Short-term distributed photovoltaic power prediction method

Abstract

The invention discloses a short-term distributed photovoltaic power prediction method, and relates to the field of photovoltaic power analysis. According to the method, the independence of input data is ensured by determining different types of photovoltaic sites related to the target site in geographic and meteorological factors; the method comprises the following steps of: determining optimal graph structure data in all graph structure data by taking a graph connectivity measurement index as an evaluation index so as to simplify the data; the directed graph convolutional neural network model is constructed and trained through the optimal graph structure data, the causal relationship between the input variables is considered, and the precision of the prediction effect is improved. According to the short-term solar photovoltaic power prediction method, the data space-time correlation can be better mined, and the prediction accuracy and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic power analysis, and in particular to a short-term distributed photovoltaic power prediction method. Background Art

[0002] Existing distributed photovoltaic power forecasting methods usually rely on single-site data and simple time series models. However, these methods have the following problems:

[0003] Insufficient data independence: Failure to fully consider the spatial information of surrounding sites results in low prediction accuracy.

[0004] Data redundancy: The data of all neighboring sites are processed indiscriminately, which increases the complexity and computational effort of the model.

[0005] Limitations of the prediction model: It ignores the causal relationship between input variables, resulting in unsatisfactory prediction results. Summary of the invention

[0006] In view of the defects in the prior art, the technical problem solved by the present invention is: to provide a short-term solar photovoltaic power prediction method based on an optimal graph structure and a hierarchical directed graph convolution model to improve the accuracy and reliability of the prediction.

[0007] To achieve the above objectives, in a first aspect, an embodiment of the present application provides a short-term distributed photovoltaic power prediction method, the method comprising the following steps:

[0008] S1: Determine different types of photovoltaic sites related to the target site in terms of geographical and meteorological factors, and construct graph structure data based on the photovoltaic sites;

[0009] S2: Using the graph connectivity metric as the evaluation indicator, determine the optimal graph structure data among all graph structure data;

[0010] S3: A hierarchical directed graph convolutional neural network model is constructed based on the optimal graph structure data. The historical spatiotemporal data of the target site on the previous day, the historical spatiotemporal data of the neighboring sites on the previous day, and the NWP data of the target site on the next day are used as input, and the predicted photovoltaic power of the target site on the next day is used as output. After training the graph convolutional neural network prediction model, the prediction result of the short-term photovoltaic power of the target site is obtained.

[0011] In combination with the first aspect, in one implementation, the process of S1 includes:

[0012] Determine the relevance characteristics of neighboring PV sites to the target site, and collect historical irradiance data and NWP data of the target PV plant and its neighboring PV plants;

[0013] According to the correlation characteristics, historical irradiance data and NWP data, different types of typical PV stations related to the target station are determined through k-means clustering and weather type matching. CDI is used as the clustering effect evaluation index in k-means clustering.

[0014] Determine the graph structure data for each typical PV station.

[0015] In combination with the first aspect, in one implementation, the CDI is calculated as follows:

[0016]

[0017] Where C represents the average distance of each cluster center, and d(Ωk) represents the average distance of the kth class.

[0018] In combination with the first aspect, in one implementation, the correlation characteristics include solar radiation intensity, temperature, and weather factors.

[0019] In combination with the first aspect, in one implementation, the process of S2 includes:

[0020] Calculate the connectivity index value of each graph structure data, including selection degree, clustering coefficient, node mesoscopicity, average path length, algebraic connectivity and natural connectivity;

[0021] The entropy weight method is used to determine the weight of each connectivity index value;

[0022] According to the connectivity index value and corresponding weight of the graph structure data, the comprehensive evaluation index of each graph structure data is determined;

[0023] The optimal graph structure data is determined based on the comprehensive evaluation index.

[0024] In combination with the first aspect, in one implementation, the calculation formula of the connectivity index is:

[0025] Node Degree: where a ij is an element in the adjacency matrix. If there is a connection between node i and node j, then a ij =1, otherwise 0;

[0026] Clustering coefficient: Where E i is the number of edges that actually exist between neighbors of node i;

[0027] Node Mesoscopicity: Where N jl (i) is the number of paths passing through node i among all the shortest paths between node j and node l; N jl is the total number of all shortest paths between node j and node l;

[0028] Average path length: Where N is the total number of nodes in the network, d ij is the shortest path length between node i and node j;

[0029] Algebraic connectivity: D = diag{d i ,i=1,2,...n}λ, where the degree matrix D is used together with the adjacency matrix A to construct the Laplacian matrix L of the graph, L=DA; d i represents the degree of node i;

[0030] Natural Connectivity: where λ i is the eigenvalue of the Laplace matrix, and N is the total number of nodes in the network.

