Photovoltaic power station maximum power generation calculation method fusing terrain and meteorological characteristics
By constructing a meteorological-photovoltaic grid node system and a hierarchical dynamic graph model, the problem of inaccurate assessment of the maximum power generation capacity of photovoltaic bases in existing technologies has been solved, achieving high-precision prediction of the maximum power generation capacity and supporting the safe and efficient operation of new power systems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTH CHINA ELECTRIC POWER UNIV
- Filing Date
- 2025-12-31
- Publication Date
- 2026-05-12
AI Technical Summary
Existing photovoltaic power prediction and assessment methods cannot accurately depict the meteorological-power mapping relationship within ultra-large-scale photovoltaic bases, resulting in inaccurate assessment of maximum power generation and affecting grid dispatch and renewable energy consumption efficiency.
By constructing a meteorological-photovoltaic grid node system, and adopting a two-stage coding architecture of CNN terrain-aware dilated convolutional network and structure-aware Transformer, combined with a hierarchical dynamic graph model, we can realize the propagation and modulation of meteorological disturbances under complex terrain and make high-precision prediction of maximum achievable power.
It has achieved high-precision, panoramic perception of photovoltaic power plants in complex terrain, and provided high-resolution prediction of maximum generating power, laying a data foundation for the safe, efficient, and economical operation of new power systems.
Smart Images
Figure CN122020604A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of photovoltaic power plant power technology, specifically involving a method for calculating the maximum power output of a photovoltaic power plant that integrates terrain and meteorological characteristics. Background Technology
[0002] Currently, my country is accelerating the construction of a new power system with new energy sources as the mainstay, and the construction of large-scale wind and solar power bases is progressing at a rapid pace. Driven by the "dual-carbon" strategic goal, several large-scale photovoltaic bases with a capacity of tens of millions of kilowatts have been planned and built in western and northern regions such as Qinghai, Ningxia, Gansu, and Inner Mongolia. These bases are vast, covering an area of hundreds or even thousands of square kilometers, with complex and diverse internal topography, encompassing various geographical units such as plateaus, mountains, hills, and deserts.
[0003] However, it is precisely this combination of massive scale and complex terrain that presents unprecedented challenges to the refined operation and management of photovoltaic power plants. Due to differences in topographical elements such as altitude, slope, and aspect, as well as the influence of meteorological processes such as local circulation and cloud movement, different areas within the base exhibit significant spatial heterogeneity in meteorological conditions. For example, at the same time, one side of the base may be bathed in sunshine, while the other side is covered by clouds; the ridge may receive ample sunlight, while the valley may be shrouded in shadow or fog. This micro-meteorological pattern, where "the weather varies every ten miles," directly leads to significant differences in the actual maximum generating capacity (i.e., the theoretical upper limit of power generation) of photovoltaic power plants in different sub-regions.
[0004] Existing photovoltaic (PV) power prediction and assessment methods are mostly based on point-to-point modeling, directly using data from a single or a few meteorological observation points to estimate the power of the entire power plant or its vicinity. These methods ignore the dynamic propagation patterns of meteorological fields in space and their coupling with complex terrain, failing to accurately depict the fine meteorological-power mapping relationship within the base. As a result, the assessment of the maximum power output of large-scale PV bases is often too broad, either overestimating the power generation potential of areas affected by shading or cloud cover, or underestimating the output capacity of areas with abundant sunlight. This inaccurate assessment not only affects the grid dispatching department's precise grasp of available power resources and reduces the level of new energy consumption, but also restricts the efficiency of electricity market transactions and may even threaten the safe and stable operation of the large power grid.
[0005] Therefore, there is an urgent need to develop a novel computational method that can deeply integrate the spatiotemporal correlation characteristics of topographic features and meteorological elements. This method must be able to systematically characterize how meteorological disturbances propagate and evolve in complex terrain, and ultimately modulate the maximum power generation capacity of each photovoltaic unit. Only in this way can we provide high-precision, high-resolution panoramic perception of the maximum generateable power for ultra-large-scale photovoltaic bases, laying a solid data and model foundation for the safe, efficient, and economical operation of new power systems. This patent is proposed against this background, aiming to solve the key bottleneck problems faced by existing technologies in complex scenarios. Summary of the Invention
[0006] The technical problem to be solved by this invention is to provide a method for calculating the maximum power output of a photovoltaic power station that integrates terrain and meteorological features. Through algorithm improvement, it is possible to systematically characterize how meteorological disturbances propagate and evolve under complex terrain, and ultimately modulate the maximum power generation capacity of each photovoltaic unit. This provides high-precision, high-resolution panoramic perception of the maximum power output for ultra-large-scale photovoltaic bases, and lays a solid data and model foundation for the safe, efficient and economical operation of new power systems.
