Joint optimization clustering distributed photovoltaic power prediction method and system

Abstract

The invention discloses a joint optimization clustering distributed photovoltaic power prediction method and system, and relates to the related field of distributed energy management, and the method comprises the steps: carrying out the local calling of data, carrying out the time sequence alignment, and constructing a multi-dimensional feature association number table; extracting geographic position information and historical output data to perform similarity analysis, performing spectral clustering, and outputting a photovoltaic cluster; collecting meteorological time series data according to the power prediction scale, inputting the meteorological time series data into the cluster power prediction model, and performing cluster power prediction; grid-connected space-time joint optimization is carried out according to the cluster-level power prediction value, and an energy storage charging and discharging scheduling strategy is output; after single-station fitting conversion is carried out and a distributed power prediction value is output, output deviation identification is carried out, and a fault node is positioned; and dynamically updating the energy storage charging and discharging scheduling strategy. The technical problems that existing power prediction is not high in precision, cross-cluster power optimization scheduling is difficult to achieve, and the power grid operation stability is poor are solved, and the technical effect of improving prediction accuracy and the power grid operation stability is achieved.

Description

Technical Field

[0001] The present application relates to the field of distributed energy management, and in particular to a jointly optimized clustered distributed photovoltaic power prediction method and system. Background Art

[0002] Accurately forecasting distributed photovoltaic power is crucial for the stable operation of the power grid and the efficient use of energy. Currently, the main approaches to addressing distributed photovoltaic power forecasting are modeling and forecasting based on the historical data of a single photovoltaic power station or unit, or using simple clustering methods to group photovoltaic power stations for forecasting. However, these methods, when faced with the reality of small or incomplete data sets for distributed photovoltaic power stations, fail to fully consider the geographical differences between photovoltaic power stations, the time-series alignment of historical output data, and the power complementarity between clusters. This results in low forecast accuracy and makes it difficult to achieve cross-cluster power optimization scheduling.

[0003] Among the current related technologies, distributed photovoltaic power prediction has the technical problem of low prediction accuracy, difficulty in achieving cross-cluster power optimization scheduling, and poor grid operation stability. This problem is particularly prominent in distributed photovoltaic power station scenarios with small or incomplete data. Summary of the Invention

[0004] The present application provides a jointly optimized clustered distributed photovoltaic power prediction method and system, adopts local call and time series alignment preprocessing of multi-source data, constructs a multi-dimensional feature association table, extracts information from the table and analyzes similarity, clusters the photovoltaic power generation unit spectrum to obtain K photovoltaic clusters, inputs meteorological time series data into the cluster power prediction model, and obtains K cluster-level power prediction values ​​through cluster power prediction. The grid-connected spatiotemporal joint optimization is performed based on the cluster-level power prediction values, and the energy storage charging and discharging scheduling strategy is output to achieve cross-cluster power complementarity. The distributed power prediction value is obtained by fitting and converting the cluster-level power prediction value to a single station, the output deviation is identified to locate the fault node, and the energy storage charging and discharging scheduling strategy is dynamically updated. The technical effect of improving the accuracy of photovoltaic power prediction, realizing cross-cluster power optimization scheduling, and thus improving the stability of power grid operation is achieved.

[0005] The present application provides a joint optimization clustering distributed photovoltaic power prediction method, comprising: locally calling multi-source data of a distributed photovoltaic power station, and constructing a multidimensional feature association number table by performing time series alignment preprocessing on the call results; extracting multiple geographical location information and multiple historical output data of multiple photovoltaic power generation units in the distributed photovoltaic power station from the multidimensional feature association number table for similarity analysis, and performing power generation unit spectrum clustering on the multiple photovoltaic power generation units based on the analysis results, outputting K photovoltaic clusters, where K is a natural number greater than or equal to 1; and collecting meteorological data obtained according to the power prediction scale. The time series data is input into the K cluster power prediction models of the K photovoltaic clusters to perform cluster power prediction and output K cluster-level power prediction values. Based on the K cluster-level power prediction values, a spatiotemporal grid-connected joint optimization is performed to output an energy storage charge and discharge scheduling strategy, and cross-cluster power complementarity of the K photovoltaic clusters is performed. After single-station fitting conversion is performed based on the K cluster-level power prediction values ​​and distributed power prediction values ​​are output, output deviation is identified based on the distributed power prediction values ​​to locate the fault node of the photovoltaic power station. The energy storage charge and discharge scheduling strategy is dynamically updated based on the fault node of the photovoltaic power station.

[0006] In a possible implementation, multiple geographic location information and multiple historical output data of multiple photovoltaic power generation units in the distributed photovoltaic power station are extracted from the multidimensional feature association number table for similarity analysis, and power generation unit spectrum clustering is performed on the multiple photovoltaic power generation units based on the analysis results, and K photovoltaic clusters are output, including: drawing multiple unit output curves based on the multiple historical output data; analyzing the multiple unit output curves through a dynamic time warping algorithm, and outputting an output curve similarity matrix of the multiple photovoltaic power generation units; constructing a spatial correlation matrix based on the multiple geographic location information; and performing power generation unit spectrum clustering by fusing and analyzing the output curve similarity matrix and the spatial correlation matrix, and clustering the multiple photovoltaic power generation units into the K photovoltaic clusters.

[0007] In a possible implementation, a spatiotemporal grid-connected optimization is performed based on the K cluster-level power prediction values, an energy storage charging and discharging scheduling strategy is output, and cross-cluster power complementarity of the K photovoltaic clusters is performed, including: dynamically loading K real-time grid-connected processing thresholds based on the K cluster-level power prediction values; obtaining K cluster-level power measured values ​​of the K photovoltaic clusters by output power monitoring and summing; performing spatiotemporal coupling optimization based on the deviation between the K cluster-level power measured values ​​and the K real-time grid-connected processing thresholds, and outputting the energy storage charging and discharging scheduling strategy; and performing cross-cluster power dynamic complementarity of the K photovoltaic clusters based on the energy storage charging and discharging scheduling strategy.

[0008] In a possible implementation, the power generation unit spectrum clustering is performed by fusing and analyzing the output curve similarity matrix and the spatial correlation matrix, and the multiple photovoltaic power generation units are clustered into the K photovoltaic clusters, including: performing exponential decay mapping on the output curve similarity matrix to generate a first normalized similarity matrix; normalizing the spatial correlation matrix according to the matrix scale of the first normalized similarity matrix to obtain a second normalized similarity matrix; superimposing the first normalized similarity matrix and the second normalized similarity matrix based on a preset matrix weight coefficient to obtain a fused similarity matrix; constructing a normalized Laplace matrix based on the fused similarity matrix; performing eigenvalue decomposition on the normalized Laplace matrix to select K low-dimensional eigenvectors corresponding to K minimum eigenvalues; after constructing a low-dimensional feature space based on the K low-dimensional eigenvectors, the multiple photovoltaic power generation units are mapped to the low-dimensional feature space; and outputting the K photovoltaic clusters by iteratively updating the cluster center points in the low-dimensional feature space.

[0009] In a possible implementation, the K photovoltaic clusters are output by iteratively updating the cluster center points in the low-dimensional feature space, including: presetting a center point spacing scale, and initializing K cluster center points in the low-dimensional feature space based on the center point spacing scale; allocating the multiple photovoltaic power generation units in the low-dimensional feature space to the K cluster center points based on the closest Euclidean distance to obtain K groups of power generation unit allocation results; performing random perturbation iterations on the K cluster center points to obtain K first iteration center points; allocating the multiple photovoltaic power generation units in the low-dimensional feature space to the K first iteration center points based on the closest Euclidean distance to obtain K groups of first iteration allocation results; comparing the sum of squares of the distances between the K groups of power generation unit allocation results and the K groups of first iteration allocation results, and performing iterative retention judgment on the K cluster center points and the K first iteration center points based on the comparison result; and outputting the K photovoltaic clusters when the iterative fluctuation of the sum of squares of the distances is less than a preset deviation.

[0010] In a possible implementation, spatiotemporal coupling optimization is performed based on the deviations between the K cluster-level power measured values ​​and the K real-time grid-connected processing thresholds to output the energy storage charging and discharging scheduling strategy, including: dividing the K photovoltaic clusters into W excess output clusters and F insufficient output clusters based on the deviation directions between the K cluster-level power measured values ​​and the K real-time grid-connected processing thresholds, where W is a natural number greater than or equal to 1 and F is a natural number greater than or equal to 1; performing local energy storage fitting on the W excess output clusters to output W excess output values; after extracting F insufficient output values ​​of the F insufficient output clusters, processing is performed based on the W excess output values ​​and the F insufficient output values ​​to obtain multiple initial charging and discharging scheduling strategies; and evaluating the multiple initial charging and discharging scheduling strategies with the goal of minimizing energy storage loss and transmission cost to screen out the energy storage charging and discharging scheduling strategy.

[0011] In a possible implementation, the meteorological time series data collected according to the power prediction scale is input into the K cluster power prediction models of the K photovoltaic clusters to perform cluster power prediction. Before outputting K cluster-level power prediction values, the method further includes: using LSTM to construct a standard power prediction model, wherein the input data of the power prediction scale are meteorological time series data and output time series data that conform to the power prediction scale, and the output result is predicted power time series data that conforms to the power prediction scale with a preset proportional factor; performing time series alignment data call based on the multiple historical output data to obtain multiple meteorological data records and multiple power generation power records; and The standard power prediction model is localized using multiple historical output data, multiple meteorological data records, and multiple power generation records as training data to obtain multiple initial power prediction sub-models of the multiple photovoltaic power generation units; model parameters of the multiple initial power prediction sub-models are federally aggregated based on the K photovoltaic clusters to obtain K federal power prediction sub-models; the K federal power prediction sub-models are replicated according to the number of clusters of the K photovoltaic clusters to obtain K groups of federal power prediction sub-models; and the K groups of federal power prediction sub-models are connected in parallel within the group and a power summation engine is configured at the output end to complete the construction of the K cluster power prediction models.

