Joint power characteristic analysis method for high-proportion distributed photovoltaic power station

Abstract

The invention discloses a joint power characteristic analysis method for a high-proportion distributed photovoltaic power station, and relates to the technical field of photovoltaic power analysis, and the method comprises the steps: collecting historical operation data, historical meteorological data and power grid topology data of each distributed photovoltaic power station in a target region; constructing a space-time diagram structure; constructing a space-time diagram neural network model to obtain a combined power characteristic twin model; and inputting the analog input data into the combined power characteristic twin model, performing scene deduction of combined power, and outputting a combined power time sequence, a fluctuation characteristic and a spatial distribution characteristic of the target area. According to the method, the technical problems of low analysis precision and difficult scheduling optimization caused by difficulty in realizing scene deduction and lack of an effective scheme for supporting scheduling decision in photovoltaic power station joint power characteristic analysis in the prior art are solved, photovoltaic joint power scene deduction is realized, and the scheduling optimization efficiency is improved. The technical effects of precise analysis of the joint power characteristics of the photovoltaic power station and scheduling decision optimization are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power analysis, in particular to a high-proportion distributed photovoltaic power station joint power characteristic analysis method. BACKGROUND

[0002] With the increasing penetration of distributed photovoltaic power stations in the power system, the output of the photovoltaic power stations is significantly volatile and uncertain due to the influence of multiple factors such as meteorological conditions and power grid topology. Traditional power analysis methods rely on a single data dimension or a simplified model, which is difficult to accurately depict the electrical, geographical and meteorological correlation characteristics among multiple power stations. As a result, the joint power time series prediction has a large deviation, the fluctuation characteristic analysis is insufficient, the spatial distribution characteristic description is not comprehensive, and there is a lack of constraint mechanism combining physical laws and deduction ability adapting to different scenarios, which cannot provide accurate and reliable power characteristic data support for power grid dispatching.

[0003] The prior art has the technical problems that in the joint power characteristic analysis of photovoltaic power stations, it is difficult to realize scenario deduction, and there is a lack of effective scheme to support dispatching decision, resulting in low analysis accuracy and difficult dispatching optimization. SUMMARY

[0004] The present application provides a high-proportion distributed photovoltaic power station joint power characteristic analysis method, which is used to solve the technical problems that in the joint power characteristic analysis of photovoltaic power stations in the prior art, it is difficult to realize scenario deduction, and there is a lack of effective scheme to support dispatching decision, resulting in low analysis accuracy and difficult dispatching optimization.

[0005] In view of the above problems, the present application provides a high-proportion distributed photovoltaic power station joint power characteristic analysis method.

[0006] The present application provides a high-proportion distributed photovoltaic power station joint power characteristic analysis method, which comprises: Collecting historical operation data, historical meteorological data and power grid topology data of each distributed photovoltaic power station in a target area; based on the power grid topology data, taking each distributed photovoltaic power station as a node and the spatial relationship between the stations as an edge, a space-time graph structure is constructed; based on the space-time graph structure, a space-time graph neural network model is constructed, and a physical constraint term is introduced to train the model using the historical operation data and the historical meteorological data, to obtain a joint power characteristic twin model; inputting simulation input data into the joint power characteristic twin model to perform scenario deduction of joint power, and outputting joint power time series, fluctuation characteristics and spatial distribution characteristics of the target area.

[0007] One or more technical solutions provided in the present application have at least the following technical effects or advantages: Historical operational data, historical meteorological data, and grid topology data of each distributed photovoltaic (PV) power station within the target area are collected. A spatiotemporal graph structure is constructed using each PV power station as a node and the spatial relationships between stations as edges. A spatiotemporal graph neural network model is built, and physical constraints are introduced. The model is trained using the historical operational data and historical meteorological data to obtain a joint power characteristic twin model. Simulated input data is input into the joint power characteristic twin model to perform joint power scenario simulation, outputting the joint power time series, fluctuation characteristics, and spatial distribution characteristics of the target area. This achieves PV joint power scenario simulation, enabling precise analysis of the joint power characteristics of PV power stations and optimization of scheduling decisions. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic flowchart of a method for analyzing the combined power characteristics of a high-proportion distributed photovoltaic power station, provided in an embodiment of this application.

[0010] Figure 2 This is a schematic diagram of the process for constructing a spatiotemporal graph structure in a method for analyzing the joint power characteristics of a high-proportion distributed photovoltaic power station provided in an embodiment of this application. Detailed Implementation

[0011] This application provides a method for analyzing the joint power characteristics of high-proportion distributed photovoltaic power plants, which addresses the technical problems in existing photovoltaic power plant joint power characteristic analysis, such as difficulty in scenario extrapolation and lack of effective solutions to support scheduling decisions, resulting in low analysis accuracy and difficulty in scheduling optimization.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0013] Examples, such as Figure 1 As shown, this application provides a method for analyzing the combined power characteristics of a high-proportion distributed photovoltaic power station, the method comprising: Step S100: Collect historical operation data, historical meteorological data, and power grid topology data of each distributed photovoltaic power station in the target area.

[0014] Specifically, first, the scope and object of data collection are determined, focusing on the target area to be analyzed, covering all distributed photovoltaic power stations in the area, and ensuring that no data is missed. For historical operation data, the actual operation parameters of each distributed photovoltaic power station at different time periods, such as daily, monthly, and different seasons, are collected, including but not limited to real-time output power, cumulative power generation, inverter working status, such as operating efficiency, start-stop record, fault alarm information, and working temperature of photovoltaic modules. These data need to be accurately matched with specific time nodes to reflect the time sequence variation law of power station operation status. For historical meteorological data, on the one hand, the overall meteorological records of the target area are collected, covering macroscopic meteorological parameters such as solar intensity, solar radiation per unit time, ambient temperature, relative humidity, wind speed, wind direction, and cloud cover. On the other hand, for the sites of each distributed photovoltaic power station, microcosmic meteorological data at the site level are supplemented to ensure that the meteorological data are accurately matched with the power station location and the time dimension is consistent with the historical operation data, so as to establish the correlation between meteorological conditions and power station power output. In the collection of power grid topology data, the structure information of the power grid in the target area is mainly obtained, including the access point location of each distributed photovoltaic power station and the power grid, the distribution and capacity of the transformer station, the model, length, and impedance parameters of the transmission line, the division of the power supply area to which each power station belongs, and the electrical connection path between sites, such as whether there is a direct electrical path, etc. These data will provide basic support for subsequent construction of spatial correlation and electrical relationship between power stations. Through systematic collection and arrangement of the three types of data, a complete original data set is formed.

[0015] Step S200: Based on the power grid topology data, the time-space graph structure is constructed with each distributed photovoltaic power station as a node and the spatial relationship between sites as an edge.