[0007] The technical solution adopted in this invention is: a method for calculating the maximum power generation of a photovoltaic power station by integrating terrain and meteorological features. The method includes the following steps: Step S10, collecting meteorological forecast information and static terrain information of the photovoltaic power station array as grid nodes, as well as the power generation, latitude and longitude location information, and installed capacity of the photovoltaic power station array in the photovoltaic base, constructing a meteorological-photovoltaic grid node system, and mapping the discrete photovoltaic power station array and meteorological forecast information to a unified geographic spatial grid; Step S20, a spatial correlation field based on a physical element mechanism, whose physical elements are jointly determined by geographical distance, historical radiation correlation, prevailing wind direction, and azimuth angle relationship, used to quantify the comprehensive correlation strength between any two grid nodes; Step... Step S30 employs a CNN terrain-aware dilated convolutional network model and a structure-aware Transformer dual-stage encoding architecture to output topological embedding vectors from the meteorological-photovoltaic grid node system; Step S40 extracts multi-scale spatiotemporal features from the topological embedding vectors of the grid nodes and meteorological forecast information; Step S50 divides the photovoltaic power station array into multi-level sub-regions and constructs a multi-grid system; Step S60 constructs a hierarchical dynamic graph structure; Step S70 deploys a spatiotemporal graph neural network on the hierarchical dynamic graph structure, using multi-source meteorological forecast data for future periods as input, and performs forward propagation through the hierarchical dynamic graph model, ultimately outputting the maximum power generation prediction value of the photovoltaic power station array from the grid nodes.
[0008] The beneficial effects of this invention are as follows: This invention mainly breaks through the limitations of traditional point-to-point modeling and constructs a new modeling paradigm that can systematically characterize the coupling relationship between the spatial propagation law of meteorological fields, the modulation effect of topography, and the response mechanism of photovoltaic power generation. First, by identifying key meteorological elements and aligning them with photovoltaic output in the spatiotemporal dimension, photovoltaic sites and multi-source meteorological data are networked and modeled to construct a "meteorology + photovoltaic" grid node system. Spatial topological features are encoded for each node to characterize the influence of topography on the relationship between meteorology and power generation. Second, considering the spatiotemporal correlation characteristics between meteorological elements, multi-scale spatiotemporal features are extracted, multi-level sub-regions are divided, and a multi-grid system is constructed to characterize the coupling relationship of meteorological processes at different spatial scales, providing structured support for subsequent modeling. Finally, based on the hierarchical dynamic graph modeling method, a meteorological-power mapping model is constructed to achieve accurate modeling of the complex spatial correlation between meteorology and power output within a large photovoltaic base. This method employs a four-pronged approach—multi-source meteorological fusion, terrain shading correction, multi-mechanism spatial correlation modeling, and CNN+Transformer structured encoding—to achieve a deep, structured representation of the physical process of how meteorological phenomena propagate spatially and are modulated by terrain, thus affecting photovoltaic power output under complex terrain conditions. The entire process is based entirely on regular grid computation, requiring no explicit graph structures or graph neural networks. It balances model expressiveness, physical interpretability, training stability, and engineering deployment efficiency, providing a new generation of data-driven modeling paradigm for refined power sensing and intelligent scheduling in high-proportion renewable energy systems. Attached Figure Description
[0009] Figure 1 This is a flowchart of the present invention. Detailed Implementation
[0010] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.
[0011] See appendix Figure 1 This invention provides a method for calculating the maximum power output of a photovoltaic power station by integrating terrain and meteorological features, comprising the following steps: Step S10: Collect meteorological forecast information and static terrain information of photovoltaic power station array as grid nodes, as well as power generation, latitude and longitude location information and installed capacity of photovoltaic power station array in photovoltaic base, and construct meteorological-photovoltaic grid node system to map discrete photovoltaic power station array and meteorological forecast information to a unified geospatial grid.