[0012] In a possible implementation, a spatial correlation matrix is ​​constructed based on the multiple geographic location information, including: enumerating the multiple photovoltaic power generation units through a combination algorithm to obtain H groups of photovoltaic power generation units, where H is a natural number greater than or equal to 1; combining the multiple geographic location information based on the H groups of photovoltaic power generation units to obtain H groups of geographic location information; wherein the geographic location information includes the latitude and longitude of the photovoltaic unit, the altitude of the photovoltaic unit, the inclination of the photovoltaic panel and the azimuth of the photovoltaic panel; after processing the H groups of geographic location information based on preset geographic normalization weights, performing Euclidean distance calculation on the processing results between groups, and outputting H spatial correlation coefficients; constructing a benchmark correlation matrix based on the multiple photovoltaic power generation units; using the H groups of photovoltaic power generation units as a filling guide, filling the H spatial correlation coefficients into the benchmark correlation matrix to complete the construction of the spatial correlation matrix.

[0013] In a possible implementation, a multi-source data local call is performed on the distributed photovoltaic power station, and the call results are pre-processed by time alignment to construct a multi-dimensional feature association table, including: performing local data calls through multiple edge acquisition terminals deployed in the multiple photovoltaic power generation units to obtain multiple meteorological data records, multiple output data records and the multiple geographic location information; presetting an output sliding window, and performing sliding mean calculation on the multiple output data records according to the output sliding window to obtain multiple downsampled output data; presetting a meteorological interpolation granularity, and using the meteorological interpolation granularity to perform linear interpolation processing on the multiple meteorological data records to obtain multiple upsampled meteorological data; after time alignment of the multiple downsampled output data and the multiple upsampled meteorological data, the multiple downsampled output data, the multiple upsampled meteorological data and the multiple geographic location information are stored based on the association of the multiple photovoltaic power generation units to complete the construction of the multi-dimensional feature association table.

[0014] The present application also provides a jointly optimized clustered distributed photovoltaic power prediction system, comprising: a time series alignment module for locally calling multi-source data of a distributed photovoltaic power station, and performing time series alignment preprocessing on the call results to construct a multi-dimensional feature association number table; a power generation unit spectrum clustering module for extracting multiple geographical location information and multiple historical output data of multiple photovoltaic power generation units in the distributed photovoltaic power station from the multi-dimensional feature association number table for similarity analysis, and performing power generation unit spectrum clustering on the multiple photovoltaic power generation units according to the analysis results, and outputting K photovoltaic clusters, where K is a natural number greater than or equal to 1; a cluster power prediction module for converting the meteorological time series data collected according to the power prediction scale into a clustered power generation unit spectrum. K cluster power prediction models are input into the K photovoltaic clusters to perform cluster power prediction and output K cluster-level power prediction values; an energy storage charge and discharge scheduling strategy output module is used to perform grid-connected spatiotemporal joint optimization based on the K cluster-level power prediction values, output an energy storage charge and discharge scheduling strategy, and perform cross-cluster power complementarity of the K photovoltaic clusters; an output deviation identification module is used to perform single-station fitting conversion based on the K cluster-level power prediction values, output a distributed power prediction value, and then perform output deviation identification based on the distributed power prediction value to locate the fault node of the photovoltaic power station; an energy storage charge and discharge scheduling strategy update module is used to dynamically update the energy storage charge and discharge scheduling strategy based on the fault node of the photovoltaic power station.

[0015] The present application proposes a joint optimization clustering distributed photovoltaic power prediction method and system. First, a multi-source data local call is made to the distributed photovoltaic power station, and the call results are pre-processed by time series alignment to construct a multi-dimensional feature association table. Then, multiple geographic location information and multiple historical output data of multiple photovoltaic power generation units in the distributed photovoltaic power station are extracted from the multi-dimensional feature association table for similarity analysis. Based on the analysis results, the multiple photovoltaic power generation units are clustered into power generation unit spectra, and K photovoltaic clusters are output. Then, the meteorological time series data obtained according to the power prediction scale are collected. The system inputs K cluster power prediction models for the K photovoltaic clusters, performs cluster power prediction, and outputs K cluster-level power prediction values. Based on these K cluster-level power prediction values, it performs spatiotemporal grid-connection optimization, outputs an energy storage charge-discharge scheduling strategy, performs cross-cluster power complementarity for the K photovoltaic clusters, and then performs single-station fitting conversion based on the K cluster-level power prediction values. After outputting distributed power prediction values, it identifies output deviations based on these distributed power prediction values, locates photovoltaic power station fault nodes, and finally dynamically updates the energy storage charge-discharge scheduling strategy based on the photovoltaic power station fault nodes. This achieves the technical effect of improving the accuracy of photovoltaic power prediction, achieving cross-cluster power optimization scheduling, and thus improving the stability of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] To more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings of the embodiments of the present application will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in precise order. On the contrary, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0017] Figure 1 A schematic flow chart of a jointly optimized clustered distributed photovoltaic power prediction method provided in an embodiment of the present application.

[0018] Figure 2 A schematic structural diagram of a jointly optimized clustered distributed photovoltaic power prediction system provided in an embodiment of the present application.

[0019] Description of the accompanying drawings: timing alignment module 10, power generation unit spectrum clustering module 20, cluster power prediction module 30, energy storage charging and discharging scheduling strategy output module 40, output deviation identification module 50, energy storage charging and discharging scheduling strategy update module 60. DETAILED DESCRIPTION

[0020] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0021] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0022] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict, and the terms “first\second” involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. The terms “including” and “having” and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application only.

[0023] The embodiment of the present application provides a joint optimization clustering distributed photovoltaic power prediction method, such as Figure 1 As shown, the method includes: Step S100 : Locally call multi-source data of the distributed photovoltaic power station, and perform time sequence alignment preprocessing on the call results to construct a multi-dimensional feature association table.

[0024] Specifically, various data types are acquired from the local systems of distributed photovoltaic power plants through API interfaces, database queries, or file reading. These include meteorological data (irradiance, temperature, wind speed, etc.), equipment operation data (inverter status, solar panel temperature, etc.), and historical output data. Because data from different data sources have different timestamps or sampling frequencies, time alignment algorithms (such as linear interpolation and time window aggregation) are used to align the data to the same time scale to ensure consistency and comparability in subsequent analysis. The aligned data is then integrated into a multidimensional data table (a multidimensional feature association table), where each row represents data at a specific time point and each column represents a feature (such as irradiance, temperature, output, etc.). This multidimensional feature association table serves as the basis for subsequent feature extraction and similarity analysis. A distributed photovoltaic power plant refers to a power system consisting of multiple photovoltaic power generation units distributed across different locations and connected to the power grid. Multi-source data refers to data from different data sources.

[0025] In one possible implementation, a multi-source data local call is performed on a distributed photovoltaic power station, and the call results are pre-processed by time alignment to construct a multi-dimensional feature association table. Step S100 further includes step S110, in which local data calls are performed through multiple edge collection terminals deployed in multiple photovoltaic power generation units to obtain multiple meteorological data records, multiple output data records and the multiple geographic location information.

[0026] Specifically, edge data collection terminals (devices deployed on-site at each photovoltaic power generation unit for data collection and preliminary processing) are deployed in each photovoltaic power generation unit. These terminals are responsible for real-time or periodic data collection. The edge data collection terminals retrieve data from local storage or sensors, including meteorological data (data describing atmospheric conditions, which can have a direct impact on photovoltaic power generation output, such as irradiance, temperature, wind speed, etc.), output data (actual output power data of the photovoltaic power generation unit, such as inverter output power), and geographic location information (information describing the location of the photovoltaic power generation unit, including latitude and longitude, altitude, photovoltaic panel tilt angle, azimuth, etc.), and store the retrieved data in the form of records.

[0027] Step S120 , presetting an output sliding window, and performing sliding mean calculation on a plurality of output data records according to the output sliding window to obtain a plurality of downsampled output data.

[0028] Specifically, according to the data analysis needs or the requirements of the prediction model, a time window size is set, such as 1 hour, 30 minutes, etc., for calculating the sliding mean. Within the set time window, the output data is averaged to obtain the downsampled output data. This method is used to smooth data fluctuations and reduce the impact of noise. The number of data points in the downsampled output data is reduced, but it can better reflect the overall trend, wherein the downsampling refers to the process of reducing the sampling frequency or the number of data points of the data. In this application, the original high-frequency output data records are converted into lower-frequency downsampled output data by performing a sliding mean calculation on the output data. The purpose of selecting downsampled meteorological data is, first, to smooth data fluctuations, reduce the impact of noise on the data, and make the data better reflect the overall trend; second, to reduce the amount of data, reduce the computational burden of subsequent processing and analysis, and improve efficiency.

[0029] Step S130 : Preset a meteorological interpolation granularity, and use the meteorological interpolation granularity to perform linear interpolation processing on multiple meteorological data records to obtain multiple upsampled meteorological data.

[0030] Among them, upsampling refers to the process of increasing the sampling frequency or number of data points of the data. In this application, the meteorological data records are linearly interpolated using meteorological interpolation granularity to obtain meteorological data at more time points, namely upsampled meteorological data. The purpose of choosing upsampled meteorological data is to improve the temporal resolution of the data, so that the meteorological data and the output data after downsampling can be better matched and aligned on the time scale, providing a more accurate and consistent data basis for subsequent analysis and modeling.

[0031] Specifically, based on the accuracy requirements of the data analysis or forecasting model, a time interval, such as 5 minutes or 10 minutes, is set for linear interpolation. Within the set time interval, the meteorological data is linearly interpolated, i.e., a linear estimate is made between known data points to obtain the upsampled meteorological data. The increased number of data points in the upsampled meteorological data improves the temporal resolution of the data. Linear interpolation is a simple and effective data interpolation method that can estimate unknown data points based on known data points.