[0016] Specifically, each distributed photovoltaic power station in the target area constructs a feature vector, each feature vector at least covers the core attribute parameters of the corresponding power station, including installed capacity, historical average unit capacity power, geographic location coordinates. Then, all feature vectors of the power stations are stacked in a predetermined order to form an initial feature matrix representing the basic information of the nodes of the graph structure, and the inherent attributes of each node are determined. Next, based on the spatial relationship between the grid topology data and the sites, including electrical distance, geographical distance, and meteorological correlation, the effective neighborhood of each node is selected and the weight of the edge is calculated: for any node, first, the K nearest nodes are selected as the candidate neighborhood set according to the geographical distance, and then by querying the grid topology data, the nodes without direct electrical access or in different power supply areas are excluded to form a neighborhood set that only retains effective associations. Then, within the effective neighborhood, the three types of weights between two nodes are calculated: 1. Electrical distance weight, which is inversely proportional to the electrical coupling impedance modulus value of the two nodes in the grid topology; 2. Geographical distance weight, which is inversely proportional to the actual geographical distance between the two nodes; 3. Meteorological correlation weight, which takes the Pearson correlation coefficient of the historical power sequence of the two nodes. After normalizing the three types of weights and weighting them according to the preset proportion, the correlation strength between the nodes is obtained, which defines the weight of the edge in the graph structure, and generates a weighted adjacency matrix representing the spatial correlation between nodes. Finally, combining the initial feature matrix and the weighted adjacency matrix, a spatio-temporal graph structure is constructed, which has node attributes, spatial correlation between nodes, and temporal characteristics. This structure can accurately reflect the inherent characteristics of each distributed photovoltaic power station and the electrical, geographical, and meteorological correlations between them.

[0017] Step S300: Based on the spatio-temporal graph structure, a spatio-temporal graph neural network model is constructed, and a physical constraint term is introduced to train the model using the historical operation data and the historical meteorological data, obtaining a joint power characteristic twin model.

[0018] Specifically, based on the constructed spatio-temporal graph structure, a spatio-temporal graph neural network model is built. The model can capture the spatial correlation between the nodes of the distributed photovoltaic power stations through the graph convolution layer, and mine the time series variation law of the power output through the time series convolution layer or the recurrent layer, so as to realize the collaborative extraction of the spatio-temporal dimension features. Subsequently, physical constraint terms conforming to the operation law of the photovoltaic power station are introduced, at least including the following one: an energy conservation constraint term, which restricts the deviation range of the total irradiance energy and the theoretical maximum power generation potential of the region, so as to avoid that the model output deviates from the energy conservation law; a spatial correlation constraint term, which restricts the power correlation between the stations in the model output to show a physical attenuation trend with the distance, so as to conform to the actual geographical and electrical correlation characteristics; a fluctuation limit constraint term, which restricts the power change rate in the model output to be within the limit value based on the physical performance of the inverter, so as to ensure that the output conforms to the equipment operation capability. Then, two types of training scene libraries are constructed by combining the historical operation data and the historical meteorological data: the first training scene library takes different meteorological scenes, such as combinations of solar intensity and temperature changes, as inputs, and takes the joint power time series, fluctuation characteristics and spatial distribution characteristics of the target region under the corresponding scene as outputs; the second training scene library takes the grid structure change parameters, such as line impedance adjustment and power station access state change, and the fixed meteorological scene as inputs, and takes the joint power characteristics under the corresponding conditions as outputs. After that, the physical constraint terms are integrated into the model training loss function, and the constraint deviation is calculated into the loss calculation by weighting, so as to improve the compliance of the model to the physical law. The first twin channel is obtained by training the model with the first training scene library, which focuses on the response of the meteorological changes to the joint power; the second twin channel is obtained by training the model with the second training scene library, which focuses on the response of the grid structure changes to the joint power; finally, the two twin channels are connected and integrated to form a joint power characteristics twin model which can adapt to meteorological and grid structure changes.

[0019] Step S400: inputting the simulation input data into the joint power characteristics twin model to perform scene deduction of the joint power, and outputting the joint power time series, fluctuation characteristics and spatial distribution characteristics of the target region.

[0020] Specifically, the input simulation data is classified, and it is judged whether it belongs to the meteorological scene type data, such as simulated different solar intensity and temperature combination scenes, or the grid structure change type data, such as simulated line impedance adjustment and power station access state change scenes, which need to be matched with fixed meteorological parameters, so as to determine the matching twin channel in the joint power characteristics twin model. If it is the meteorological scene type data, the first twin channel which focuses on the response of the meteorological changes is triggered; if it is the grid structure change type data, the second twin channel which focuses on the response of the grid structure adjustment is triggered.

[0021] Subsequently, after the analog input data is preprocessed in the format required by the model, it is input into the matched twin channel for joint power scenario deduction: the spatio-temporal graph neural network in the channel will dynamically calculate the power output of each distributed photovoltaic power station in the target area based on the spatio-temporal correlation rules and physical constraint rules learned in the early training, combined with the characteristics of the analog input data, while taking into account the influence of the electrical, geographical, and meteorological correlations between the stations on the joint power.

[0022] Finally, the model outputs three types of core results: first, the joint power time sequence, which presents the total joint power change of the target area at different time periods under the deduction scenario in the form of a time series curve; second, the fluctuation characteristics, including power fluctuation amplitude, fluctuation frequency, etc., reflecting the stability of the joint power; third, the spatial distribution characteristics, which show the power output differences and spatial correlation state of each distributed photovoltaic power station under the deduction scenario in the form of a heat map or data table, providing accurate data support for photovoltaic power dispatching, power grid planning, and risk prediction in the target area.

[0023] In one possible implementation manner, step S200 further includes: Step S210: the spatial relationship includes electrical distance, geographical distance, and meteorological correlation.

[0024] Specifically, the spatial relationship includes electrical distance, geographical distance, and meteorological correlation, which are the core dimensions for quantifying the correlation strength between nodes of each distributed photovoltaic power station when constructing the spatio-temporal graph structure. The electrical distance is calculated based on the power grid topology data, reflecting the electrical coupling tightness of two power stations in the power grid system, and the correlation weight is inversely proportional to the electrical coupling impedance modulus value between the two power stations. The geographical distance is determined based on the geographical location coordinates of the power stations, representing the physical spatial interval between the two power stations, and the correlation weight is inversely proportional to the actual geographical distance, which is a key basis for filtering node candidate neighbors. The meteorological correlation is quantified by analyzing the Pearson correlation coefficient of the historical power sequence of the two power stations, reflecting the synchronicity of the two power stations affected by meteorological conditions. The higher the correlation coefficient, the more obvious the power output coordination change characteristics. The three complement each other from the aspects of power grid electrical characteristics, physical spatial location, and meteorological influence coordination, providing comprehensive basis for defining the weight of edges in the spatio-temporal graph structure and constructing the weighted adjacency matrix.