[0031] In combination with the first aspect, in one implementation, the comprehensive evaluation index of the graph structure data is obtained based on the sum of each connectivity index value of the graph structure data multiplied by its weight, and the optimal graph structure data is the graph structure data with the largest comprehensive evaluation index.

[0032] In combination with the first aspect, in one implementation, the process of S3 includes:

[0033] Construct the historical spatiotemporal data input layer and future meteorological data input layer of the target site based on the optimal graph structure data;

[0034] In each historical spatiotemporal data input layer and future meteorological data input layer, k adjacency matrices are obtained according to various thresholds ρ, and the adjacency matrix with the maximum connectivity is determined;

[0035] The convolutional neural network prediction model is constructed through the adjacency matrix of maximum connectivity, and the directed edges are used to indicate the unidirectional information flow for visualization, and a directed graph structure is obtained; the directed graph structure includes: the light intensity nodes of the target site and the adjacent photovoltaic stations, the NWP meteorological factor node of the target site, and the photovoltaic power node of the target site;

[0036] According to the number of time steps required for prediction, the predicted PV power and the number of PV stations, the inputs of the light intensity node, the NWP meteorological factor node and the PV power node, as well as the output of the target node are determined.

[0037] In combination with the first aspect, in one implementation, the light intensity node, the NWP meteorological factor node, and the photovoltaic power node are defined as X, W, and P, respectively, and K is the number of time steps required for prediction. is the predicted PV power at time t+1, N is the number of PV stations, and M is the number of selected NWP factors;

[0038] Input sample X t,i =[x t-H+1,i ,x t-H+2,i ,…,x t,i ]i∈[1,N];

[0039] The input sample W of W t,i =[w t+1,i ,w t+2,i ,…,w t+K,i ]i∈[1,M];

[0040] The input sample P of P t,i =[p t-H+1 ,p t-H+2 ,…,p t ];

[0041] Output of the target node

[0042] In combination with the first aspect, in one implementation, the predicted photovoltaic power P t The calculation formula is:

[0043]

[0044] Where P stc Represents the output of photovoltaic modules under standard conditions, Istc = 1000W / m2, Tstc = 25℃; I r,t represents the solar radiation intensity at time t; α T Represents the power temperature coefficient of the photovoltaic module; T t Indicates the actual temperature of the PV module at time t;

[0045] The forward operation formula of the convolutional neural network prediction model is:

[0046]

[0047] Where f(G) represents the prediction model based on graph G, which includes node set V and adjacency matrix A;

[0048] The graph convolution operation and back propagation calculation formula are as follows:

[0049] Θ* G x=Ug0U T x;

[0050] Where *G is the graph convolution operator, x is the signal of the node domain on the graph, g0 is the convolution kernel, and U is the eigenvector matrix of the graph Laplacian matrix;

[0051]

[0052] Where X L+1is the node feature matrix of the l+1th layer, which is the output after the graph convolution calculation; is a matrix with self-connection, In is the identity matrix, D is the degree matrix of A, X (l) is the node feature matrix of the lth layer, which is the input node feature; W (l) is the trainable weight matrix of the lth layer, which controls the weighting of each node feature; σ() represents the sigmoid activation function in the nonlinear model.

[0053] Compared with the prior art, the advantages of the present invention are:

[0054] The present invention ensures the independence of input data by determining different types of photovoltaic sites related to target sites in terms of geographical and meteorological factors; determines the optimal graph structure data among all graph structure data by taking graph connectivity metrics as evaluation indicators to streamline data; constructs and trains a directed graph convolutional neural network model through the optimal graph structure data, takes into account the causal relationship between input variables, and improves the accuracy of prediction effects.

[0055] Therefore, the present invention can better mine the short-term solar photovoltaic power prediction method of the temporal and spatial correlation of data, and improve the accuracy and reliability of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0057] Figure 1 A schematic diagram of a process flow of a short-term distributed photovoltaic power prediction method involved in an embodiment of the present application;

[0058] Figure 2 This is a schematic diagram of constructing the historical spatiotemporal data input layer and the future meteorological data input layer of the model in the embodiment of the present application.