[0032] In step S140 , after time-series alignment of the multiple downsampled output data and the multiple upsampled meteorological data, the multiple downsampled output data, the multiple upsampled meteorological data, and the multiple geographic location information are stored based on the association of the multiple photovoltaic power generation units to complete the construction of the multidimensional feature association table.

[0033] Specifically, the downsampled output data and upsampled meteorological data are aligned according to timestamps to ensure that the data at each time point corresponds. The aligned data and geographic location information are then associated and stored by photovoltaic power generation unit to form a multidimensional feature association table. This multidimensional feature association table serves as the basis for subsequent feature extraction, similarity analysis, and power prediction.

[0034] For example, if the output sliding window is set to 30 minutes, the average output for each 30-minute period can be calculated and distributed to each minute within that 30-minute period. If the original meteorological data is recorded at a 15-minute interval, linear interpolation can be used to interpolate the meteorological data to each minute. This process aligns the downsampled output data and upsampled meteorological data on a minute-by-minute basis, providing an accurate and consistent data foundation for subsequent analysis and modeling. This implementation uses edge acquisition terminals to access local data and perform sliding mean calculation, linear interpolation, and time series alignment to construct a multidimensional feature association table containing multiple features. This table provides an accurate and consistent data foundation for subsequent feature extraction, similarity analysis, and power forecasting, helping to improve forecast accuracy and grid connection stability.

[0035] In step S200, multiple geographic location information and multiple historical output data of multiple photovoltaic power generation units in a distributed photovoltaic power station are extracted from the multidimensional feature association number table for similarity analysis, and power generation unit spectral clustering is performed on the multiple photovoltaic power generation units based on the analysis results to output K photovoltaic clusters, where K is a natural number greater than or equal to 1.

[0036] Specifically, geographic location information associated with each PV power generation unit, such as longitude, latitude, and altitude, is extracted from the multidimensional feature association table for subsequent similarity analysis or clustering. Output data for each PV power generation unit at different time points is extracted from the multidimensional feature association table to form a historical output sequence. This data forms the basis for building a power prediction model. The geographic location information and historical output data are integrated, and similarity metrics (such as Euclidean distance and cosine similarity) are used to calculate the similarity between different PV power generation units. Based on the results of the similarity analysis, a similarity matrix is ​​constructed, and the PV power generation units are divided into K clusters using a spectral clustering algorithm (such as K-means spectral clustering or Normalized Cuts). Spectral clustering is a graph-theory-based clustering algorithm that clusters data by constructing a similarity matrix and performing spectral decomposition on the matrix. Spectral clustering algorithms can handle non-spherical data distributions and are robust to noise.

[0037] In one possible implementation, multiple geographic location information and multiple historical output data of multiple photovoltaic power generation units in a distributed photovoltaic power station are extracted from a multidimensional feature association number table for similarity analysis, and power generation unit spectrum clustering is performed on the multiple photovoltaic power generation units based on the analysis results to output K photovoltaic clusters. Step S200 further includes step S210, drawing multiple unit output curves based on the multiple historical output data.

[0038] Specifically, historical output data for each photovoltaic power generation unit (the actual output power data of the photovoltaic power generation unit over the past period) is extracted from the multidimensional feature association table. For each photovoltaic power generation unit, its historical output data is arranged chronologically. Using time series analysis tools or programming languages ​​(such as Python's matplotlib library), a time-to-output curve (unit output curve) is plotted for each unit, with time as the horizontal axis and output as the vertical axis. This curve reflects the change in the output power of the photovoltaic power generation unit over time.

[0039] In step S220 , a dynamic time warping algorithm is used to analyze and output a similarity matrix of output curves of multiple photovoltaic power generation units.

[0040] Specifically, the Dynamic Time Warping (DTW) algorithm is used to calculate the similarity between two time series. It can handle nonlinear changes in time series and uses dynamic programming to find the best matching path between the two time series to calculate their similarity. The DTW algorithm is used to calculate the similarity between each pair of photovoltaic power generation unit output curves. The calculated similarity values ​​are entered into a matrix to form a similarity matrix. The similarity matrix is ​​a two-dimensional array, where each element represents the similarity between the output curves of two photovoltaic power generation units.

[0041] Step S230: constructing a spatial correlation matrix based on the plurality of geographical location information.

[0042] Specifically, the spatial correlation between photovoltaic power generation units is calculated using their geographic location information (including latitude and longitude, altitude, photovoltaic panel inclination, and azimuth). While geographic location information reflects the spatial distribution characteristics of photovoltaic power generation units, spatial correlation describes the strength of the correlation between output fluctuations due to geographic proximity between different units. The impact of geographic proximity is numerically represented by constructing a spatial correlation matrix. The spatial correlation matrix is ​​a two-dimensional array in which each element represents the strength of the correlation between output fluctuations due to geographic proximity between two photovoltaic power generation units.

[0043] Step S240 , performing spectral clustering of power generation units by fusing and analyzing the output curve similarity matrix and the spatial correlation matrix, and clustering the multiple photovoltaic power generation units into K photovoltaic clusters.

[0044] Specifically, the output curve similarity matrix and the spatial correlation matrix are weighted and fused to form a comprehensive similarity matrix. The spectral clustering algorithm, a clustering algorithm based on graph theory, uses the comprehensive similarity matrix to construct a graph structure and performs clustering through spectral decomposition of the graph. The spectral clustering algorithm is used to cluster photovoltaic power generation units, resulting in K photovoltaic clusters. A photovoltaic cluster is a collection of photovoltaic power generation units grouped according to similarity. This implementation method accurately captures the differences and similarities in output characteristics between photovoltaic power generation units by plotting unit output curves and calculating the similarity matrix using the dynamic time warping algorithm. Furthermore, the spatial correlation matrix, which takes into account geographic location information, can reflect the strength of output fluctuations caused by geographic proximity. Fusion analysis of these two matrices and spectral clustering can produce a more reasonable and effective photovoltaic cluster partitioning result. This partitioning facilitates subsequent cluster-level power forecasting and the development of energy storage charging and discharging scheduling strategies, thereby improving the overall operational efficiency and grid-connected stability of distributed photovoltaic power plants.

[0045] In a possible implementation, a spatial correlation matrix is ​​constructed according to the plurality of geographic location information, and step S230 further includes step S231, enumerating a plurality of photovoltaic power generation units through a combination algorithm to obtain H groups of photovoltaic power generation units.

[0046] Specifically, the number of photovoltaic power generation units to be combined is determined (e.g., two units in a group), and a combination algorithm is used to enumerate all possible combinations of photovoltaic power generation units to form H groups of combinations, such as using a loop or recursive algorithm to enumerate all possible combinations, where H is a natural number greater than or equal to 1.

[0047] Step S232 : combining multiple pieces of geographical location information according to H groups of photovoltaic power generation units to obtain H groups of geographical location information.

[0048] The geographic location information includes the longitude and latitude of the photovoltaic unit, the altitude of the photovoltaic unit, the inclination angle of the photovoltaic panel, and the azimuth angle of the photovoltaic panel. Specifically, geographic location information corresponding to H groups of photovoltaic power generation units is extracted from a multidimensional feature association table or a geographic location information database. The extracted geographic location information is organized into H groups to form a set of H groups of geographic location information. The geographic location information includes information describing the spatial location of the photovoltaic power generation unit, such as the longitude and latitude of the photovoltaic unit, the altitude, the inclination angle of the photovoltaic panel, and the azimuth angle of the photovoltaic panel.

[0049] Step S233 : After processing H groups of geographic location information based on preset geographic normalization weights, perform Euclidean distance calculation between groups on the processing results, and output H spatial correlation coefficients.

[0050] Specifically, geographic location information includes multiple dimensions, such as longitude and latitude, altitude, photovoltaic panel inclination and azimuth, etc. These dimensions have different dimensions and ranges. In order to make them comparable when calculating spatial correlation, they need to be normalized. The preset geographic normalization weights are used to adjust the weights of different dimensions to reflect their importance in spatial correlation. According to the preset geographic normalization weights, H groups of geographic location information are normalized to eliminate the dimensional differences between data of different dimensions. The Euclidean distance formula is used to calculate the distance between each group of geographic location information as a measure of spatial correlation. The calculated distance value is converted into a correlation coefficient. The smaller the distance, the higher the correlation.

[0051] The example calculation is as follows: Assume that there are two groups of photovoltaic power generation units with geographical location information as shown in Table 1.

[0052] Table 1: Geographic location information

[0053] The default geographic normalization weights are: longitude: 0.3, latitude: 0.3, altitude: 0.2, photovoltaic panel tilt: 0.1, photovoltaic panel azimuth: 0.1. Scale the data of each dimension to the interval [0,1]. For example, the normalization formula for longitude is: ; Assuming that the minimum longitude in all data is 110.0° and the maximum longitude is 111.0°, then: Normalized longitude 1 = =0.2, normalized longitude 2= =0.5. The normalized data is weighted and summed according to the preset weights to obtain a comprehensive normalized value. For example: comprehensive normalized value 1 = 0.3×0.2+0.3×0.5+0.2×0.5+0.1×0.3+0.1×0.8=0.36, comprehensive normalized value 2 = 0.3×0.5+0.3×0.8+0.2×0.6+0.1×0.2+0.1×0.9=0.49. Use the comprehensive normalized value to calculate the Euclidean distance between two sets of geographic location information: Euclidean distance = =0.13. Convert the Euclidean distance into a correlation coefficient. The smaller the distance, the higher the correlation. For example, you can use the formula: , assuming the maximum Euclidean distance is 1.0, then: correlation coefficient = 1- =0.87. Finally, the spatial correlation coefficient after processing is 0.87, indicating that the two groups of photovoltaic power generation units have a high correlation in their geographical locations.

[0054] Step S234 : constructing a reference correlation matrix based on the plurality of photovoltaic power generation units.

[0055] Specifically, a baseline correlation matrix is ​​constructed based on the number of photovoltaic power generation units, with rows and columns corresponding to different photovoltaic power generation units. The baseline correlation matrix is ​​initialized to a zero matrix or initialized according to a specific rule. The baseline correlation matrix is ​​used to store the spatial correlation coefficients between photovoltaic power generation units and serves as a reference for subsequent filling.