[0025] In one possible implementation manner, as shown in Figure 2 Step S200 further includes: Step S220: constructing a feature vector for each distributed photovoltaic power station, each feature vector including at least the installed capacity, the historical average unit capacity power, and the geographical location coordinates of the corresponding power station.

[0026] Step S230: Based on the power grid topology data, stack the feature vectors of the respective distributed photovoltaic power stations as nodes to form an initial feature matrix of nodes.

[0027] Step S240: Based on the spatial relationship, calculate the correlation strength between nodes, define the weight of the edge, perform weighted linear combination, generate a weighted adjacency matrix representing the spatial relationship between nodes, and establish the spatio-temporal graph structure.

[0028] Specifically, the construction object of the feature vector is all distributed photovoltaic power stations in the target area, ensuring that each station participating in joint power characteristic analysis is covered without omission. Then, the core attribute parameters of each power station are extracted one by one to form a feature vector, wherein the installed capacity parameter directly uses the rated installed capacity of the power station design or record, which is used to intuitively reflect the maximum power generation potential of the power station; the historical average unit capacity power parameter is calculated by calling the historical operation data of the power station, such as the actual output power record in the past 1 year or 3 quarters, to calculate the average output power per kilowatt in the statistical period, thereby reflecting the long-term power generation efficiency of the power station; the geographic location coordinate parameter is obtained by GPS positioning or grid file query, accurately recording the longitude and latitude information of the power station, laying a foundation for subsequent calculation of geographical distance between stations and analysis of spatial correlation; finally, the three types of parameters are integrated into a vector form in a predetermined order to form a feature vector exclusive to each distributed photovoltaic power station, ensuring that the vector can fully and accurately represent the key inherent attributes of the corresponding power station, providing standardized basic data for subsequent construction of the initial feature matrix of nodes.

[0029] Based on the power grid topology data, the access position and correlation of each distributed photovoltaic power station in the target area in the power grid system are determined, and the node identity of each station in the spatio-temporal graph structure is determined based on this, ensuring that the node division matches the actual structure of the power grid; then, the feature vector constructed for each distributed photovoltaic power station is called, each vector containing core parameters such as installed capacity, historical average unit capacity power, and geographic location coordinates, and the ordered arrangement of all feature vectors is performed according to the preset node sorting rule and the access order of the power station in the power grid topology, the order from east to west / south to north of the geographic location; finally, the stacked and integrated feature vectors are arranged in rows or columns, if each feature vector is m-dimensional, such as 3 core parameters corresponding to 3 dimensions, and there are n distributed photovoltaic power stations in the target area, then the stacked matrix is n rows m columns or m rows n columns, the matrix is the initial feature matrix of nodes, and each row (or column) in the matrix corresponds to the complete feature information of a power station node, clearly presenting the inherent attributes of all nodes.

[0030] With the spatial relationship including electrical distance, geographical distance and meteorological correlation as the core basis, for the determined distributed photovoltaic power station nodes, the effective neighborhood of each node is first screened, the K nodes closest in geographical distance are selected as candidate neighborhoods, and nodes without direct electrical access or in different power supply areas are excluded in combination with the power grid topology data to ensure that only node pairs with actual relevance are calculated; then, the nodes in the effective neighborhood are calculated for three types of association weights: the electrical distance weight is inversely proportional to the electrical coupling impedance modulus of the two nodes, the geographical distance weight is inversely proportional to the actual geographical distance of the two nodes, and the meteorological correlation weight is the Pearson correlation coefficient of the historical power sequence of the two nodes. The three types of weights are normalized and weighted and summed according to a predetermined proportion to obtain the association strength between nodes, and the association strength is used to define the weight of the edge between the two nodes; then, the edge weights of all node pairs are integrated through weighted linear combination to generate a weighted adjacency matrix of n rows and n columns, n being the total number of nodes. Each element in the matrix corresponds to the association weight of a pair of nodes, directly reflecting the closeness of the spatial association between nodes; finally, the initial feature matrix of the nodes formed and the weighted adjacency matrix generated in this step are combined to construct a spatio-temporal graph structure that combines the inherent properties of nodes and the spatial association information between nodes.

[0031] In one possible implementation manner, the step S240 further includes: Step S241: calculating an electrical distance weight between the nodes, the electrical distance weight being inversely proportional to the electrical coupling impedance modulus of the two nodes in the power grid topology.

[0032] Step S242: calculating a geographical distance weight between the nodes, the geographical distance weight being inversely proportional to the geographical distance of the two nodes.

[0033] Step S243: calculating a meteorological correlation weight between the nodes, the meteorological correlation weight being the Pearson correlation coefficient of the historical power sequence of the two nodes.

[0034] Step S244: normalizing and weighting the electrical distance weight, the geographical distance weight and the meteorological correlation weight to obtain the association strength between the nodes.

[0035] Step S245: defining the weight of the edge between the nodes with the association strength to construct the weighted adjacency matrix.

[0036] Specifically, the calculation object is all distributed photovoltaic power station nodes in the target area, i.e., the two-by-two combination of each power station, ensuring that all possible node pairs with electrical association are covered; then, based on the power grid topology data collected in the early stage, the electrical connection parameters of each pair of nodes (power stations) in the power grid system are extracted, and key data such as the impedance of the transmission line between the two nodes and the equivalent impedance of the transformer are obtained. The electrical coupling impedance modulus between the two nodes, i.e., the absolute value of the impedance, is calculated through circuit theory, reflecting the degree of hindrance of electrical connection; subsequently, a calculation model is constructed according to the rule that the electrical distance weight is inversely proportional to the electrical coupling impedance modulus, in the form of electrical distance weight = 1 / (electrical coupling impedance modulus + ε), where ε is a very small positive number to avoid a zero denominator, so that the smaller the electrical coupling impedance modulus between the two nodes, the less the loss and interference of electric energy transmission between the two nodes, the closer the electrical association, and the greater the electrical distance weight; finally, the electrical distance weight between all pairs of nodes is calculated one by one through the model.

[0037] The calculation object is the two-by-two combination of all distributed photovoltaic power station nodes in the target area, ensuring that all node pairs that need to be analyzed for spatial correlation are covered; then, the geographic location coordinates, i.e., latitude and longitude data, in the feature vector constructed by each node are retrieved, and the actual straight-line distance between each pair of nodes in physical space is calculated one by one using the Euclidean distance formula selected according to the size of the region, obtaining the geographic distance quantization value of the two nodes; subsequently, a calculation model is constructed according to the rule that the geographic distance weight is inversely proportional to the geographic distance, for example, in the form of geographic distance weight = 1 / (geographic distance + ε), where ε is a very small positive number to avoid abnormal weights due to geographic distance approaching zero, so that the closer the geographic distance between the two nodes, the more similar the influence of the same regional microclimate, such as local cloud coverage, small-scale wind speed changes, and terrain conditions, the stronger the basic correlation of power output characteristics, and the greater the geographic distance weight; finally, the geographic distance weight between all pairs of nodes is calculated through the model, providing a quantitative basis for the physical space level for subsequent fusion of multi-dimensional weights and determination of the comprehensive correlation strength of the nodes.