[0059] Figure 3 This is a schematic diagram of a hierarchical directed graph structure model in an embodiment of the present application.

[0060] Figure 4 This is a schematic diagram of the definition of three types of nodes in the graph structure data of the embodiment of the present application;

[0061] Figure 5 This is a schematic diagram of the hardware structure of the short-term distributed photovoltaic power prediction device involved in the embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0063] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0064] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0065] First, see Figure 1 As shown, the embodiment of the present application provides a short-term distributed photovoltaic power prediction method, the steps of the method include:

[0066] S1: Identify different types of PV sites related to the target site in terms of geographical and meteorological factors to ensure the independence of input data; construct graph structure data based on the identified PV sites;

[0067] S2: Using the graph connectivity metric as the evaluation indicator, determine the optimal graph structure data among all graph structure data;

[0068] S3: A hierarchical directed graph convolutional neural network model is constructed based on the optimal graph structure data. The historical spatiotemporal data of the target site on the previous day, the historical spatiotemporal data of the neighboring sites on the previous day, and the NWP (Numerical Weather Prediction) data of the target site on the next day are used as input, and the predicted photovoltaic power of the target site on the next day is used as output. After training the graph convolutional neural network prediction model, the prediction result of the short-term photovoltaic power of the target site is obtained.

[0069] In one embodiment, the process of S1 includes:

[0070] S101: Based on the basic principles of photovoltaic power generation, geographical location and weather factors, analyze the correlation characteristics of neighboring photovoltaic sites to the target site; collect historical irradiance data and NWP (Numerical Weather Prediction) data of the target photovoltaic power station and its neighboring photovoltaic power stations.

[0071] S102: Based on the correlation characteristics, historical irradiance data and NWP data, different types of typical PV stations related to the target station are determined through k-means clustering and weather type matching. K-means clustering uses CDI (clustering distance index) as the clustering effect evaluation indicator. CDI takes into account both intra-class compactness and inter-class dispersion.

[0072] S103: Determine graph structure data of each typical photovoltaic station.

[0073] Specifically, the above correlation characteristics include solar radiation intensity, temperature, and weather factors. Weather factors affect the solar radiation received by photovoltaic power stations. Solar radiation is affected by many external factors, such as the position of the sun, cloud cover, and weather changes. Due to the movement of clouds, the radiation of neighboring photovoltaic power stations has a certain correlation. Geographical factors are mainly affected by latitude and longitude as well as terrain conditions, among which longitude affects the time delay of the photovoltaic output sequence, and latitude affects the height of the outer envelope of the photovoltaic output, that is, the output amplitude. In addition, terrain conditions, that is, altitude, affect the air pressure, air quality and temperature of the area. Therefore, from the perspective of geographical factors, the main influencing factors of spatial correlation should be geographical location and terrain conditions.

[0074] Specifically, the calculation formula of CDI is:

[0075]

[0076] Among them, C represents the average distance of each cluster center, d(Ωk) represents the average distance of the kth class; the smaller the CDI value, the better the clustering effect. The K-means algorithm is used for multiple clustering, and the K value with the smallest CDI index is taken as the final division result.

[0077] In one embodiment, the process of S2 includes:

[0078] S201: Calculate the connectivity index value of each graph structure data, the connectivity index includes selection degree, clustering coefficient, node mesoscopicity, average path length, algebraic connectivity and natural connectivity; among them, node degree, clustering coefficient and node mesoscopicity are local feature indicators, which are used to evaluate the importance of the target node in the entire graph network. On the other hand, average path length, algebraic connectivity and natural connectivity are global feature indicators, which are used to evaluate the robustness and connectivity of the network.

[0079] Specifically, the calculation formula of the connectivity index is:

[0080] Node Degree: where a ij is an element in the adjacency matrix. If there is a connection between node i and node j, then a ij =1, otherwise 0;

[0081] Clustering coefficient: Where E i is the number of edges that actually exist between neighbors of node i;

[0082] Node Mesoscopicity: Where N jl (i) is the number of paths passing through node i among all the shortest paths between node j and node l; N jl is the total number of all shortest paths between node j and node l;

[0083] Average path length: Where N is the total number of nodes in the network, d ij is the shortest path length between node i and node j;

[0084] Algebraic connectivity: D = diag{d i ,i=1,2,...n}λ=λ2, where the degree matrix D is used together with the adjacency matrix A to construct the Laplacian matrix L of the graph, L=DA; the algebraic connectivity is the second smallest eigenvalue (λ2) of the Laplacian matrix L; d i represents the degree of node i (i.e. the number of connections of node i);

[0085] Natural Connectivity: where λ i is the eigenvalue of the Laplace matrix, and N is the total number of nodes in the network.