[0056] In step S235 , H groups of photovoltaic power generation units are used as filling guides to fill H spatial correlation coefficients into the reference correlation matrix, thereby completing the construction of the spatial correlation matrix.

[0057] Specifically, based on the combination of H groups of photovoltaic power generation units, the filling positions of the spatial correlation coefficients in the baseline correlation matrix are determined. The H spatial correlation coefficients are filled into the corresponding positions of the baseline correlation matrix according to the filling guidelines. The filling guidelines are the rules or basis for guiding the filling of the spatial correlation coefficients in the baseline correlation matrix. The spatial correlation matrix is ​​the baseline correlation matrix after the spatial correlation coefficients are filled in, and is used to represent the spatial correlation between photovoltaic power generation units. This implementation method combines the enumeration of photovoltaic power generation units, extracts geographic location information, calculates the spatial correlation coefficients, and fills them into the baseline correlation matrix, resulting in a matrix that accurately reflects the spatial correlation between photovoltaic power generation units. Such a matrix facilitates subsequent spectral clustering of power generation units, clustering photovoltaic power generation units with similar geographic locations and output characteristics into the same photovoltaic cluster, thereby improving the accuracy and reliability of cluster power forecasts. Furthermore, the spatial correlation matrix can provide an important basis for spatiotemporal optimization of grid connection, facilitate the development of more reasonable energy storage charging and discharging scheduling strategies, and achieve cross-cluster power complementarity within photovoltaic clusters.

[0058] In one possible implementation, power generation unit spectrum clustering is performed by fusing and analyzing the output curve similarity matrix and the spatial correlation matrix, and the multiple photovoltaic power generation units are clustered into the K photovoltaic clusters. Step S240 further includes step S241, performing exponential decay mapping on the output curve similarity matrix to generate a first normalized similarity matrix.

[0059] Specifically, an exponential decay function is applied to the elements in the output curve similarity matrix to reduce the influence of large similarity values ​​and enhance the discrimination of small similarity values. For each element in the similarity matrix, an exponential decay function is applied such as exp(-α Similarity) is an exponential decay function, where α is the decay coefficient and similarity is the original similarity value. The processed matrix is ​​normalized so that the element values ​​in the matrix are within a specific range (e.g., between 0 and 1). Exponential decay mapping is a mathematical transformation method used to adjust the values ​​of matrix elements to change their distribution characteristics. The first normalized similarity matrix is ​​the output curve similarity matrix after exponential decay mapping and normalization.

[0060] Step S242 : normalizing the spatial correlation matrix according to the matrix scale of the first normalized similarity matrix to obtain a second normalized similarity matrix.

[0061] Specifically, the scale of the first normalized similarity matrix (e.g., the range of element values) is analyzed. Based on the scale of the first normalized similarity matrix, the spatial correlation matrix is ​​normalized accordingly to make the two matrices numerically comparable. That is, based on the numerical range of the first normalized similarity matrix (e.g., maximum / minimum values), the elements in the spatial correlation matrix are linearly scaled or normalized (e.g., min-max normalization or Z-score normalization) to match the numerical range of the first normalized similarity matrix. The second normalized similarity matrix is ​​the normalized spatial correlation matrix.

[0062] Step S243 : The first normalized similarity matrix and the second normalized similarity matrix are superimposed based on a preset matrix weight coefficient to obtain a fused similarity matrix.

[0063] Specifically, the matrix weight coefficient is a pre-set coefficient used to balance the contribution of the first normalized similarity matrix and the second normalized similarity matrix in the fusion process. The two normalized similarity matrices are weighted and summed according to the weight coefficient to obtain the fused similarity matrix.

[0064] Step S244: construct a normalized Laplacian matrix based on the fused similarity matrix.

[0065] Specifically, we first calculate the degree matrix of the fused similarity matrix (i.e., a diagonal matrix whose diagonal elements are the sums of each row of the fused similarity matrix), then calculate the Laplacian matrix (i.e., the degree matrix minus the fused similarity matrix), and finally normalize the Laplacian matrix. The normalized Laplacian matrix is ​​the matrix used for spectral clustering.

[0066] Step S245 , performing eigenvalue decomposition on the normalized Laplace matrix to select K low-dimensional eigenvectors corresponding to the K minimum eigenvalues.

[0067] Specifically, the normalized Laplace matrix is ​​decomposed into eigenvalues ​​and eigenvectors. The eigenvectors corresponding to the K smallest eigenvalues ​​are selected as basis vectors of the low-dimensional feature space to construct the low-dimensional feature space.

[0068] Step S246 : After constructing a low-dimensional feature space based on the K low-dimensional feature vectors, multiple photovoltaic power generation units are mapped to the low-dimensional feature space.

[0069] Specifically, the low-dimensional feature space consists of K low-dimensional feature vectors, which are used to represent the low-dimensional features of photovoltaic power generation units. The original high-dimensional data (i.e., the output data and geographic location information of the photovoltaic power generation units) is mapped into the low-dimensional feature space. This mapping process is achieved by calculating the projection of each photovoltaic power generation unit's feature vector into the low-dimensional feature space. Specifically, for each photovoltaic power generation unit, the dot product operation is performed on its high-dimensional feature vector and K low-dimensional feature vectors to obtain the coordinate representation of the photovoltaic power generation unit in the low-dimensional feature space. In this way, each photovoltaic power generation unit has a corresponding low-dimensional feature vector in the low-dimensional feature space, thus achieving a high-dimensional to low-dimensional mapping. High-dimensional data contains multiple complex and redundant information dimensions, such as photovoltaic power generation unit output data, geographic location, and other parameters. Low-dimensionality is the process of extracting the most representative and concise key feature dimensions from the high-dimensional data through mathematical methods. Using a small number of core features to describe the originally complex high-dimensional data makes the data easier to process and analyze while retaining key information. Low-dimensional feature vectors are feature vectors whose number of dimensions is less than a dimensionality threshold, which can be set to 5.

[0070] Step S247 , outputting K photovoltaic clusters by iteratively updating cluster centers in the low-dimensional feature space.

[0071] Specifically, in a low-dimensional feature space, cluster centers are updated through an iterative algorithm (such as the K-means algorithm) to achieve more accurate clustering results. Based on the iterative results, photovoltaic power generation units are divided into K clusters. This implementation method integrates two types of similarity information (output curve similarity and spatial correlation) to more comprehensively reflect the similarities between photovoltaic power generation units, thereby improving clustering accuracy and providing strong support for subsequent power forecasting and grid connection optimization.

[0072] In one possible implementation, the K photovoltaic clusters are output by iteratively updating cluster centers in the low-dimensional feature space. Step S247 further includes step S2471, where a center-point spacing scale is preset and K cluster centers are initialized in the low-dimensional feature space based on the center-point spacing scale. Specifically, a center-point spacing scale is first set, which determines the relative distance or distribution between the initial cluster centers. Then, K points are selected in the low-dimensional feature space as initial cluster centers based on this scale, either randomly or according to a specific strategy (e.g., uniform distribution). Step S2472, where the multiple photovoltaic power generation units in the low-dimensional feature space are assigned to the K cluster centers based on the closest Euclidean distance, results in K groups of power generation unit assignments. Specifically, the Euclidean distance between each photovoltaic power generation unit and the K cluster centers is calculated, and each power generation unit is assigned to the cluster center closest to it. In this way, each cluster center has a group of photovoltaic power generation units closest to it.

[0073] In step S2473, random perturbations are iteratively performed on the K cluster centers to obtain K first-iteration centers. Specifically, the current K cluster centers are slightly randomly perturbed (i.e., randomly moved a short distance near their current positions) to obtain K new iteration centers. In step S2474, multiple photovoltaic power generation units in the low-dimensional feature space are assigned to the K first-iteration centers based on the closest Euclidean distance, obtaining K groups of first-iteration allocation results. Specifically, this is similar to step S2472, but this time the allocation is based on the new K iteration centers. In step S2475, the sum of the squared distances between the K groups of power generation unit allocation results and the K groups of first-iteration allocation results is calculated. Based on this calculation, the K cluster centers and the K first-iteration centers are determined to be retained in the iteration. Specifically, the sum of the squared distances between the current allocation result (i.e., the result of step S2472) and the iterative allocation result (i.e., the result of step S2474) is calculated. If the sum of squared distances after the iteration is smaller, it means that the iteration has improved the clustering effect, so the new iteration center point is retained; otherwise, the original cluster center point is retained. The sum of squared distances refers to the sum of the squares of the distances from all points to a certain center point, which is used to measure the compactness of the cluster.

[0074] In step S2476, when the iterative fluctuation of the sum of squared distances is less than a preset deviation, K photovoltaic clusters are output. Specifically, steps S2473 to S2475 are repeated until the iterative fluctuation of the sum of squared distances (i.e., the difference between the sum of squared distances between two consecutive iterations) is less than a preset deviation value (a threshold used to determine whether the iterations have converged). At this point, clustering is considered to have converged, and the final K photovoltaic clusters are output. This implementation method uses random perturbations and iterative updates of cluster centers, as well as comparisons of the sum of squared distances to determine whether the clustering effect has improved, ultimately obtaining compact and accurate clustering results. Accurate clustering can improve the accuracy and reliability of power forecasting, thereby optimizing energy storage charging and discharging scheduling strategies and achieving cross-cluster power complementarity among photovoltaic clusters.

[0075] Step S300 : inputting meteorological time series data collected according to the power prediction scale into K cluster power prediction models of K photovoltaic clusters, performing cluster power prediction, and outputting K cluster-level power prediction values.