[0012] The meteorological forecast information in step S10 includes total surface radiation flux, cloud cover, temperature, and wind speed; Static terrain information includes elevation, slope, and aspect; The formula for mapping discrete photovoltaic sites and weather forecast information to a unified geospatial grid is S. ij (t): In the above formula, Meteorological characteristics, Let G be the terrain state feature, t be the time, and G be the time value. i E represents the total surface radiation flux. i Cloud cover, value range 0–100%, T i For temperature, Wind speed; For altitude, α i and β i These are slope and aspect, respectively.
[0013] Step S20: The spatial correlation field is based on the physical element mechanism fusion. Its physical elements are determined by geographical distance, historical radiation correlation, dominant wind direction and azimuth relationship, and are used to quantify the comprehensive correlation strength between any two grid nodes.
[0014] Step S201: Collect the geographical distance d between each grid node. nm Pearson correlation coefficient ρ between the historical total radiation sequences of grid nodes nm The prevailing wind direction angle ψ of the meteorological system near the ground at the grid node n The azimuth angle θ between grid nodes nm The typical propagation scale L of cloud clusters or meteorological disturbances; Step S202: Calculate the meteorological correlation strength between each grid node. In the formula, To adjust the weights, λ1, λ2, and λ3 are adjustable weight coefficients.
[0015] Step S30: A two-stage encoding architecture, consisting of a CNN terrain-aware dilated convolutional network model and a structure-aware Transformer, is adopted to enable the meteorological-photovoltaic grid node system to output a topology embedding vector.
[0016] Step S301: A CNN terrain-aware dilated convolutional network is used to aggregate local features of the input tensor to capture multi-scale terrain-weather coupling patterns. In the formula, *e(l) represents the expansion rate e (l) The convolution operation outputs... It contains rich local contextual information and has initially integrated the topographical shading effect and meteorological co-variation characteristics of the adjacent area; ReLU() is an activation function. It is a parametric operator. It is a bias operator, where S represents the set of static terrain information and weather forecast information obtained in step S10; Step S302: Introduce a structure-aware Transformer two-stage encoding architecture, using the spatial correlation field as a structure bias injection attention mechanism to model long-range dependencies under physical constraints. In the formula, Attention m () represents the attention computation operator, MHSA() represents multi-head attention, head n Let Q represent the nth attention head, K be the query vector, K be the key vector, and V be the value vector. For the original attention score, B∈R N×N The bias matrix is defined as follows: In the formula, B nm The element in the nth row and mth column represents the element in the nth row and mth column, and ò represents the weighting parameter. Step S303: Output a topological embedding vector that fuses local details and global structural context at each mesh node location. In the formula, MLP() represents a fully connected neural network.
[0017] Step S40, multi-scale spatiotemporal feature extraction, includes feature extraction at both the temporal and spatial scales. At the temporal scale, parallel one-dimensional convolutional kernel groups are used to extract short-term fluctuations, daily cycles, and weather system evolution features at the hourly, daily, and weather process levels, respectively. At the spatial scale, three spatial scales are defined: micro (a single photovoltaic power station array), meso (a cluster of photovoltaic power station arrays), and macro (the entire photovoltaic base).
[0018] Step S50: Divide the photovoltaic power station array into multi-level sub-regions and construct a multi-grid system.