[0038] The calculation object is all pairwise combinations of distributed photovoltaic power station nodes in the target area, ensuring that all node pairs related to weather are covered; then, the historical operation data of each node is called, and the power output records corresponding to the same time period are extracted to form the historical power sequence of each node, which is recorded in hourly or minute granularity to ensure that the power sequences of two nodes are completely aligned in the time dimension; then, the Pearson correlation coefficient calculation formula is used to analyze the correlation of the historical power sequence of each node pair, and the correlation coefficient with a value range of [-1, 1] is obtained by calculating the ratio of the covariance of the two sequences to the product of the standard deviations. The closer the coefficient is to 1, the more consistent the power output of the two nodes is with the change trend of the weather conditions such as solar intensity, temperature, and cloud coverage, and the stronger the synchronization of the influence of the same weather factor; finally, the Pearson correlation coefficient calculated is directly used as the weather correlation weight of the node pair, which quantifies the power coordination characteristics of the two nodes caused by the correlation of weather conditions.

[0039] The three types of weights calculated, i.e., electrical distance weight, geographical distance weight, and weather correlation weight, are normalized respectively. The min-max normalization method is used to map the value range of each type of weight to the [0, 1] interval, i.e., for all node pairs in each type of weight, the formula (original value-minimum value) / (maximum value-minimum value) is used to calculate, which eliminates the influence of different weight dimensions caused by the difference in dimensions and numerical range, and ensures that the three types of weights are comparable and superimposable; then, according to the actual operation characteristics and analysis requirements of photovoltaic power stations in the target area, the proportion coefficients of the three types of weights in the comprehensive correlation strength are preset, such as the electrical distance weight coefficient 0.4, the geographical distance weight coefficient 0.3, and the weather correlation weight coefficient 0.3. The sum of the coefficients is 1, and the coefficients can be dynamically adjusted according to the grid dispatching emphasis or historical data analysis effect; then, for each node pair, the normalized three types of weights are multiplied by the corresponding coefficients, and the product results are added up to obtain the correlation strength of the node pair, which fully integrates the correlation information of the three dimensions of electrical coupling, physical space, and weather driving; finally, the above process is completed to calculate the correlation strength between all nodes, which provides a comprehensive quantitative basis for defining the weight of the edge and constructing the weighted adjacency matrix.

[0040] The calculated correlation strength between each pair of nodes is taken as the weight value of the corresponding node pair edge in the space-time graph structure, which comprehensively reflects the correlation closeness of the two nodes in the electrical, geographical, and meteorological dimensions; then, according to the preset ordering of the distributed photovoltaic power station nodes in the target region, such as according to the power station number or the access order in the power grid topology, the correlation strength of all node pairs is arranged in matrix form according to the order of the corresponding nodes in the row and column. If there are n nodes in the target region, an n-row and n-column matrix is constructed, and the element value of the ith row and jth column in the matrix is the correlation strength between the ith node and the jth node, that is, the weight of the edge. The diagonal element, the correlation of the node with itself, can be set to 0 or 1, which is determined according to the model requirements. Finally, the matrix formed by the above arrangement is the weighted adjacency matrix, which completely and accurately represents the spatial correlation relationship between all nodes, and together with the formed node initial feature matrix, it constitutes the core data of the space-time graph structure, providing structured graph data input for the subsequent training of the space-time graph neural network model.

[0041] In one possible implementation manner, step S240 further includes: For any node, the K closest nodes are selected as the candidate neighborhood set based on geographical distance.

[0042] The candidate neighborhood set is traversed, and nodes without direct electrical access or in different power supply areas are removed by querying the power grid topology data to form an effective neighborhood set.

[0043] In the effective neighborhood set, the electrical distance weight, geographical distance weight, and meteorological correlation weight of the any node and other nodes in the set are calculated.

[0044] Specifically, the geographical position coordinates (longitude and latitude data) of the target node and all other nodes in the target region are retrieved, and the Euclidean distance formula adapted according to the region range is used to calculate the geographical straight-line distance between the target node and each of the other nodes one by one to obtain the geographical distance quantization results of the target node and all other nodes. Then, these geographical distances are sorted in ascending order of numerical value, and the first K nodes with the smallest distance are selected from the sorting results, where K is a neighborhood size parameter preset according to actual analysis requirements, such as K = 5 or K = 10, which can be dynamically adjusted according to the power station density and power grid structure complexity of the target region. Finally, the K closest nodes are integrated to form the candidate neighborhood set of the target node, and the potential neighbor nodes most closely associated with the target node in the physical space level are preliminarily locked.

[0045] For any determined distributed photovoltaic power station node (target node) and its candidate neighborhood set, each node in the set is traversed one by one, and each to-be-verified node and the target node are taken as a group of to-be-checked node pairs; then, the power grid topology data collected in the early stage is called, and the connection relationship of the node pair in the power grid system is queried, on the one hand, whether the two nodes have a direct electrical path, such as being directly connected through a power transmission line, being connected through the same distribution branch line, etc., is checked, and the case that there is no direct line connection or the two nodes belong to disconnected branches of the power grid is excluded, and on the other hand, whether the two nodes are in the same power supply area, that is, whether they are supplied by the same transformer substation or the same distribution transformer, is confirmed, and the case that the two nodes belong to different power supply management units is excluded; finally, the nodes that do not meet the condition of "having a direct electrical path and belonging to the same power supply area" after checking are removed from the candidate neighborhood set, and the remaining nodes are integrated to form an effective neighborhood set of the target node, so as to ensure that the nodes in the set and the target node have both geographical proximity and correlation basis in the actual power grid operation level.

[0046] For any distributed photovoltaic power station node (target node) and its effective neighborhood set, other nodes in the set are selected one by one to form node pairs with the target node, and three types of weights are calculated, first, the electrical coupling impedance modulus value of the node pair is extracted based on the power grid topology data, and the electrical distance weight is calculated by the formula 1 / (impedance modulus+ε) according to the rule that the electrical distance weight is inversely proportional to the impedance modulus, wherein ε is a very small positive number; second, the geographical position coordinates of the two nodes are called, the actual geographical distance is calculated by the Euclidean distance formula, and the geographical distance weight is calculated by 1 / (geographical distance+ε) according to the rule that the geographical distance weight is inversely proportional to the geographical distance, wherein ε is a very small positive number; finally, the historical power sequences of the two nodes aligned in time dimension are called, the correlation coefficient is calculated by the Pearson correlation coefficient formula, the ratio of the covariance and the standard deviation product is obtained, and the correlation coefficient with a value range of [-1, 1] is taken as the weather correlation weight, and finally the three types of weights of the target node and all other nodes in the effective neighborhood set are calculated, which provides accurate local weight data for subsequent integration of correlation strength and construction of a weighted adjacency matrix.