[0076] Specifically, based on the characteristics of the PV cluster and forecasting requirements, an appropriate power forecasting model is selected, such as a time series model (ARIMA, Autoregressive Integrated Moving Average, LSTM, Long Short-Term Memory), a machine learning model (Random Forest, Support Vector Machine, etc.), or a deep learning model. The power forecasting model is used to predict the output of the PV cluster in the future. The selected model is trained using the cluster's historical output data and corresponding meteorological time series data as a training set. The training process includes steps such as feature selection and parameter tuning. The trained model is validated using an independent validation set to evaluate its forecasting performance (such as accuracy and stability). Based on the power forecasting time scale (such as hours, days, or weeks), corresponding meteorological time series data is collected and input into the trained cluster power forecasting model to perform power forecasting. Each model outputs a cluster-level power forecast value.

[0077] In one possible implementation, meteorological time series data collected according to the power prediction scale is input into K cluster power prediction models for K photovoltaic clusters to perform cluster power prediction. Before outputting K cluster-level power prediction values, step S300 further includes step S310, where a standard power prediction model is constructed using an LSTM. The input data for the power prediction scale are meteorological time series data and output time series data that conform to the power prediction scale, and the output is predicted power time series data that conforms to the power prediction scale with a preset scaling factor. In this application, the preset scaling factor is set to 1 / 12, meaning that the prediction time resolution output by the power prediction model is 12 times finer than the input time scale. For example, if the input is an hourly forecast (60 minutes), the output is 5-minute (60 divided by 12) power time series data. Specifically, a long short-term memory network (LSTM) is selected as the underlying architecture of the power prediction model. LSTM excels at processing time series data and can capture long-term dependencies in the data. The input and output of the LSTM model are designed. The input data is meteorological time series data and output time series data that meet the power prediction scale. After preprocessing, these data are input into the LSTM model in chronological order. The output result is the predicted power time series data that meets the power prediction scale with the preset proportional factor, that is, the model can predict the power value in a shorter time period in the future. Among them, the power prediction scale refers to the time range of the predicted power, such as hourly level, minute level, etc. In this application, the preset proportional factor can be set to 1 / 12. Meteorological time series data refers to meteorological data that changes over time, such as irradiance, temperature, etc. Output time series data refers to the output data of photovoltaic power generation units that changes over time.

[0078] Step S320 , performing time-series alignment data call based on multiple historical output data to obtain multiple meteorological data records and multiple power generation records.

[0079] Specifically, the time points corresponding to the meteorological data are extracted from the historical output data to ensure that the meteorological data and power generation data are aligned in time. Stored meteorological data and power generation data records, obtained from historical observations or simulations, are then used for model training and validation.

[0080] Step S330 , using the plurality of historical output data, the plurality of meteorological data records and the plurality of power generation records as training data to localize the standard power prediction model, and obtain a plurality of initial power prediction sub-models of the plurality of photovoltaic power generation units.

[0081] Specifically, historical output data, meteorological data records, and power generation records are extracted and retrieved as training datasets. The LSTM model parameters are adjusted during the training process to better fit the training data and improve prediction accuracy. The localization process includes steps such as data preprocessing, model structure adjustment, and hyperparameter optimization. Localization refers to the process of adapting a general model to a specific region or dataset to improve its accuracy and applicability.

[0082] Step S340 : performing federated aggregation of model parameters on multiple initial power prediction sub-models based on the K photovoltaic clusters to obtain K federated power prediction sub-models.

[0083] Specifically, the initial power prediction sub-models are grouped into clusters based on the PV cluster divisions. Federated aggregation is performed on the initial power prediction sub-models within each cluster. Federated aggregation involves merging or averaging their model parameters to obtain a federated power prediction sub-model representing the cluster. Federated aggregation improves the model's generalization capabilities by combining the wisdom of multiple sub-models.

[0084] Step S350 : replicating K federal power prediction sub-models according to the number of K photovoltaic clusters to obtain K groups of federal power prediction sub-models.

[0085] Specifically, for each photovoltaic cluster, its corresponding federal power prediction sub-model is replicated according to the number of photovoltaic power generation units in the cluster, so that each cluster has an independent federal power prediction sub-model group.

[0086] Step S360 : K groups of federated power prediction sub-models are connected in parallel within the group and a power summing engine is configured at the output end to complete the construction of K cluster power prediction models.

[0087] Specifically, the federated power prediction sub-models for each cluster are connected in parallel, running these models simultaneously and processing their outputs. A power summing engine is configured at the output end to sum the output power of the parallel models to obtain the predicted power for each cluster. This allows the predicted power of each cluster to be calculated independently and ultimately combined to form the predicted power for the entire power plant. This implementation utilizes the LSTM model and federated aggregation technology, combined with historical output data and meteorological time series data, to improve the accuracy of power forecasts and provide accurate and reliable power forecast support for subsequent grid connection optimization and energy storage scheduling.

[0088] Step S400 : performing spatiotemporal grid-connected joint optimization based on K cluster-level power prediction values, outputting energy storage charging and discharging scheduling strategies, and performing cross-cluster power complementarity among the K photovoltaic clusters.

[0089] Specifically, the power of different clusters fluctuates over time and space. Optimization algorithms (such as linear programming, dynamic programming, and genetic algorithms) are used to optimize grid-connected power scheduling, aiming to smooth power fluctuations and improve grid stability. Based on the optimization results, a charging and discharging strategy for the energy storage system is formulated. When the power forecast for a cluster is high, the energy storage system can be scheduled to charge; when the power forecast is low, the energy storage system can be scheduled to discharge to supplement power demand. Through the energy storage system's charging and discharging scheduling, power complementarity between different clusters is achieved, further improving the grid-connected performance and economic benefits of the entire distributed photovoltaic power plant.

[0090] In one possible implementation, based on the K cluster-level power prediction values, a spatiotemporal grid connection optimization is performed to output an energy storage charging and discharging scheduling strategy, enabling cross-cluster power complementarity among the K photovoltaic clusters. Step S400 further includes step S410, dynamically loading K real-time grid connection thresholds based on the K cluster-level power prediction values. Specifically, the system adjusts the grid connection thresholds in real time based on the K cluster-level power prediction values ​​(i.e., the power output prediction values ​​of each photovoltaic cluster for a future time period derived from the cluster power prediction model). The grid connection thresholds are used to determine whether the cluster power meets grid requirements. Adjustment of the grid connection thresholds can be based on factors such as the grid's capacity, the cluster's historical power fluctuation range, and current grid load demand.

[0091] An example of real-time adjustment of the grid-connected processing threshold is as follows: Assume that there are three photovoltaic clusters with power prediction values ​​as shown in Table 2.

[0092] Table 2: Cluster-level power predictions

[0093] The grid's capacity to accommodate each cluster is no more than 120 kW, taking into account the historical power fluctuation range of the cluster (assuming the fluctuation range is ±20 kW) and the current grid load demand (assuming the grid wants the cluster power to be stable near the predicted value but not exceed the capacity).

[0094] The steps for adjusting the grid connection threshold are as follows: Cluster 1's predicted power value is 100 kW, with a fluctuation range of ±20 kW. Therefore, the adjusted grid connection threshold is [80, 120] kW. Cluster 2's predicted power value is 150 kW, with a fluctuation range of ±20 kW, but the grid's upper limit is 120 kW. Therefore, the lower threshold should be the predicted value minus the fluctuation range, but it must not exceed the grid's upper limit. The adjusted grid connection threshold should be [120, 120] kW (both the lower and upper limits are the grid's upper limit). Cluster 3's predicted power value is 80 kW, with a fluctuation range of ±20 kW. Therefore, the adjusted grid connection threshold is [60, 100] kW.

[0095] Finally, the real-time grid connection processing threshold is shown in Table 3.

[0096] Table 3: Real-time grid connection processing thresholds

[0097] Step S420: K cluster-level power measurements for the K photovoltaic clusters are obtained by monitoring and summing the output power. Specifically, a power meter or data acquisition system installed at the cluster's grid connection point monitors the actual output power of each photovoltaic cluster in real time. The monitored actual output power of each photovoltaic power generation unit is summed to obtain the actual power output value of each photovoltaic cluster.

[0098] In step S430, a spatiotemporal coupled optimization is performed based on the deviations between the K measured cluster-level power values ​​and the K real-time grid-connection thresholds, outputting an energy storage charging and discharging scheduling strategy. Specifically, the measured cluster-level power values ​​are compared with the grid-connection thresholds to calculate the difference or deviation. An optimization algorithm is used to determine the optimal energy storage charging and discharging strategy, taking into account both temporal (e.g., power demand variations over time) and spatial (e.g., power complementarity between clusters). Based on the optimization results, a specific energy storage charging and discharging scheduling strategy is generated, including charging and discharging time, charging and discharging power, and other information.

[0099] Step S440, dynamic cross-cluster power complementation of K photovoltaic clusters is performed according to the energy storage charge and discharge scheduling strategy. Specifically, the charge and discharge behavior of the energy storage system is controlled according to the energy storage charge and discharge scheduling strategy. When a photovoltaic cluster has excess power, the excess power is absorbed by the energy storage system; when a cluster has insufficient power, the energy storage system releases the stored power to supplement it, thereby achieving power balance between clusters. This implementation method dynamically loads the grid-connected processing threshold, monitors power output in real time, performs spatiotemporal coupling optimization, and dynamically complements cross-cluster power. The system can flexibly respond to fluctuations and changes in photovoltaic cluster power, reduce the impact and instability of the power grid, and achieve harmonious coexistence between photovoltaic clusters and the power grid.

[0100] In one possible implementation, spatiotemporal coupling optimization is performed based on the deviations between the K cluster-level power measured values ​​and the K real-time grid-connection processing thresholds, and the energy storage charging and discharging scheduling strategy is output. Step S430 further includes step S431, which divides the K photovoltaic clusters into W excess output clusters and F insufficient output clusters based on the direction of the deviations between the K cluster-level power measured values ​​and the K real-time grid-connection processing thresholds. Specifically, W is a natural number greater than or equal to 1, and F is a natural number greater than or equal to 1. For each photovoltaic cluster, its cluster-level power measured value is compared with the real-time grid-connection processing threshold. If the measured value is greater than the real-time grid-connection processing threshold, it is determined to be excess output; if the measured value is less than the real-time grid-connection processing threshold, it is determined to be insufficient output. Based on the direction of the deviation, the K photovoltaic clusters are divided into W excess output clusters and F insufficient output clusters, where the sum of W and F is equal to K.