[0086] S202: Use the entropy weight method to determine the weight of each connectivity index value (the weight of the same connectivity index of each graph structure data is the same, that is, the weight is a constant value); determine the comprehensive evaluation index of each graph structure data according to the connectivity index value of the graph structure data and the corresponding weight, and determine the optimal graph structure data for subsequent training according to the comprehensive evaluation index, so as to achieve higher prediction performance.

[0087] Specifically: the comprehensive evaluation index of the graph structure data is obtained according to the sum of each connectivity index value of the graph structure data multiplied by its weight, and the optimal graph structure data is the graph structure data with the largest comprehensive evaluation index.

[0088] In one embodiment, the process of S3 includes:

[0089] S301: See Figure 2 As shown, the historical spatiotemporal data input layer and future meteorological data input layer of the target site are constructed based on the optimal graph structure data.

[0090] S302: In each historical spatiotemporal data input layer and future meteorological data input layer, k adjacency matrices are obtained according to various thresholds ρ; and the optimal adjacency matrix (i.e., the adjacency matrix with maximum connectivity) is determined through the above-mentioned graph connectivity index.

[0091] S303: Construct a convolutional neural network prediction model through the adjacency matrix of maximum connectivity, use directed edges to indicate unidirectional information flow for visualization, and obtain a directed graph structure; see Figure 3 As shown, the directed graph structure contains three different types of nodes: the light intensity node of the target site and the adjacent photovoltaic stations, the NWP meteorological factor node of the target site, and the photovoltaic power node of the target site.

[0092] S304: Determine the inputs of the light intensity node, the NWP meteorological factor node and the photovoltaic power node, and the output of the target node according to the number of time steps required for prediction, the predicted photovoltaic power and the number of photovoltaic stations.

[0093] Specifically, participate Figure 4 As shown in the figure, the light intensity node, NWP meteorological factor node and photovoltaic power node are defined as X, W and P respectively, and K is the number of time steps required for prediction. is the predicted PV power at time t+1, N is the number of PV stations, and M is the number of selected NWP factors;

[0094] Input sample X t,i =[x t-H+1,i ,x t-H+2,i ,…,x t,i ]i∈[1,N];

[0095] The input sample W of W t,i =[w t+1,i ,w t+2,i ,…,w t+K,i ]i∈[1,M];

[0096] The input sample P of P t,i =[p t-H+1 ,p t-H+2 ,…,p t ];

[0097] Output of the target node

[0098] S305: Taking the historical spatiotemporal data of the target station and its neighboring stations on the previous day and the NWP meteorological data of the target station on the next day as input, and the predicted photovoltaic power of the target station on the next day as output, after training the convolutional neural network prediction model, the prediction result of the short-term photovoltaic power of the target station is obtained.

[0099] Specifically, the predicted photovoltaic power Pt The calculation formula is:

[0100]

[0101] Where P stc Represents the output of photovoltaic modules under standard conditions, Istc = 1000W / m2, Tstc = 25℃); I r,t represents the solar radiation intensity at time t; α T Represents the power temperature coefficient of the photovoltaic module, indicating the effect of temperature on the output power; T t Indicates the actual temperature of the photovoltaic module at time t; the actual output power P t As a prediction target, the formula can clarify which factors affect it, providing a theoretical basis for the input of subsequent model training.

[0102] The forward operation formula of the convolutional neural network prediction model is:

[0103]

[0104] Where f(G) represents the prediction model based on the graph G, G includes the information of nodes and edges, and G is a graph structure, that is, G includes the node set V and the adjacency matrix A.

[0105] The graph convolution operation and back propagation calculation formula are as follows:

[0106] Θ* G x=Ug0U T x;

[0107] Where *G is the graph convolution operator, x is the signal of the node domain on the graph, g0 is the convolution kernel, and U is the eigenvector matrix of the graph Laplacian matrix.