[0047] In a possible implementation manner, step S300 further includes: Step S310: The physical constraint term includes at least one of an energy conservation constraint term, a spatial correlation constraint term, and a fluctuation limit constraint term.

[0048] The energy conservation constraint term is used to constrain the deviation range between the total irradiation energy of the region and the theoretical maximum power generation potential; the spatial correlation constraint term is used to constrain the power correlation between the model output sites to comply with the physical decay law of distance; and the fluctuation limit constraint term is used to constrain the power variation rate of the model output to be less than the limit value based on the physical performance of the inverter.

[0049] Specifically, the physical constraint term is the key to ensure that the joint power characteristic twin model output conforms to the actual operation law of the photovoltaic power station. It contains at least one of the energy conservation constraint term, the spatial correlation constraint term and the fluctuation limit constraint term, and the specific definitions and functions are as follows: the energy conservation constraint term is based on the total irradiance energy of the target area, combined with the installed capacity, photoelectric conversion efficiency and other parameters of each distributed photovoltaic power station to calculate the theoretical maximum power generation potential of the region. By setting a deviation threshold, such as ±5% of the theoretical value, the deviation range of the joint power total energy of the model output and the theoretical maximum power generation potential is constrained to avoid unreasonable output of the model that deviates from the law of conservation of energy due to data noise or overfitting; the spatial correlation constraint term is based on the physical law that the closer the distance between power stations, the stronger the synchronization of power output affected by weather and geographical factors, and the higher the correlation. By pre-setting a correlation decay function, such as exponential decay with increasing distance, the power correlation between different stations of the model output needs to conform to the decay trend to prevent the result of appearing high power correlation in stations far away from each other which violates the actual physical logic; the fluctuation limit constraint term determines the power change rate limit value based on the physical performance parameters of the inverter, such as the maximum climbing rate and the maximum descending rate, and constrains the power change rate of the model output at any time, i.e. the ratio of the power difference between adjacent time steps to the time step, not to exceed the limit value, to ensure that the power fluctuation of the model output conforms to the actual operation capacity of the equipment and avoid extreme fluctuation results that exceed the adjustment range of the inverter. The three types of constraint terms can be used singly or in combination according to the model training scene requirements to provide physical level guarantee for the rationality of the model output.

[0050] In one possible implementation manner, the step S300 further includes: Step S320: determining a training loss function of the spatio-temporal graph neural network model.

[0051] Step S330: combining the physical constraint term with the training loss function, and training the model by using the historical operation data and the historical meteorological data to obtain the joint power characteristic twin model.

[0052] Specifically, according to the core task target of the model: accurately predicting the joint power time sequence, fluctuation characteristics and spatial distribution characteristics of the target area, it is clear that the loss function needs to be around the deviation of the quantitative model prediction value and the true value, and needs to adapt to the continuous regression characteristics of the power data; then, the type of the basic loss function is selected, and the mean square error (MSE) loss function is preferred, the calculation logic of which is: extracting the real joint power data in the historical operation data, such as the total joint power of each time step area and the power output value of each power station, as a label, comparing the predicted power data output by the model, calculating the square difference of each group of predicted value and true value, and then taking the average of all square differences, so as to highlight the influence of large deviation on the loss and strengthen the fitting accuracy of the model to the key power nodes; at the same time, auxiliary loss terms can be supplemented according to the training scene requirements, for example, the mean absolute error (MAE) loss is added for fluctuation characteristic prediction to calculate the absolute difference average of the predicted and true fluctuation amplitude, or the cosine similarity loss is added for spatial distribution characteristics to quantify the similarity deviation of the predicted power station power spatial distribution and the true distribution, avoiding the fitting deviation caused by a single loss function; finally, the basic loss function and the auxiliary loss term are combined with a preset weight, such as 0.7 for MSE and 0.3 for MAE, to form a complete training loss function, ensuring that it can comprehensively cover the prediction dimensions that the model needs to optimize.

[0053] The determined physical constraint term is converted into a quantifiable constraint loss term, for example, the energy conservation constraint term is calculated by the deviation square of the total power of the region output by the model and the theoretical maximum power generation potential as the constraint loss, the spatial correlation constraint term is calculated by the deviation of the predicted station power correlation and the distance attenuation law as the constraint loss, and the fluctuation limit constraint term is calculated by the absolute value of the power change rate exceeding the limit as the constraint loss; then, the weight coefficient of each constraint loss term is allocated, and the basic training loss function is weighted and summed to construct a comprehensive loss function containing physical rules; subsequently, the historical meteorological data, such as time series data of sunshine, temperature, wind speed, etc., and the power grid topology characteristics are used as the model input, the actual power output in the historical operation data, including the power time sequence of each power station and the total power of the region, are used as the label, the data is divided into training set and validation set in time sequence, and the parameters of the spatio-temporal graph neural network model are iteratively optimized through the back propagation algorithm; during the training process, the comprehensive loss function value and the satisfaction degree of each physical constraint term are monitored in real time, when the prediction error of the model on the validation set continues to decrease and the physical constraint deviation is stable within the preset threshold, such as energy conservation deviation < 5%, fluctuation exceeding limit times < 1%, the iteration is stopped; finally, the model parameters at this time are saved, and a joint power characteristic twin model that can accurately fit the historical data and strictly follow the physical law is obtained, providing a reliable tool for subsequent joint analysis and prediction of regional photovoltaic power.

[0054] In one possible implementation manner, step S320 further includes: Step S321: constructing a first training scenario library and a second training scenario library based on the historical meteorological data and the historical operation data.

[0055] The input of each training scenario in the first training scenario library is a meteorological scenario, and the output is the joint power time sequence, fluctuation characteristics and spatial distribution characteristics of the target region. The input of each training scenario in the second training scenario library is a power grid structure change parameter and a fixed meteorological scenario, and the output is the joint power time sequence, fluctuation characteristics and spatial distribution characteristics of the target region.

[0056] Step S322: combining the physical constraint term with the training loss function, training the spatio-temporal graph neural network model based on the first training scenario library to obtain a first twin channel, and training the spatio-temporal graph neural network model based on the second training scenario library to obtain a second twin channel.

[0057] Step S323: connecting the first twin channel and the second twin channel to generate the joint power characteristic twin model.