[0019] Step S50 includes the following steps: Step S501: Normalize the actual output of the photovoltaic power station array. In the formula, P represents the normalized power output. i t P is the actual photovoltaic output of the i-th photovoltaic array at time t. i,capacity This represents the installed capacity of the photovoltaic array; The latitude and longitude location information of the photovoltaic power station array, as well as the mean, maximum value, and standard deviation of the annual photovoltaic power output sequence, are used as feature inputs (p). i Construct the sample matrix P. feature(p i )=[p i,longitude ,p i,latitude mean(p) i,annual ),max(p i,annual ),std(p i,annual )], In the formula, p i,annual This represents the normalized photovoltaic power series for one year, p i,longitude and p i,latitude These represent the longitude and latitude of the i-th photovoltaic array, respectively. Step S502: Construct a graph structure based on the similarity matrix. Construct an undirected graph based on graph nodes; If there is a connection between two graph nodes, then the weight w of the edge is... ij Not zero; otherwise, zero; The elements w of the adjacency matrix ij The formula is, The graph structure is divided into k subgraphs using the ratio-cut criterion, which represent the sub-regions of the photovoltaic power station array. In the formula, k represents the number of subgraphs. Represents C x The supplement, This represents the and Cutting; Step S503: Transformation and solution of the spectral clustering optimization problem. Λ = Dl, In the formula, the degree matrix Λ is the Laplace matrix, H is a matrix composed of the eigenvectors corresponding to the first k smallest non-zero eigenvalues of the Laplace matrix L, D is the degree matrix, and the diagonal elements d of the degree matrix D are... ii It is the sum of the weights of all connected edges of grid node i; H∈R n×k The matrix makes Tr(H) T LH) is the smallest; Using the K-means clustering algorithm, feature(p) iDivide into the corresponding sub-regions; Initially, k subgraphs are randomly selected as cluster centers, and then the cluster centers are iteratively updated to ensure that all features (p) are clustered. i The data is then grouped into the corresponding cluster center category until convergence. Finally, each actual sample feature(p) is assigned a value. i Based on its position in the low-dimensional embedding space, it is assigned to a certain sub-region, which means that a certain photovoltaic power station array belongs to a certain sub-graph, thus completing the overall photovoltaic power station array division of the base based on spectral clustering. Step S504: Aggregation and hierarchical construction of secondary sub-regions; Based on the division of sub-regions, further evaluate the output synergy, spatial proximity and meteorological consistency among the sub-regions, and perform secondary clustering of the sub-regions to form intermediate regions; This process can be carried out recursively, ultimately constructing a multi-layered grid system of "photovoltaic array → sub-region → intermediate region → overall base".
[0020] Step S60: Construct a hierarchical dynamic graph structure. The hierarchical dynamic graph structure in Step S60 includes graph nodes and edge weights. Graph nodes represent the divided sub-regions, including the photovoltaic power station array as the bottom-level nodes, the primary sub-regions as the middle-level nodes, and the intermediate-level regions or the entire base as the high-level nodes. Edge weights represent the meteorological correlation strength between each grid node. Edge weights are no longer pre-set but are calculated online based on the collaborative change patterns of real-time or recent historical power output sequences. Specifically, the model continuously evaluates the dynamic similarity of the power output curves of two sub-regions through a sliding time window (e.g., using dynamic time-warped distance, mutual information, or time-varying Pearson correlation coefficient), and updates the graph adjacency matrix accordingly, enabling the graph structure to adaptively reflect the actual correlation strength and propagation direction under current weather disturbances.
[0021] Step S70: Deploy a spatiotemporal graph neural network on the hierarchical dynamic graph structure. Use multi-source meteorological forecast data for future periods as input, and perform forward propagation through the hierarchical dynamic graph model. Finally, the grid nodes output the predicted value of the maximum power generation of the photovoltaic power station array.
[0022] Step S70 includes the following steps: Step S701: An LSTM (Long Short-Term Memory) network is introduced to explore the temporal variation pattern of the sub-region output pattern, where the sub-region output pattern is... h t =o t ☉tanh(s t ), In the formula, A t Jt ,o t ,s t ,h t These represent the module's input gate, forget gate, output gate, memory gate, and hidden gate, respectively. x ,Θ h ,b} are the specific model parameters of the gate, σ(·) is the activation function, and X t Represents the module output. X, as spatiotemporal related information t Input t = 1, 2, 3 ⊙ represents the matrix multiplication operator, Θ W These are the corresponding matrix multiplication operators; Step S702: This matrix containing timing information The input is fed into the spatiotemporal graph neural network, and based on the spatiotemporal information of the current time step... The mining yielded spatiotemporal correlation information X with a dimension of 48×1. sub1 ; In the formula, X represents the output of the convolutional layer. Θ is the input to the convolutional layer, and Θ represents the parameters of the convolutional kernel to be trained. Let the node degree matrix be... Let I be the adjacency matrix at time t, and let Θ be the identity matrix. sub1 These are the corresponding subregion multiplication operators; Step S703: By analogy, define the spatiotemporal correlation information extracted from different sub-regions as X. sub1 ,X sub2 ,X sub3 …; X is obtained by modeling the global spatiotemporal correlation and fusing the features of all nodes through a fully connected layer. region The maximum power generation prediction value of the photovoltaic power station that can be obtained is Θ. p It is the corresponding global spatiotemporal correlation multiplication operator; p t:t+15 =X region ☉Θ p .
[0023] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a method for calculating the maximum power output of a photovoltaic power station by integrating terrain and meteorological features.
