios as conditional information into the graph attention module, utilizing the similar uncertainty... The scene adjustment module calculates the attention weights of the graph attention module and extracts spatial features from multiple distributed photovoltaic power stations based on the adjusted attention weights to obtain scene-aware spatial features; the second extraction module 600 is used to input the meteorological time series into the convolutional feature extraction module for feature extraction to obtain local meteorological features; the fusion module 700 is used to fuse the scene-aware spatial features with the local meteorological features to generate a fused feature sequence; the output module 800 is used to input the fused feature sequence into the time modeling module, capture the spatiotemporal evolution law through the time modeling module, and output the power prediction value of the target photovoltaic power station.

[0068] It should be noted that the foregoing explanation of the embodiment of the distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention also applies to the distributed photovoltaic power spatiotemporal prediction device based on scene-conditional graph attention in this embodiment, and will not be repeated here.

[0069] Example 4 Figure 4 This is a schematic diagram of the structure of the electronic device 900 provided in this embodiment. The electronic device 900 includes: a memory 901, a processor 902, and a computer program stored in the memory 901 and executable on the processor 902.

[0070] When the processor 902 executes the program, it implements the distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention provided in the above embodiments.

[0071] Furthermore, the electronic device 900 also includes a communication interface 903 for communication between the memory 901 and the processor 902.

[0072] The memory 901 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage device.

[0073] If the memory 901, processor 902, and communication interface 903 are implemented independently, then the communication interface 903, memory 901, and processor 902 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0074] Optionally, in a specific implementation, if the memory 901, processor 902, and communication interface 903 are integrated on a single chip, then the memory 901, processor 902, and communication interface 903 can communicate with each other through an internal interface.

[0075] The processor 902 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present invention.

[0076] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for spatiotemporal prediction of distributed photovoltaic power based on scene-conditional graph attention.

[0077] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0078] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0079] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0080] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0081] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0082] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for spatiotemporal prediction of distributed photovoltaic power based on scene-conditional graph attention, characterized in that, include: Obtain historical power sequences of multiple distributed photovoltaic power stations and corresponding meteorological time series of the region; An uncertainty feature system is constructed based on the historical power sequence; Based on the aforementioned uncertainty feature system, historical data are clustered to obtain similar uncertainty scenarios that represent different uncertainty patterns; Construct a spatiotemporal graph network that includes a graph attention module, a convolutional feature extraction module, and a temporal modeling module; The similar uncertainty scenarios are injected into the graph attention module as conditional information. The attention weight calculation of the graph attention module is adjusted using the similar uncertainty scenarios. Based on the adjusted attention weights, spatial features are extracted from multiple distributed photovoltaic power stations to obtain the spatial features of scene perception. The meteorological time series is input into the convolutional feature extraction module for feature extraction to obtain local meteorological features; The spatial features of the scene perception and the local meteorological features are fused to generate a fused feature sequence; The fused feature sequence is input into the time modeling module, which captures the spatiotemporal evolution pattern and outputs the power prediction value of the target photovoltaic power station.

2. The distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention according to claim 1, characterized in that, The uncertainty feature system includes a discrete intensity index and a discrete rate index constructed from the volatility dimension; The discrete intensity index is calculated according to the following formula: in, Sampling points The corresponding average value, The standard deviation of the sampled data; The dispersion index is calculated using the following formula: in, Sampling points The power value, The sampling time interval, Sampling points The power value.

3. The distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention according to claim 2, characterized in that, The uncertainty feature system also includes constructing a mutation intensity index and a mutation frequency index from the dimension of mutation. The absolute power difference of the mutation intensity index is calculated using the following formula: The mutation frequency index is calculated using the following formula: in, Indicates the number of valid mutation events. This indicates the total number of rating statistics.

4. The distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention according to claim 3, characterized in that, The uncertainty feature system also includes constructing a disorder intensity index and a disorder frequency index from the disorder dimension; The disorder intensity index is calculated using the following formula: in, Indicates power falling into Possibilities within the range The number of power intervals to be divided; The disorder frequency index is calculated using the following formula: in, express The proportion of total power.

5. The distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention according to claim 1, characterized in that, The historical data is clustered based on the aforementioned uncertainty feature system. include: Construct a feature attention module, where the dimension of any day is... D Uncertainty eigenvectors The interaction between features is learned through nonlinear transformation, and a normalized attention weight vector is generated, which is calculated as follows: in, and This is the weight matrix. and It is the bias vector; The similar uncertainty scene embedding is obtained by weighted summation of the original feature vector using the attention weight vector, and its calculation formula is as follows: in, For the first Attention weights corresponding to each uncertain feature; For the first The value of the original uncertain feature; Clustering algorithms are used to cluster the embeddings of the similar uncertainty scenarios to generate the similar uncertainty scenarios.