[0058] Specifically, the historical meteorological data used to build the scenario library includes time series data of solar intensity, temperature, wind speed, cloud coverage, etc. at different time periods, and covers typical weather types such as sunny, overcast, cloudy, and rainy. The historical operation data includes real-time power output of each distributed photovoltaic power station in the target area, regional total joint power, inverter operation state, and other data, and the two types of data are completely aligned in the time dimension, which can accurately match the corresponding relationship between meteorological conditions and power output. Next, based on the core purpose of separating the influence of meteorological variables and grid structure variables on joint power, two types of training scenario libraries are constructed. The first training scenario library focuses on the influence of meteorological factors on joint power. The input of each training scenario is a meteorological scenario, which is the meteorological time series data in a continuous time window extracted from historical meteorological data, such as a 1-hour time step, a 24-hour solar-temperature-wind speed sequence, and a weather type label. The output is the joint power time series, fluctuation characteristics, and spatial distribution characteristics of the target area in the corresponding time window extracted from the historical operation data, where the joint power time series is a 24-hour hourly total power value, the fluctuation characteristics include the maximum increase, decrease, fluctuation frequency, and fluctuation amplitude standard deviation within 24 hours, and the spatial distribution characteristics are the proportion distribution of each power station hourly power to the total power of the region. In this way, scenario mapping of meteorological input and multi-dimensional power output is achieved. The second training scenario library focuses on the influence of grid structure changes on joint power. To exclude the interference of meteorological variables, the input of each training scenario is set as grid structure change parameters and a fixed meteorological scenario. The grid structure change parameters are extracted from grid operation records, including power station access or exit status, transmission line impedance adjustment, transformer capacity change, etc. The fixed meteorological scenario is selected from the stable standard scenario in the historical meteorological data, such as a 7-day continuous sunny weather time series without fluctuation, to ensure uniform meteorological conditions. The output is also the joint power time series, fluctuation characteristics, and spatial distribution characteristics of the target area extracted from the historical operation data under the corresponding grid structure and fixed meteorological scenario, to achieve scenario mapping of grid structure change and fixed meteorological combined input and multi-dimensional power output. Finally, the two types of scenario libraries are constructed.

[0059] The determined physical constraint term is fused with the determined training loss function, at least one selected from the energy conservation constraint term, the spatial correlation constraint term and the wave limit constraint term is converted into a quantifiable calculation constraint loss, each constraint loss is assigned a corresponding weight coefficient, and the constraint loss is weighted and summed with the basic training loss function to form a comprehensive loss function that takes into account the prediction accuracy and the physical law. Then, for the first training scene library, the meteorological scene of each training scene in the first training scene library is taken as input, and the corresponding target region joint power time sequence, fluctuation characteristic and spatial distribution characteristic are taken as output label. The comprehensive loss function is used to iteratively train the spatio-temporal graph neural network model, the model parameters are continuously optimized through back propagation, the model gradually learns the mapping relationship between the meteorological factors and the joint power multi-dimensional characteristics, and finally the first twin channel focusing on fitting the influence law of meteorological conditions on joint power is obtained. At the same time, for the second training scene library, the grid structure change parameter and the fixed meteorological scene of each training scene in the second training scene library are taken as input, and the corresponding target region joint power time sequence, fluctuation characteristic and spatial distribution characteristic are taken as output label. The same structure of the comprehensive loss function is used to train another spatio-temporal graph neural network model, and the fitting ability of the model to the relationship between the grid structure change and the joint power response is optimized, so as to obtain the second twin channel focusing on capturing the influence law of the grid structure change on the joint power. Thus, the two twin channels have precise modeling capability for different influencing factors.

[0060] Based on the output characteristics of the first twin channel and the second twin channel, a channel fusion mechanism is designed. The mechanism adopts an attention weight distribution strategy to dynamically adjust the fusion ratio of the output results of the two channels by analyzing the input data type, such as enhancing the weight of the first channel when the scene is dominated by meteorological data, and enhancing the weight of the second channel when the scene is dominated by grid structure change. Then, a feature splicing layer and a collaborative optimization layer are set in the model architecture. The feature splicing layer splices the intermediate feature vectors of the joint power time sequence, fluctuation characteristic and spatial distribution characteristic output by the two channels according to the dimensions to form comprehensive features containing double influencing factors. The collaborative optimization layer performs nonlinear transformation on the spliced features through a multilayer perceptron to further eliminate the redundant information between the output of the two channels and strengthen the complementarity, so that the fusion result not only retains the power dynamic response characteristics under the driving of meteorological conditions, but also embodies the power distribution adjustment law brought by the grid structure change. Finally, the collaborative optimized features are decoded by the output layer to generate joint power time sequence, fluctuation characteristic and spatial distribution characteristic that meet the dual influence of meteorological conditions and grid structure, and finally form a complete joint power characteristic twin model. The model can adaptively call the advantages of the two channels according to the actual input scene, and realize the precise simulation and prediction of the joint power characteristics under complex working conditions.

[0061] In one possible implementation manner, the step S400 further includes: Step S410: identifying the data category of the simulation input data, judging the triggered matching twin channel.

[0062] Step S420: inputting the simulation input data into the matching twin channel for scene deduction, outputting the joint power time sequence, fluctuation characteristics and spatial distribution characteristics of the target area.

[0063] Specifically, the input simulation input data is analyzed for features, core parameter types contained therein are extracted, and the scene attributes reflected by the data are determined. If the simulation input data only contains meteorological related parameters, such as future period sunshine intensity prediction sequence, temperature variation trend, wind speed and duration distribution, cloud cover dynamic, etc., and does not involve any power grid structure change information, the data category thereof is identified as meteorology driven data, and it is determined that the first twin channel focusing on the mapping relationship between meteorological factors and joint power needs to be triggered. If the simulation input data contains power grid structure change parameters, such as new power station access scheme, line impedance adjustment value, transformer capacity change parameter, part of power station exit operation state identifier, etc., and is accompanied by fixed meteorological scene parameters, such as setting to historical same period typical sunny weather or standardized meteorological condition, the data category thereof is identified as power grid structure driven data, and it is determined that the second twin channel focusing on the response relationship between power grid structure change and joint power needs to be triggered. If the simulation input data contains both meteorological parameters and power grid structure change parameters, the scene priority rules of the model are preset, such as according to application requirements, the meteorological parameters are given priority to trigger the first twin channel, and the power grid structure parameters are used as auxiliary correction terms; or in the power grid reconstruction scene, the second twin channel is triggered preferentially, and the meteorological parameters remain fixed reference value, and finally the unique matching twin channel is determined to ensure that the subsequent scene deduction can call the most suitable model channel for calculation.

[0064] According to the determined matching twin channel, the simulated input data is preprocessed, if the first twin channel focusing on meteorological-power mapping is matched, the simulated meteorological data such as future period sunshine, temperature time series is time series aligned and numerically standardized according to the format standard in the channel training to ensure that the data dimension, time step and channel input requirements are consistent, if the second twin channel focusing on power grid structure-power mapping is matched, the simulated power grid structure change parameters such as power station access state, line impedance value and fixed meteorological scene parameters are integrated in format, and the power grid topology related features are supplemented to adapt to the channel input logic. Then, the preprocessed simulated input data is input into the matching twin channel, and the spatio-temporal graph neural network model in the channel will call the mapping rule learned in the training stage, i.e. the power response rule under meteorological driving or the power adjustment rule under power grid structure change to perform scene deduction: in the deduction process, the model will combine the built-in physical constraint term to check the intermediate output result in real time, for example, through the energy conservation constraint term to correct the deviation of regional total power and irradiance energy, through the fluctuation limit constraint term to limit the power change rate exceeding the inverter performance, to ensure that the deduction process conforms to the physical law. Finally, after decoding by the model output layer, the joint power time series of the target region is generated, such as the future 24-hour hourly total power change curve, including peak power, valley power and occurrence time, fluctuation characteristics such as maximum power fluctuation amplitude, fluctuation frequency statistics and over-limit risk assessment of adjacent time steps, and spatial distribution characteristics such as power proportion of each distributed photovoltaic power station at different time periods, spatial clustering of high-power output stations, and complete output of scene deduction results to support power grid dispatching, operation and maintenance decision-making and other application requirements.