[0024] The present invention also provides a non-transitory computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a method for calculating the maximum power output of a photovoltaic power station by integrating terrain and meteorological features.
[0025] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for calculating the maximum generating power of a photovoltaic power station by integrating topographic and meteorological features, characterized in that, The method includes the following steps: Step S10: Collect meteorological forecast information and static terrain information of photovoltaic power station array as grid nodes, as well as power generation, latitude and longitude location information and installed capacity of photovoltaic power station array in photovoltaic base, and construct meteorological-photovoltaic grid node system to map discrete photovoltaic power station array and meteorological forecast information to a unified geospatial grid. Step S20: The spatial correlation field is based on the physical element mechanism fusion. Its physical elements are determined by geographical distance, historical radiation correlation, dominant wind direction and azimuth relationship, and are used to quantify the comprehensive correlation strength between any two grid nodes. Step S30: A two-stage encoding architecture of CNN terrain-aware dilated convolutional network model and structure-aware Transformer is adopted to enable the meteorological-photovoltaic grid node system to output a topology embedding vector. Step S40: Extract multi-scale spatiotemporal features from the topological embedding vectors of the grid nodes and meteorological forecast information; Step S50: Divide the photovoltaic power station array into multi-level sub-regions and construct a multi-grid system. Step S60: Construct a hierarchical dynamic graph structure; Step S70: Deploy a spatiotemporal graph neural network on the hierarchical dynamic graph structure. Use multi-source meteorological forecast data for future periods as input, and perform forward propagation through the hierarchical dynamic graph model. Finally, the grid nodes output the predicted value of the maximum power generation of the photovoltaic power station array.
2. The method for calculating the maximum power output of a photovoltaic power station according to claim 1, characterized in that, The meteorological forecast information in step S10 includes total surface radiation flux, cloud cover, temperature and wind speed; Static terrain information includes elevation, slope, and aspect; The formula for mapping discrete photovoltaic sites and weather forecast information to a unified geospatial grid is S. ij (t): In the above formula, Meteorological characteristics, Let G be the terrain state feature, t be the time, and G be the time value. i E represents the total surface radiation flux. i Cloud cover, value range 0–100%, T i For temperature, Wind speed; For altitude, α i and β i These are slope and aspect, respectively.
3. The method for calculating the maximum power output of a photovoltaic power station according to claim 1, characterized in that, Step S20 includes the following steps: Step S201: Collect the geographical distance d between each grid node. nm Pearson correlation coefficient ρ between the historical total radiation sequences of grid nodes nm The prevailing wind direction angle ψ of the meteorological system near the ground at the grid node n The azimuth angle θ between grid nodes nm The typical propagation scale L of cloud clusters or meteorological disturbances; Step S202: Calculate the meteorological correlation strength between each grid node. In the formula, To adjust the weights, λ1, λ2, and λ3 are adjustable weight coefficients.
4. The method for calculating the maximum power output of a photovoltaic power station according to claim 1, characterized in that, Step S30 includes the following steps: Step S301: A CNN terrain-aware dilated convolutional network is used to aggregate local features of the input tensor to capture multi-scale terrain-weather coupling patterns. In the formula, *e(l) represents the expansion rate e (l) The convolution operation outputs... It contains rich local contextual information and has initially integrated the topographical shading effect and meteorological co-variation characteristics of the adjacent area; ReLU() is an activation function. It is a parametric operator. It is a bias operator, where S represents the set of static terrain information and weather forecast information obtained in step S10; Step S302: Introduce a structure-aware Transformer two-stage encoding architecture, using the spatial correlation field as a structure bias injection attention mechanism to model long-range dependencies under physical constraints. In the formula, Attention m () represents the attention computation operator, MHSA() represents multi-head attention, head n Let Q represent the nth attention head, K be the query vector, K be the key vector, and V be the value vector. For the original attention score, B∈R N×N The bias matrix is defined as follows: In the formula, B nm Represents the element in the nth row and mth column. Represents the weighted parameters; Step S303: Output a topological embedding vector that fuses local details and global structural context at each mesh node location. In the formula, MLP() represents a fully connected neural network.
5. The method for calculating the maximum power output of a photovoltaic power station according to claim 1, characterized in that, The multi-scale spatiotemporal feature extraction in step S40 includes feature extraction at both the temporal and spatial scales.