6. The distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention according to claim 1, characterized in that, The graph attention module employs a graph attention mechanism, representing multiple photovoltaic power plants as a graph, where each power plant is a node, and the initial feature of each node is its historical power sequence. It calculates the geographical distance between power plants based on their geographical coordinates and selects the closest node for each power plant based on this geographical distance. Each power station is used as a neighbor node to construct a neighbor set. ; The step of injecting the similar uncertainty scenarios as conditional information into the graph attention module includes: For the central node and its neighboring nodes in the graph structure, the attention coefficient is calculated through a linear transformation that incorporates scene information. The formula is as follows: in, Attention coefficient and These are the characteristics of the central node and its neighboring nodes, respectively. The scene vector is obtained by embedding the similar uncertainty scenarios. , and For learnable parameters, This indicates vector concatenation; The attention coefficients of all neighbors of a node are normalized to obtain the final attention weight, which is calculated as follows: in, The set of neighbors of the central node; The transformed features of neighboring nodes are weighted and summed according to the attention weights, and then nonlinearly activated to obtain the spatial features for scene perception. The calculation formula is as follows: In the formula, This represents the next layer of features after transforming neighboring nodes.

7. The distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention according to claim 1, characterized in that, The feature fusion of the spatial features perceived in the scene and the local meteorological features includes: A one-dimensional convolutional neural network is used as the convolutional feature extraction module to process the meteorological time series. Apply convolution kernel group The local meteorological features are extracted, and their calculation formula is as follows: in, Indicates the first k Each convolutional kernel at time step t The output characteristics, Indicates at time step Meteorological data input, Indicates the first k Each convolutional kernel is located at... The weight, Indicates the first k The bias term of each convolution kernel, For the first Each convolutional kernel length, K This represents the total number of convolutional kernels. Indicates local meteorological characteristics. This represents a vector concatenation operation. Use the Sigmoid activation function; The spatial features of scene perception output by the graph attention module are concatenated with the local meteorological features as vectors, and then linear mapping and normalization are applied to obtain a fused feature sequence, the calculation formula of which is: in, This is a feature of intermediate splicing. This is the final fused feature sequence. The output features of the graph attention module, This is a local meteorological feature. This represents the normalization after a linear mapping. The weights and biases are for the linear mapping.

8. The distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention according to claim 1, characterized in that, The step of inputting the fused feature sequence into the time modeling module includes: A Long Short-Term Memory (LSTM) network is used as the time modeling module to perform gated recursive computation on the fused feature sequence. The gating and output satisfy the following: in, , and These are the forget gate, input gate, and output gate, respectively. Indicates the state of candidate cells. Indicates the current cell state. Indicates the current hidden state. This indicates the hidden state at the previous moment; These are the corresponding weight matrices; The hidden state of the previous moment The corresponding cyclic weight matrix; These are the corresponding bias vectors; The hyperbolic tangent activation function is used. For Hadamah accumulation; Based on the hidden state, the predicted power value of the target photovoltaic power station is output, and its calculation formula is: in, To predict power, and These are the corresponding weights and biases, respectively.

9. A distributed photovoltaic power spatiotemporal prediction device based on scene-conditional graph attention, characterized in that, include: The acquisition module is used to acquire historical power sequences of multiple distributed photovoltaic power stations and corresponding meteorological time sequences of the region; The first construction module is used to construct an uncertainty feature system based on the historical power sequence; The clustering module is used to cluster historical data according to the uncertainty feature system to obtain similar uncertainty scenarios that represent different uncertainty patterns; The second building module is used to construct a spatiotemporal graph network that includes a graph attention module, a convolutional feature extraction module, and a temporal modeling module; The first extraction module is used to inject the similar uncertainty scenario as conditional information into the graph attention module, adjust the attention weight calculation of the graph attention module using the similar uncertainty scenario, and extract spatial features of multiple distributed photovoltaic power stations based on the adjusted attention weight to obtain the spatial features of scene perception. The second extraction module is used to input the meteorological time series into the convolutional feature extraction module for feature extraction to obtain local meteorological features; The fusion module is used to fuse the spatial features of the scene perception with the local meteorological features to generate a fused feature sequence. The output module is used to input the fused feature sequence into the time modeling module, capture the spatiotemporal evolution law through the time modeling module, and output the power prediction value of the target photovoltaic power station.

10. An electronic device, characterized in that, The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the operations performed by the distributed photovoltaic power spatiotemporal prediction method based on scene-conditional graph attention as described in any one of claims 1 to 8.

Images