[0065] In one possible implementation manner, the step S300 further includes: Step S340: generating a plurality of groups of regional joint power simulation data under different historical meteorological conditions by using the joint power characteristic twin model.

[0066] Step S350: analyzing the plurality of groups of regional joint power simulation data by using unsupervised time series clustering to automatically identify power curve categories with different morphological characteristics.

[0067] Step S360: for each identified power curve category, calculating a center trajectory as a representative power curve, and statistically calculating a fluctuation range, an occurrence probability and main associated meteorological characteristics to establish each typical scene.

[0068] Step S370: storing each typical scene and mapping it with historical power dispatching decisions to establish a typical scene library for random optimization dispatching.

[0069] Specifically, diverse historical weather conditions are screened from the collected historical weather data, which need to cover typical weather types such as sunny, overcast, cloudy, and rainfall in the target area, and also include special weather conditions such as dust and haze, to ensure coverage of weather scenarios with different radiation intensity, temperature, wind speed, and cloud cover characteristics. Then, according to the input format requirements of the joint power characteristic twin model, each type of historical weather condition is preprocessed, the time step of the weather data is uniformly arranged, the time sequence dimensions of radiation intensity, temperature, wind speed, and other parameters are kept consistent with the input standard during model training, and the data values are standardized to ensure that the parameter range adapts to the model calculation logic. Then, the preprocessed historical weather conditions are input into the trained joint power characteristic twin model, which calls the mapping rules between weather factors and joint power learned in the training phase, and combines the built-in physical constraints to verify the deduction process, ensuring that the output results comply with physical laws such as energy conservation and power fluctuation limits. Finally, for each type of historical weather condition, multiple sets of regional joint power simulation data are generated through multiple model deductions, each set of data containing complete power time series information, and ultimately forming a regional joint power simulation data set covering different historical weather conditions, sufficient in quantity, and reflecting the power variation characteristics under various weather scenarios.

[0070] The generated multiple sets of regional joint power simulation data are preprocessed, the power time series corresponding to each set of data are converted to standardized vector format, the time sequence length and time step of all data are unified, and the analysis interference caused by differences in time length and different numerical magnitudes between different data is eliminated, ensuring that all power curves to be analyzed are in the same data dimension. Then, an unsupervised clustering algorithm suitable for time series data characteristics is selected, and the morphological characteristics of the power curve are used as the core clustering basis. By calculating the similarity between different power time series, including the distance measurement of key features such as the dynamic change trend of the power curve, the time position of the peak and valley, and the rate of power rise and fall, power curves with similar morphological characteristics are classified into the same cluster. The entire clustering process does not require manual pre-labeling of class labels, and the algorithm will automatically determine the reasonable number of clusters based on the morphological rules inherent in the data, such as clearly distinguishing power curves with obvious morphological differences such as "morning power rapidly rising, noon reaching high value and maintaining stable", "overall low power and small fluctuations throughout the day", and "power rising and falling sharply with changing cloud cover" into different categories. Finally, the clustering results are verified for reasonableness by using clustering effectiveness evaluation indicators, ensuring that each power curve category has high morphological consistency within the category and significant morphological differences between categories, and finally automatically identifying power curve categories with different morphological characteristics.

[0071] For each power curve category identified automatically, the power time sequence corresponding to all regional joint power simulation data in this category is extracted. Based on time step, the mean of all power values at each time node is calculated. These means are connected in time order to form the central trajectory of this category, which is determined as the representative power curve of this power curve category, intuitively reflecting the core trend of the power change in this category. Then, the fluctuation range of the power curve in this category is calculated. By calculating the deviation of all power values at each time node from the corresponding mean, the power fluctuation interval of each time node is determined. The fluctuation interval of all time nodes is integrated to obtain the power fluctuation range in the entire time sequence period, clearly presenting the stability characteristics of the power output in this category. Subsequently, the occurrence probability of this power curve category is calculated. By counting the proportion of the number of power simulation data groups in this category to the total number of regional joint power simulation data, the occurrence probability of this category under historical weather conditions is obtained, reflecting its frequency of occurrence. At the same time, the corresponding historical weather conditions when generating the power simulation data in this category are traced back, and the weather features highly associated with the power features in this category are analyzed and refined, such as high irradiance intensity and low wind speed often associated with "midday high power smooth output" power curve category, and cloudy and intermittent irradiance often associated with "multi-period power fluctuation" power curve category. Finally, the representative power curve, fluctuation range, occurrence probability and main associated weather features of each power curve category are integrated to form a typical scene containing the core information of this category, completing the construction of each typical scene.

[0072] The constructed each typical scene is processed in a standardized format. The core information of the typical scene, such as representative power curve, fluctuation range, occurrence probability and main associated meteorological characteristics, is arranged according to a preset data structure, such as a structured data table or an extensible markup language format, to ensure that the classification of various types of information is clear and the fields are uniform, facilitating subsequent storage and calling. Then, all the processed typical scenes are stored in batches in a designated database. A hierarchical storage strategy is adopted to classify and archive the typical scenes according to their main associated meteorological characteristics, such as sunny, cloudy, cloudy fluctuation or occurrence probability level, to improve the efficiency of scene retrieval. Subsequently, the historical power dispatching decision database is called to extract historical dispatching cases matching the features of each typical scene. These cases need to include power scene description, dispatching strategy, such as energy storage charging and discharging adjustment, conventional power output optimization, grid-connected line power distribution, and dispatching effect evaluation data, in the corresponding period. By comparing the power characteristics and meteorological characteristics of the typical scenes with the scene conditions of the historical dispatching cases, a mapping relationship between the two is established. For example, the typical scene of “high power smooth output and high occurrence probability” is associated with the historical dispatching decision of “preferably full generation on grid and reduce energy storage investment”, and the typical scene of “frequent power fluctuation and large fluctuation range” is associated with the historical dispatching decision of “strengthening energy storage frequency modulation and dynamically adjusting conventional power reserve capacity”. Finally, all the typical scenes and their matched historical power dispatching decisions are integrated to form a complete data set containing scene information, mapping decisions and decision effects, and a typical scene library for random optimization dispatching is constructed. The scene library can provide historical experience reference and data support for the formulation of dispatching strategies in different power scenarios in subsequent random optimization dispatching of the power grid.