[0108]

[0109] Where X L+1 is the node feature matrix of the l+1th layer, which is the output after the graph convolution calculation; is a matrix with self-connection, In is the identity matrix, D is the degree matrix of A, X (l) is the node feature matrix of the lth layer, which is the input node feature; W (l) is the trainable weight matrix of the lth layer, which controls the weighting of each node feature; σ() represents the sigmoid activation function in the nonlinear model.

[0110] In a second aspect, an embodiment of the present application provides a short-term distributed photovoltaic power prediction device, which can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0111] Reference Figure 5 , Figure 5 The hardware structure diagram of the short-term distributed photovoltaic power prediction device involved in the embodiment of the present application is shown in FIG. In the embodiment of the present application, the short-term distributed photovoltaic power prediction device may include a processor, a memory, a communication interface, and a communication bus.

[0112] The communication bus may be of any type and is used to interconnect the processor, the memory, and the communication interface.

[0113] The communication interface includes input / output (I / O) interface, physical interface and logical interface, etc., which are used to realize the interconnection of devices inside the short-term distributed photovoltaic power prediction device, and the interface used to realize the interconnection of the short-term distributed photovoltaic power prediction device with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, a fiber optic interface, an ATM interface, etc.; the user device can be a display, a keyboard, etc.

[0114] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0115] The processor may be a general-purpose processor, which may call a short-term distributed photovoltaic power prediction program stored in a memory and execute the short-term distributed photovoltaic power prediction method provided in the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the short-term distributed photovoltaic power prediction program is called may refer to the various embodiments of the short-term distributed photovoltaic power prediction method of the present application, which will not be described in detail here.

[0116] Those skilled in the art will understand that Figure 5The hardware structure shown in the figure does not constitute a limitation on the present application, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0117] In a third aspect, an embodiment of the present application also provides a computer-readable storage medium.

[0118] The computer-readable storage medium of the present application stores a short-term distributed photovoltaic power prediction program, wherein when the short-term distributed photovoltaic power prediction program is executed by a processor, the steps of the short-term distributed photovoltaic power prediction method as described above are implemented.

[0119] Among them, the method implemented when the short-term distributed photovoltaic power prediction program is executed can refer to the various embodiments of the short-term distributed photovoltaic power prediction method of the present application, and will not be repeated here.

[0120] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0121] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0122] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0123] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0124] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0125] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0126] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0127] The above are only specific implementations of the embodiments of the present invention, but the protection scope of the embodiments of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the embodiments of the present invention, and these modifications or replacements should be included in the protection scope of the embodiments of the present invention. Therefore, the protection scope of the embodiments of the present invention shall be based on the protection scope of the claims.

Claims

1. A short-term distributed photovoltaic power prediction method, characterized in that: The method comprises the following steps: S1: Determine different types of photovoltaic sites related to the target site in terms of geographical and meteorological factors, and construct graph structure data based on the photovoltaic sites; S2: Using the graph connectivity metric as the evaluation indicator, determine the optimal graph structure data among all graph structure data; S3: A hierarchical directed graph convolutional neural network model is constructed based on the optimal graph structure data. The historical spatiotemporal data of the target site on the previous day, the historical spatiotemporal data of the neighboring sites on the previous day, and the NWP data of the target site on the next day are used as input, and the predicted photovoltaic power of the target site on the next day is used as output. After training the graph convolutional neural network prediction model, the prediction result of the short-term photovoltaic power of the target site is obtained.

2. The short-term distributed photovoltaic power prediction method according to claim 1, characterized in that: The S1 process includes: Determine the relevance characteristics of neighboring PV sites to the target site, and collect historical irradiance data and NWP data of the target PV plant and its neighboring PV plants; According to the correlation characteristics, historical irradiance data and NWP data, different types of typical PV stations related to the target station are determined through k-means clustering and weather type matching. CDI is used as the clustering effect evaluation index in k-means clustering. Determine the graph structure data for each typical PV station.

3. The short-term distributed photovoltaic power prediction method according to claim 2, characterized in that: The calculation formula of the CDI is: Where C represents the average distance of each cluster center, and d(Ωk) represents the average distance of the kth class.

4. The short-term distributed photovoltaic power prediction method according to claim 2, characterized in that: The correlation characteristics include solar radiation intensity, temperature, and weather factors.