6. The method for calculating the maximum power output of a photovoltaic power station according to claim 1, characterized in that, Step S50 includes the following steps: Step S501: Normalize the actual output of the photovoltaic power station array. In the formula, P represents the normalized power output. i t P is the actual photovoltaic output of the i-th photovoltaic array at time t. i,capacity This represents the installed capacity of the photovoltaic array; The latitude and longitude location information of the photovoltaic power station array, as well as the mean, maximum value, and standard deviation of the annual photovoltaic power output sequence, are used as feature inputs (p). i Construct the sample matrix P. feature(p i )=[p i,longitude ,p i,latitude ,mean(p i,annual ),max(p i,annual ),std(p i,annual )], In the formula, p i,annual This represents the normalized photovoltaic power series for one year, p i,longitude and p i,latitude These represent the longitude and latitude of the i-th photovoltaic array, respectively. Step S502: Construct a graph structure based on the similarity matrix. Construct an undirected graph based on graph nodes; If there is a connection between two graph nodes, then the weight w of the edge is... ij Non-zero; otherwise zero; the element w of the adjacency matrix ij The formula is, The graph structure is divided into k subgraphs using the ratio-cut criterion, which represent the sub-regions of the photovoltaic power station array. In the formula, k represents the number of subgraphs. Represents C x The supplement, This represents the and Cutting; Step S503: Transformation and solution of the spectral clustering optimization problem. Λ = Dl, In the formula, the degree matrix Λ is the Laplace matrix, H is a matrix composed of the eigenvectors corresponding to the first k smallest non-zero eigenvalues of the Laplace matrix L, D is the degree matrix, and the diagonal elements d of the degree matrix D are... ii It is the sum of the weights of all connected edges of grid node i; H∈R n×k The matrix makes Tr(H) T LH) is the smallest; Using the K-means clustering algorithm, feature(p) i ) Divide into the corresponding sub-regions, Step S504: Aggregation and hierarchical construction of secondary sub-regions; Based on the division of sub-regions, further evaluate the output synergy, spatial proximity and meteorological consistency among the sub-regions, and perform secondary clustering of the sub-regions to form intermediate regions; This process can be carried out recursively, ultimately constructing a multi-layered grid system of "photovoltaic array → sub-region → intermediate region → overall base".
7. The method for calculating the maximum power output of a photovoltaic power station according to claim 1, characterized in that, The hierarchical dynamic graph structure in step S60 includes graph nodes and edge weights; The nodes in the diagram represent the various sub-regions, including the photovoltaic power station array as the bottom layer node, the primary sub-region as the middle layer node, and the intermediate region or the entire base as the high layer node. The edge weights represent the strength of meteorological correlations between grid nodes.
8. The method for calculating the maximum power output of a photovoltaic power station according to claim 1, characterized in that, Step S70 includes the following steps: Step S701: An LSTM (Long Short-Term Memory) network was introduced to explore the temporal variation of sub-region output patterns. The sub-region output pattern is as follows: h t =o t ⊙tanh(s t ), In the formula, A t J t ,o t ,s t ,h t These represent the module's input gate, forget gate, output gate, memory gate, and hidden gate, respectively. x ,Θ h ,b} are the specific model parameters of the gate, σ(·) is the activation function, and X t Represents the module output. As spatiotemporal related information X t Input t = 1, 2, 3, ⊙ represents the matrix multiplication operator, Θ W These are the corresponding matrix multiplication operators; Step S702: This matrix containing timing information The input is fed into the spatiotemporal graph neural network, and based on the spatiotemporal information of the current time step... The mining yielded spatiotemporal correlation information X with a dimension of 48×1. sub1 ; X sub1 =X⊙Θ sub1 , In the formula, X represents the output of the convolutional layer. Θ is the input to the convolutional layer, and Θ represents the parameters of the convolutional kernel to be trained. Let the node degree matrix be... Let I be the adjacency matrix at time t, and let Θ be the identity matrix. sub1 These are the corresponding subregion multiplication operators; Step S703: By analogy, define the spatiotemporal correlation information extracted from different sub-regions as X. sub1 ,X sub2 ,X sub3 …; X is obtained by modeling the global spatiotemporal correlation and fusing the features of all nodes through a fully connected layer. region The maximum power generation prediction value of the photovoltaic power station that can be obtained is Θ. p It is the corresponding global spatiotemporal correlation multiplication operator. p t：t+15 =X region ⊙Θ p 。 9. A non-transitory computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-8.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-8.