[0101] Step S432 performs local energy storage fitting on the W excess output clusters, outputting W excess output values. Specifically, for each excess output cluster, the excess power is calculated (i.e., the difference between the measured value and the real-time grid connection threshold) to determine the amount of excess power that can be stored. The calculated excess power is output as the excess output value for subsequent power matching.

[0102] In step S433, after extracting the F output deficit values ​​for the F under-output clusters, a power complement matching enumeration is performed based on the W excess output values ​​and the F under-output values ​​to obtain multiple initial charge and discharge scheduling strategies. Specifically, for each under-output cluster, its power deficit (i.e., the difference between the real-time grid connection processing threshold and the measured value) is calculated as the under-output value. Based on the W excess output values ​​and the F under-output values, a combination algorithm is used to enumerate various possible charge and discharge scheduling schemes, including allocating the excess power of a cluster with excess output to one or more clusters with insufficient output. All possible enumerated charge and discharge scheduling schemes are output as the initial charge and discharge scheduling strategy.

[0103] In step S434, multiple initial charge-discharge scheduling strategies are evaluated with the goal of minimizing energy storage loss and transmission costs, and a selected energy storage charge-discharge scheduling strategy is selected. Specifically, the evaluation goal is to minimize energy storage loss and transmission costs. Energy storage loss includes energy lost during the charge-discharge process, while transmission costs include the costs of power transmission. Each initial charge-discharge scheduling strategy is evaluated, and its energy storage loss and transmission costs are calculated. Based on the evaluation results, the charge-discharge scheduling strategy with the lowest energy storage loss and transmission costs is selected as the final energy storage charge-discharge scheduling strategy.

[0104] An example of performing spatiotemporal coupled optimization is as follows: Assume that there are three photovoltaic clusters whose measured power values ​​and real-time grid-connected processing thresholds are shown in Table 4.

[0105] Table 4: Measured power values ​​and real-time grid-connected processing thresholds

[0106] Cluster 1: The measured value is 110 kW, and the grid-connection threshold is [80, 120] kW, which is within the normal range. Cluster 2: The measured value is 130 kW, and the grid-connection threshold is [120, 120] kW, indicating excess output (measured value > upper limit). Cluster 3: The measured value is 50 kW, and the grid-connection threshold is [60, 100] kW, indicating insufficient output (measured value < lower limit). Based on the direction of the deviation, the three PV clusters are divided into: Cluster 2 with excess output and Cluster 3 with insufficient output. Cluster 2: Excess output = measured value - upper limit = 130 kW - 120 kW = 10 kW. The excess output of cluster 2 is 10 kW. Cluster 3: Underoutput = lower limit - measured value = 60 kW - 50 kW = 10 kW (indicating a need for additional output of 10 kW).

[0107] Possible charging and discharging scheduling schemes are enumerated as follows: Scheme 1: All of Cluster 2's 10kW excess power is allocated to Cluster 3. Scheme 2: Part of Cluster 2's 10kW excess power is allocated to Cluster 3, and the remainder is stored in the local energy storage system (for example, 5kW is allocated to Cluster 3 and 5kW is stored in the local energy storage system).

[0108] The calculation formula for energy storage loss and transmission cost is: energy storage loss = charging and discharging amount × loss coefficient (assuming the loss coefficient is 0.1); transmission cost = transmission amount × transmission unit price (assuming the transmission unit price is 0.5 yuan / kW).

[0109] Evaluation Scheme 1: Energy storage loss: 0 (all used to supplement power shortages), transmission cost: 10kW × 0.5 yuan / kW = 5 yuan, total cost: 5 yuan. Evaluation Scheme 2: Energy storage loss: 5kW × 0.1 = 0.5kW, transmission cost: 5kW × 0.5 yuan / kW = 2.5 yuan, total cost: 0.5kW × 0.5 yuan / kW + 2.5 yuan = 2.75 yuan. Scheme 2's total cost of 2.75 yuan is less than Scheme 1's 5 yuan, so Scheme 2 is selected as the final energy storage charging and discharging scheduling strategy. The final scheduling strategy is: Cluster 2's 5kW excess power is allocated to Cluster 3, and Cluster 2's remaining 5kW is stored in the local energy storage system. This implementation determines the direction of cluster power deviation, classifies clusters with excess and insufficient output, enumerates power complementarity matches, and evaluates and screens to determine the optimal energy storage charging and discharging scheduling strategy. This system can flexibly respond to fluctuations and changes in PV cluster power, improve renewable energy utilization, optimize grid operation, and ensure the stability and reliability of power supply.

[0110] Step S500 : After performing single-station fitting conversion based on K cluster-level power prediction values ​​and outputting distributed power prediction values, output deviation identification is performed based on the distributed power prediction values ​​to locate the fault node of the photovoltaic power station.

[0111] Specifically, cluster-level power forecasts are allocated to each PV plant using either a weighted average or proportional allocation method. If the weighted average method is used, the cluster-level power forecast is assigned to each PV plant based on its historical power contribution to the cluster. Weights can be calculated based on factors such as historical power generation, geographic location, and equipment capacity. If the proportional allocation method is used, the cluster-level power forecast is assigned to each PV plant based on the equipment capacity ratio of each PV plant. Using a threshold comparison method, the difference (deviation) between the actual power and the forecasted power is calculated for each PV plant. A reasonable deviation threshold (such as 10% or 20%) is set. If the deviation exceeds the deviation threshold, the plant is considered to have an anomaly. Combining geographic location information with historical data, plants with deviations exceeding the deviation threshold are further analyzed to identify the faulty node. This includes examining environmental factors surrounding the faulty plant (such as whether there are obstructions or whether it is in a shadowed area). Comparing the plant's historical power data with the faulty plant's power data determines whether the failure is sudden or a chronic issue.

[0112] For example, suppose there is a PV cluster consisting of three PV plants (A, B, and C), with a cluster-level power forecast of 300 kW. Using a weighted average method, assume that the historical power contributions of plants A, B, and C are 30%, 40%, and 30%, respectively. The allocated power forecasts are: Plant A: 300 kW x 30% = 90 kW; Plant B: 300 kW x 40% = 120 kW; and Plant C: 300 kW x 30% = 90 kW. Assume the actual measured power values ​​are: Plant A: 80 kW; Plant B: 125 kW; and Plant C: 88 kW. The calculated deviations are: Plant A: |80 kW - 90 kW | = 10 kW; Plant B: |125 kW - 120 kW | = 5 kW; and Plant C: |88 kW - 90 kW | = 2 kW. Assuming a deviation threshold of 10 kW, the deviation of Plant A is greater than or equal to the deviation threshold, making it identified as an anomaly. Checking Plant A's geographic location reveals that newly planted trees are located nearby, potentially obscuring some of the solar panels.

[0113] Step S600: Dynamically update the energy storage charging and discharging scheduling strategy according to the faulty node of the photovoltaic power station.

[0114] Specifically, the energy storage system's charge and discharge schedule is adjusted based on the power deviation of the faulty node. If the faulty power station's power is insufficient, the energy storage system increases discharge to compensate for the power shortfall. Power meters and data acquisition systems monitor the actual power of each photovoltaic power station in real time. Based on this real-time data, the energy storage system's charge and discharge schedule is dynamically updated to ensure stable grid operation.

[0115] For example, suppose Power Station A is identified as a faulty node with insufficient power. Its actual power is 80 kW, while its predicted power is 90 kW, a deviation of 10 kW. The energy storage system increases its discharge by 10 kW to compensate for Power Station A's power shortfall. Based on real-time monitoring data, the energy storage system dynamically adjusts its discharge capacity to ensure stable grid operation. Table 5 shows an example of a dynamic update of the energy storage charging and discharging scheduling strategy.

[0116] Table 5: Example of dynamic update of energy storage charging and discharging scheduling strategy

[0117] The embodiment of the present application uses local call and time series alignment preprocessing of multi-source data to construct a multi-dimensional feature association table, extract information from the table for similarity analysis, cluster the spectrum of photovoltaic power generation units to obtain K photovoltaic clusters, input meteorological time series data into the cluster power prediction model, cluster power prediction to obtain K cluster-level power prediction values, perform grid-connected spatiotemporal joint optimization based on the cluster-level power prediction values, output energy storage charge and discharge scheduling strategy to achieve cross-cluster power complementarity, obtain distributed power prediction values ​​based on single-station fitting conversion of cluster-level power prediction values, identify output deviations to locate faulty nodes, and dynamically update energy storage charge and discharge scheduling strategy, etc., to achieve the following technical effects: 1. By locally calling and preprocessing multi-source data with time series alignment, and combining geographic location information and historical output data for similarity analysis and spectral clustering, an accurate cluster power prediction model is constructed. This solves the problem of the existing technology that cannot accurately predict distributed photovoltaic power, and achieves the technical effect of improving the accuracy of photovoltaic power prediction.

[0118] 2. Based on the cluster-level power forecast value, the grid-connected time and space joint optimization is carried out, and the energy storage charging and discharging scheduling strategy is output to achieve dynamic power complementarity across clusters. This solves the problem of cross-cluster power optimization scheduling that is difficult to achieve in existing technologies, and achieves the technical effect of improving the stability of grid operation and economic benefits.

[0119] 3. By dynamically monitoring power output and adjusting the energy storage charging and discharging strategy in real time, it can quickly respond to power deviations, compensate for power gaps or absorb excess power. This solves the problem of poor grid operation stability in existing technologies, achieving the technical effect of reducing grid shocks and ensuring stable grid operation.

[0120] 4. Single-station fitting conversion and output deviation identification are used, combined with geographic location information and historical data analysis to accurately locate photovoltaic power stations with abnormal power. This solves the problem of the inability to effectively identify and locate faulty nodes in existing technologies, achieving the technical effect of quickly discovering and handling faults and reducing maintenance costs.