5. The short-term distributed photovoltaic power prediction method according to claim 1, characterized in that: The S2 process includes: Calculate the connectivity index value of each graph structure data, including selection degree, clustering coefficient, node mesoscopicity, average path length, algebraic connectivity and natural connectivity; The entropy weight method is used to determine the weight of each connectivity index value; According to the connectivity index value and corresponding weight of the graph structure data, the comprehensive evaluation index of each graph structure data is determined; The optimal graph structure data is determined based on the comprehensive evaluation index.

6. The short-term distributed photovoltaic power prediction method according to claim 5, characterized in that: The calculation formula of the connectivity index is: Node Degree: where a ij is an element in the adjacency matrix. If there is a connection between node i and node j, then a ij =1, otherwise 0; Clustering coefficient: Where E i is the number of edges that actually exist between neighbors of node i; Node Mesoscopicity: Where N jl (i) is the number of paths passing through node i among all the shortest paths between node j and node l; N jl is the total number of all shortest paths between node j and node l; Average path length: Where N is the total number of nodes in the network, d ij is the shortest path length between node i and node j; Algebraic connectivity: D = diag{d i ,i=1,2,...n}λ, where the degree matrix D is used together with the adjacency matrix A to construct the Laplacian matrix L of the graph, L=DA; d i represents the degree of node i; Natural Connectivity: where λ i is the eigenvalue of the Laplace matrix, and N is the total number of nodes in the network.

7. The short-term distributed photovoltaic power prediction method according to claim 5, characterized in that: The comprehensive evaluation index of the graph structure data is obtained according to the sum of each connectivity index value of the graph structure data multiplied by its weight, and the optimal graph structure data is the graph structure data with the largest comprehensive evaluation index.

8. The short-term distributed photovoltaic power prediction method according to claim 1, characterized in that: The S3 process includes: Construct the historical spatiotemporal data input layer and future meteorological data input layer of the target site based on the optimal graph structure data; In each historical spatiotemporal data input layer and future meteorological data input layer, k adjacency matrices are obtained according to various thresholds ρ, and the adjacency matrix with the maximum connectivity is determined; The convolutional neural network prediction model is constructed through the adjacency matrix of maximum connectivity, and the directed edges are used to indicate the unidirectional information flow for visualization, and a directed graph structure is obtained; the directed graph structure includes: the light intensity nodes of the target site and the adjacent photovoltaic stations, the NWP meteorological factor node of the target site, and the photovoltaic power node of the target site; According to the number of time steps required for prediction, the predicted PV power and the number of PV stations, the inputs of the light intensity node, the NWP meteorological factor node and the PV power node, as well as the output of the target node are determined.

9. The short-term distributed photovoltaic power prediction method according to claim 8, characterized in that: Define the light intensity node, NWP meteorological factor node and photovoltaic power node as X, W and P respectively, K is the number of time steps required for prediction, is the predicted PV power at time t+1, N is the number of PV stations, and M is the number of selected NWP factors; Input sample X t,i =[x t-H+1,i ,x t-H+2,i ,…,x t,i ]i∈[1,N]; The input sample W of W t,i =[w t+1,i ,w t+2,i ,…,w t+K,i ]i∈[1,M]; The input sample P of P t,i =[p y-H+1 ,p t-H+2 ,…,p t ]; Output of the target node 10. The short-term distributed photovoltaic power prediction method according to claim 9, characterized in that: The predicted photovoltaic power P t The calculation formula is: Where P stc Represents the output of photovoltaic modules under standard conditions, Istc = 1000W / m2, Tstc = 25℃; I r,t represents the solar radiation intensity at time t; α T Represents the power temperature coefficient of the photovoltaic module; T t Indicates the actual temperature of the PV module at time t; The forward operation formula of the convolutional neural network prediction model is: Where f(G) represents the prediction model based on graph G, which includes node set V and adjacency matrix A; The graph convolution operation and back propagation calculation formula are: T* G x=Ug0U T x; Where *G is the graph convolution operator, x is the signal of the node domain on the graph, g0 is the convolution kernel, and U is the eigenvector matrix of the graph Laplacian matrix; Where X L+1 is the node feature matrix of the l+1th layer, which is the output after the graph convolution calculation; is a matrix with self-connection, In is the identity matrix, D is the degree matrix of A, X (l) is the node feature matrix of the lth layer, which is the input node feature; W (l) is the trainable weight matrix of the lth layer, which controls the weighting of each node feature; σ() represents the sigmoid activation function in the nonlinear model.

Images