[0073] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. Moreover, the above describes specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0074] The above only describes the preferred embodiments of the present application and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0075] The present application is only an exemplary description of the present application, and should be considered to cover any and all modifications, changes, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A high proportion distributed photovoltaic power station combined power characteristic analysis method, characterized in that, The method comprises the following steps: Collecting historical operation data, historical meteorological data, and power grid topology data of each distributed photovoltaic power station in a target area; Based on the power grid topology data, the spatial-temporal graph structure is constructed with each distributed photovoltaic power station as a node and the spatial relationship between the stations as an edge; Based on the spatial-temporal graph structure, a spatial-temporal graph neural network model is constructed, and a physical constraint term is introduced to train the model using the historical operation data and the historical meteorological data, thereby obtaining a joint power characteristic twin model; The simulation input data is input into the joint power characteristic twin model to perform joint power scenario deduction, and the joint power time sequence, fluctuation characteristics, and spatial distribution characteristics of the target area are output.

2. The method for analyzing the combined power characteristics of a high proportion distributed photovoltaic power station according to claim 1, characterized in that, The spatial relationship includes electrical distance, geographical distance, and meteorological correlation.

3. The method of claim 2, wherein the high proportion distributed photovoltaic power station combined power characteristic analysis method is characterized in that, Based on the power grid topology data, the spatial-temporal graph structure is constructed with each distributed photovoltaic power station as a node and the spatial relationship between the stations as an edge, which comprises: A feature vector is constructed for each distributed photovoltaic power station, and each feature vector at least includes the installed capacity, historical average unit capacity power, and geographical position coordinates of the corresponding power station; Based on the power grid topology data, the feature vectors are stacked to form an initial feature matrix of the nodes with each distributed photovoltaic power station as a node; Based on the spatial relationship, the correlation strength between each pair of nodes is calculated, the weight of the edge is defined, a weighted linear combination is performed, a weighted adjacency matrix representing the spatial relationship between the nodes is generated, and the spatial-temporal graph structure is established.

4. The method of claim 3, wherein the high proportion distributed photovoltaic power station combined power characteristic analysis method is characterized in that, Based on the spatial relationship, the correlation strength between each pair of nodes is calculated, the weight of the edge is defined, and a weighted linear combination is performed, which comprises: The electrical distance weight between each pair of nodes is calculated, and the electrical distance weight is inversely proportional to the electrical coupling impedance modulus value of the two nodes in the power grid topology; The geographical distance weight between each pair of nodes is calculated, and the geographical distance weight is inversely proportional to the geographical distance between the two nodes; The meteorological correlation weight between each pair of nodes is calculated, and the meteorological correlation weight is the Pearson correlation coefficient of the historical power sequence of the two nodes; The electrical distance weight, the geographical distance weight, and the meteorological correlation weight are normalized and then summed to obtain the correlation strength between each pair of nodes; The correlation strength is used to define the weight of the edge between each pair of nodes to construct the weighted adjacency matrix.

5. The method of claim 4, wherein the high proportion distributed photovoltaic power station combined power characteristic analysis method is characterized in that, Based on the spatial relationship, the correlation strength between each pair of nodes is calculated, and the weight of the edge is defined, which further comprises: For any node, the K nearest nodes are selected as a candidate neighborhood set based on the geographical distance; The candidate neighborhood set is traversed, and nodes without direct electrical access or in different power supply areas are removed by querying the power grid topology data to form an effective neighborhood set; The electrical distance weight, the geographical distance weight, and the meteorological correlation weight between the any node and other nodes in the effective neighborhood set are calculated.

6. The method of claim 1, wherein the high proportion distributed photovoltaic power station combined power characteristic analysis method is characterized in that, The physical constraint term includes at least one of an energy conservation constraint term, a spatial correlation constraint term, and a fluctuation limit constraint term. The energy conservation constraint term is used to constrain the deviation range between the total irradiation energy of the region and the theoretical maximum power generation potential; the spatial correlation constraint term is used to constrain the inter-site power correlation of the model output to conform to the physical decay law; and the fluctuation limit constraint term is used to constrain the power variation rate of the model output to be less than the limit value based on the physical performance of the inverter.

7. The method of claim 6, wherein the method is characterized by, Based on the spatio-temporal graph structure, a spatio-temporal graph neural network model is constructed, and a physical constraint term is introduced to train the model by using the historical operation data and the historical meteorological data, so as to obtain a joint power characteristic twin model, including: determining a training loss function of the spatio-temporal graph neural network model; combining the physical constraint term with the training loss function, and training the model by using the historical operation data and the historical meteorological data, so as to obtain the joint power characteristic twin model.

8. The method of claim 7, wherein the high proportion distributed photovoltaic power station combined power characteristic analysis method is characterized in that, Training the model by using the historical operation data and the historical meteorological data includes: constructing a first training scene library and a second training scene library by using the historical meteorological data and the historical operation data; wherein the input of each training scene in the first training scene library is a meteorological scene, and the output is the joint power time sequence, fluctuation characteristic and spatial distribution characteristic of the target region; the input of each training scene in the second training scene library is a power grid structure change parameter and a fixed meteorological scene, and the output is the joint power time sequence, fluctuation characteristic and spatial distribution characteristic of the target region; combining the physical constraint term with the training loss function, training the spatio-temporal graph neural network model by using the first training scene library to obtain a first twin channel, and training the spatio-temporal graph neural network model by using the second training scene library to obtain a second twin channel; connecting the first twin channel and the second twin channel to generate the joint power characteristic twin model.

9. The method of claim 8, wherein the high proportion distributed photovoltaic power station combined power characteristic analysis method is characterized in that, inputting simulation input data into the joint power characteristic twin model to perform scene deduction of joint power, and outputting the joint power time sequence, fluctuation characteristic and spatial distribution characteristic of the target region, including: identifying the data category of the simulation input data to determine the matched twin channel triggered; inputting the simulation input data into the matched twin channel to perform scene deduction, and outputting the joint power time sequence, fluctuation characteristic and spatial distribution characteristic of the target region.

10. The method of claim 1, wherein the high proportion distributed photovoltaic power station combined power characteristic analysis method is characterized in that, After obtaining the joint power characteristic twin model, further including: generating multiple groups of regional joint power simulation data under different historical meteorological conditions by using the joint power characteristic twin model; performing analysis on the multiple groups of regional joint power simulation data by using unsupervised time series clustering, and automatically identifying power curve categories with different morphological characteristics; for each identified power curve category, calculating a center trajectory as a representative power curve, and statistically calculating a fluctuation range, occurrence probability and main associated meteorological features to establish each typical scene; storing each typical scene, and mapping the historical power dispatching decisions to establish a typical scene library for random optimization dispatching.

Images