[0121] In the above, refer to Figure 1 A method for predicting distributed photovoltaic power based on clustering optimization is described in detail. Figure 2 A jointly optimized clustered distributed photovoltaic power prediction system according to an embodiment of the present application is described.

[0122] According to an embodiment of the present application, a jointly optimized clustered distributed photovoltaic power prediction system is used to address the technical problems of low prediction accuracy in existing distributed photovoltaic power prediction, difficulty in achieving cross-cluster power optimization scheduling, and thus poor grid operation stability. The system achieves the technical effects of improving the accuracy of photovoltaic power prediction, achieving cross-cluster power optimization scheduling, and thus improving grid operation stability. The jointly optimized clustered distributed photovoltaic power prediction system includes: a timing alignment module 10, a power generation unit spectrum clustering module 20, a cluster power prediction module 30, an energy storage charge and discharge scheduling strategy output module 40, an output deviation identification module 50, and an energy storage charge and discharge scheduling strategy update module 60.

[0123] The timing alignment module 10 is used to locally call multi-source data of the distributed photovoltaic power station and perform timing alignment preprocessing on the call results to construct a multi-dimensional feature association number table; the power generation unit spectrum clustering module 20 is used to extract multiple geographical location information and multiple historical output data of multiple photovoltaic power generation units in the distributed photovoltaic power station from the multi-dimensional feature association number table to perform similarity analysis, and perform power generation unit spectrum clustering on the multiple photovoltaic power generation units based on the analysis results, and output K photovoltaic clusters, where K is a natural number greater than or equal to 1; the cluster power prediction module 30 is used to input the meteorological time series data collected according to the power prediction scale into the K clusters of the K photovoltaic clusters A power prediction model is provided for cluster power prediction and outputting K cluster-level power prediction values; an energy storage charge and discharge scheduling strategy output module 40 is provided for performing spatiotemporal joint optimization of grid connection based on the K cluster-level power prediction values, outputting an energy storage charge and discharge scheduling strategy, and performing cross-cluster power complementarity of the K photovoltaic clusters; an output deviation identification module 50 is provided for performing single-station fitting conversion based on the K cluster-level power prediction values, outputting distributed power prediction values, and then identifying output deviations based on the distributed power prediction values ​​to locate fault nodes in photovoltaic power stations; and an energy storage charge and discharge scheduling strategy update module 60 is provided for dynamically updating the energy storage charge and discharge scheduling strategy based on the fault nodes in photovoltaic power stations.

[0124] The specific configuration of the power generation unit spectrum clustering module 20 will be described in detail below. As described above, multiple geographic location information and multiple historical output data of multiple photovoltaic power generation units in the distributed photovoltaic power station are extracted from the multidimensional feature association number table for similarity analysis, and the multiple photovoltaic power generation units are subjected to power generation unit spectrum clustering based on the analysis results to output K photovoltaic clusters. The power generation unit spectrum clustering module 20 may further include: a unit output curve drawing unit for drawing multiple unit output curves based on the multiple historical output data; an output curve similarity matrix output unit for analyzing and outputting the output curve similarity matrix of the multiple photovoltaic power generation units through a dynamic time warping algorithm; a spatial correlation matrix construction unit for constructing a spatial correlation matrix based on the multiple geographic location information; and a power generation unit spectrum clustering unit for performing power generation unit spectrum clustering by fusing and analyzing the output curve similarity matrix and the spatial correlation matrix to cluster the multiple photovoltaic power generation units into the K photovoltaic clusters.

[0125] The specific configuration of the energy storage charge and discharge scheduling strategy output module 40 will be described in detail below. As described above, the grid-connected spatiotemporal optimization is performed based on the K cluster-level power prediction values, and the energy storage charge and discharge scheduling strategy is output to perform cross-cluster power complementarity of the K photovoltaic clusters. The energy storage charge and discharge scheduling strategy output module 40 may further include: a real-time grid-connected processing threshold loading unit for dynamically loading K real-time grid-connected processing thresholds based on the K cluster-level power prediction values; a cluster-level power measured value acquisition unit for obtaining K cluster-level power measured values ​​of the K photovoltaic clusters by summing the output power monitoring; a spatiotemporal coupling optimization unit for performing spatiotemporal coupling optimization based on the deviation between the K cluster-level power measured values ​​and the K real-time grid-connected processing thresholds, and outputting the energy storage charge and discharge scheduling strategy; and a cross-cluster power dynamic complementation unit for performing cross-cluster power dynamic complementarity of the K photovoltaic clusters based on the energy storage charge and discharge scheduling strategy.

[0126] Among them, the power generation unit spectrum clustering is performed by fusing and analyzing the output curve similarity matrix and the spatial correlation matrix, and the multiple photovoltaic power generation units are clustered into the K photovoltaic clusters. The power generation unit spectrum clustering unit can further include: a first normalized similarity matrix generation subunit is used to perform exponential decay mapping on the output curve similarity matrix to generate a first normalized similarity matrix; a second normalized similarity matrix generation subunit is used to normalize the spatial correlation matrix according to the matrix scale of the first normalized similarity matrix to obtain a second normalized similarity matrix; a similarity matrix superposition subunit is used to superimpose the spatial correlation matrix based on a preset matrix weight coefficient. The first normalized similarity matrix and the second normalized similarity matrix are combined to obtain a fused similarity matrix; the normalized Laplace matrix construction subunit is used to construct a normalized Laplace matrix based on the fused similarity matrix; the eigenvalue decomposition subunit is used to perform eigenvalue decomposition on the normalized Laplace matrix to select K low-dimensional eigenvectors corresponding to K minimum eigenvalues; the mapping subunit is used to construct a low-dimensional feature space based on the K low-dimensional feature vectors, and then map the multiple photovoltaic power generation units to the low-dimensional feature space; the cluster center point iterative update subunit is used to iteratively update the cluster center points in the low-dimensional feature space to output the K photovoltaic clusters.

[0127] The K photovoltaic clusters are outputted by iteratively updating the cluster center points in the low-dimensional feature space. The cluster center point iterative updating subunit may further include: a cluster center point initialization component for presetting the center point spacing scale, and initializing the K cluster center points in the low-dimensional feature space based on the center point spacing scale; a photovoltaic power generation unit allocation component for allocating the multiple photovoltaic power generation units in the low-dimensional feature space to the K cluster center points based on the closest Euclidean distance, and obtaining K groups of power generation unit allocation results; a random perturbation iteration component for iteratively performing random perturbations on the K cluster center points. generations, obtaining K first iteration center points; the photovoltaic power generation unit allocation component is used to allocate the multiple photovoltaic power generation units in the low-dimensional feature space to the K first iteration center points based on the closest Euclidean distance, and obtain K groups of first iteration allocation results; the center point iterative retention judgment component is used to compare the sum of squares of the distances between the K groups of power generation unit allocation results and the K groups of first iteration allocation results, and perform iterative retention judgment on the K cluster center points and the K first iteration center points based on the comparison results; the photovoltaic cluster output component is used to output the K photovoltaic clusters when the iterative fluctuation of the sum of squares of the distances is less than a preset deviation.

[0128] Wherein, according to the deviation between the K cluster-level power measured values ​​and the K real-time grid-connected processing thresholds, spatiotemporal coupling optimization is performed to output the energy storage charging and discharging scheduling strategy. The spatiotemporal coupling optimization unit may further include: a photovoltaic cluster division subunit for dividing the K photovoltaic clusters into W excess output clusters and F insufficient output clusters according to the deviation direction between the K cluster-level power measured values ​​and the K real-time grid-connected processing thresholds, where W is a natural number greater than or equal to 1 and F is a natural number greater than or equal to 1; a local energy storage fitting subunit for performing local energy storage fitting on the W excess output clusters and outputting W excess output values; a power complementary matching enumeration subunit for extracting F insufficient output values ​​of the F insufficient output clusters and then performing power complementary matching enumeration based on the W excess output values ​​and the F insufficient output values ​​to obtain multiple initial charging and discharging scheduling strategies; and an energy storage charging and discharging scheduling strategy acquisition subunit for evaluating the multiple initial charging and discharging scheduling strategies with the goal of minimizing energy storage loss and transmission cost, and screening out the energy storage charging and discharging scheduling strategy.

[0129] The specific configuration of the cluster power prediction module 30 will be described in detail below. As described above, the meteorological time series data collected according to the power prediction scale is input into the K cluster power prediction models of the K photovoltaic clusters to perform cluster power prediction. Before outputting K cluster-level power prediction values, the cluster power prediction module 30 may further include: a standard power prediction model construction unit for constructing a standard power prediction model using LSTM, wherein the input data of the power prediction scale are meteorological time series data and output time series data that conform to the power prediction scale, and the output result is the predicted power time series data of the power prediction scale that conforms to the preset proportional factor; a data calling unit for performing time series alignment data calling based on the multiple historical output data to obtain multiple meteorological data records and multiple power generation power records; a standard power prediction model localization unit The unit is used to localize the standard power prediction model using the multiple historical output data, multiple meteorological data records and multiple power generation power records as training data to obtain multiple initial power prediction sub-models of the multiple photovoltaic power generation units; the parameter federal aggregation unit is used to perform model parameter federal aggregation on the multiple initial power prediction sub-models based on the K photovoltaic clusters to obtain K federal power prediction sub-models; the sub-model replication unit is used to replicate the K federal power prediction sub-models according to the number of clusters of the K photovoltaic clusters to obtain K groups of federal power prediction sub-models; the cluster power prediction model construction unit is used to complete the construction of the K cluster power prediction models by paralleling the K groups of federal power prediction sub-models within the group and configuring a power summation engine at the output end.

[0130] Wherein, a spatial correlation matrix is ​​constructed based on the multiple geographic location information, and the spatial correlation matrix construction unit may further include: a photovoltaic power generation unit combination enumeration subunit for enumerating the multiple photovoltaic power generation units through a combination algorithm to obtain H groups of photovoltaic power generation units, where H is a natural number greater than or equal to 1; a geographic location information acquisition subunit for combining the multiple geographic location information based on the H groups of photovoltaic power generation units to obtain H groups of geographic location information, wherein the geographic location information includes the longitude and latitude of the photovoltaic unit, the altitude of the photovoltaic unit, the inclination of the photovoltaic panel and the azimuth of the photovoltaic panel; an inter-group Euclidean distance calculation subunit for processing the H groups of geographic location information based on preset geographic normalization weights, performing inter-group Euclidean distance calculation on the processing results, and outputting H spatial correlation coefficients; a benchmark correlation matrix construction subunit for constructing a benchmark correlation matrix based on the multiple photovoltaic power generation units; a correlation coefficient filling subunit for filling the H spatial correlation coefficients into the benchmark correlation matrix using the H groups of photovoltaic power generation units as a filling guide to complete the construction of the spatial correlation matrix.

[0131] The specific configuration of the timing alignment module 10 will be described in detail below. As described above, a distributed photovoltaic power station performs local multi-source data calls and performs timing alignment preprocessing on the call results to construct a multi-dimensional feature association table. The timing alignment module 10 may further include: a local data call unit for performing local data calls through multiple edge acquisition terminals deployed in the multiple photovoltaic power generation units to obtain multiple meteorological data records, multiple output data records, and the multiple geographic location information; a sliding mean calculation unit for presetting an output sliding window and performing sliding mean calculation on the multiple output data records according to the output sliding window to obtain multiple downsampled output data; a linear interpolation processing unit for presetting a meteorological interpolation granularity and performing linear interpolation processing on the multiple meteorological data records using the meteorological interpolation granularity to obtain multiple upsampled meteorological data; and an associated storage unit for performing timing alignment on the multiple downsampled output data and the multiple upsampled meteorological data, and then associatively storing the multiple downsampled output data, the multiple upsampled meteorological data, and the multiple geographic location information based on the multiple photovoltaic power generation units to complete the construction of the multi-dimensional feature association table.

[0132] A jointly optimized clustered distributed photovoltaic power prediction system provided in an embodiment of the present application can execute a jointly optimized clustered distributed photovoltaic power prediction method provided in any embodiment of the present application, and has functional modules and beneficial effects corresponding to the execution method.

[0133] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present application.

[0134] The above specific embodiments do not constitute a limitation to the scope of protection of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of this application should be included in the scope of protection of this application. In some cases, the actions or steps recorded in this application can be performed in an order different from that in the embodiments and can still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A joint optimization clustering distributed photovoltaic power prediction method, characterized in that: The method comprises: Locally call multi-source data of distributed photovoltaic power plants and perform time series alignment preprocessing on the call results to construct a multi-dimensional feature association table; Extracting multiple geographic location information and multiple historical output data of multiple photovoltaic power generation units in the distributed photovoltaic power station from the multidimensional feature association number table to perform similarity analysis, and performing power generation unit spectral clustering on the multiple photovoltaic power generation units based on the analysis results, and outputting K photovoltaic clusters, where K is a natural number greater than or equal to 1; Inputting meteorological time series data collected according to the power prediction scale into K cluster power prediction models of the K photovoltaic clusters to perform cluster power prediction and output K cluster-level power prediction values; Performing spatiotemporal grid-connected joint optimization based on the K cluster-level power prediction values, outputting an energy storage charging and discharging scheduling strategy, and performing cross-cluster power complementarity among the K photovoltaic clusters; After performing single-station fitting conversion based on the K cluster-level power prediction values ​​and outputting distributed power prediction values, output deviation identification is performed based on the distributed power prediction values ​​to locate the fault node of the photovoltaic power station; The energy storage charging and discharging scheduling strategy is dynamically updated according to the fault node of the photovoltaic power station.

2. A joint optimization clustering distributed photovoltaic power prediction method according to claim 1, characterized in that: Extracting multiple geographic location information and multiple historical output data of multiple photovoltaic power generation units in the distributed photovoltaic power station from the multidimensional feature association number table to perform similarity analysis, and performing power generation unit spectral clustering on the multiple photovoltaic power generation units based on the analysis results to output K photovoltaic clusters, including: Drawing a plurality of unit output curves according to the plurality of historical output data; Analyzing the output curves of the plurality of units using a dynamic time warping algorithm, and outputting a similarity matrix of the output curves of the plurality of photovoltaic power generation units; constructing a spatial correlation matrix based on the plurality of geographic location information; The power generation unit spectrum clustering is performed by fusing and analyzing the output curve similarity matrix and the spatial correlation matrix, and the multiple photovoltaic power generation units are clustered into the K photovoltaic clusters.

3. The method for predicting distributed photovoltaic power by clustering based on joint optimization according to claim 1, wherein: Performing spatiotemporal grid-connected joint optimization based on the K cluster-level power prediction values, outputting an energy storage charging and discharging scheduling strategy, and performing cross-cluster power complementarity among the K photovoltaic clusters, including: Dynamically loading K real-time grid connection processing thresholds according to the K cluster-level power prediction values; Obtaining K cluster-level power measured values ​​of the K photovoltaic clusters by summing the output power monitoring; Performing spatiotemporal coupling optimization based on the deviations between the K cluster-level power measured values ​​and the K real-time grid-connected processing thresholds to output the energy storage charging and discharging scheduling strategy; Dynamic cross-cluster power complementarity of the K photovoltaic clusters is performed according to the energy storage charging and discharging scheduling strategy.

4. A joint optimization clustering distributed photovoltaic power prediction method according to claim 2, characterized in that: The power generation unit spectrum clustering is performed by fusing and analyzing the output curve similarity matrix and the spatial correlation matrix, and the plurality of photovoltaic power generation units are clustered into the K photovoltaic clusters, including: Performing exponential decay mapping on the output curve similarity matrix to generate a first normalized similarity matrix; Normalizing the spatial correlation matrix according to the matrix scale of the first normalized similarity matrix to obtain a second normalized similarity matrix; Superimposing the first normalized similarity matrix and the second normalized similarity matrix based on a preset matrix weight coefficient to obtain a fused similarity matrix; constructing a normalized Laplace matrix based on the fused similarity matrix; Performing eigenvalue decomposition on the normalized Laplace matrix to select K low-dimensional eigenvectors corresponding to the K smallest eigenvalues; After constructing a low-dimensional feature space based on the K low-dimensional feature vectors, mapping the plurality of photovoltaic power generation units to the low-dimensional feature space; The K photovoltaic clusters are output by iteratively updating cluster centers in the low-dimensional feature space.

5. A joint optimization clustering distributed photovoltaic power prediction method according to claim 4, characterized in that: The K photovoltaic clusters are output by iteratively updating the cluster center points in the low-dimensional feature space, including: Preset a center point spacing scale, and initialize K cluster center points in the low-dimensional feature space based on the center point spacing scale; Based on the closest Euclidean distance, the plurality of photovoltaic power generation units in the low-dimensional feature space are allocated to the K cluster centers to obtain K groups of power generation unit allocation results; Performing random perturbation iteration on the K cluster center points to obtain K first iteration center points; Based on the closest Euclidean distance, the plurality of photovoltaic power generation units in the low-dimensional feature space are allocated to the K first iteration center points to obtain K groups of first iteration allocation results; Calculating the sum of squares of the distances between the K groups of power generation unit allocation results and the K groups of first iteration allocation results, and making an iterative retention judgment of the K cluster centers and the K first iteration center points based on the calculation results; When the iterative fluctuation of the sum of squared distances is less than a preset deviation, the K photovoltaic clusters are output.

6. A joint optimization clustering distributed photovoltaic power prediction method according to claim 3, characterized in that: Based on the deviations between the K cluster-level power measured values ​​and the K real-time grid-connected processing thresholds, a spatiotemporal coupling optimization is performed to output the energy storage charging and discharging scheduling strategy, including: According to the deviation direction of the K cluster-level power measured values ​​and the K real-time grid-connection processing thresholds, the K photovoltaic clusters are divided into W excess output clusters and F insufficient output clusters, where W is a natural number greater than or equal to 1, and F is a natural number greater than or equal to 1; Performing local energy storage fitting on the W excess output clusters, and outputting W excess output values; After extracting the F output deficiency values ​​of the F output deficiency clusters, processing is performed according to the W excess output values ​​and the F output deficiency values ​​to obtain a plurality of initial charge and discharge scheduling strategies; With the goal of minimizing energy storage loss and transmission cost, the multiple initial charge and discharge scheduling strategies are evaluated and the energy storage charge and discharge scheduling strategy is screened out.

7. The method for predicting distributed photovoltaic power by clustering based on joint optimization according to claim 1, wherein: Before inputting meteorological time series data collected according to the power prediction scale into K cluster power prediction models of the K photovoltaic clusters to perform cluster power prediction and output K cluster-level power prediction values, the method further includes: A standard power prediction model is constructed using LSTM, wherein the input data of the power prediction scale is meteorological time series data and output time series data that conform to the power prediction scale, and the output result is predicted power time series data that conforms to the power prediction scale with a preset scale factor; Performing time-series alignment data calls based on the multiple historical output data to obtain multiple meteorological data records and multiple power generation records; localizing the standard power prediction model using the plurality of historical output data, the plurality of meteorological data records, and the plurality of power generation records as training data to obtain a plurality of initial power prediction sub-models for the plurality of photovoltaic power generation units; According to the K photovoltaic clusters, performing federal aggregation of model parameters on the multiple initial power prediction sub-models to obtain K federal power prediction sub-models; replicating the K federal power prediction sub-models according to the number of the K photovoltaic clusters to obtain K groups of federal power prediction sub-models; The K groups of federated power prediction sub-models are connected in parallel within the group and a power summing engine is configured at the output end to complete the construction of the K cluster power prediction models.

8. The method for predicting distributed photovoltaic power by clustering based on joint optimization according to claim 2, wherein: Constructing a spatial correlation matrix according to the plurality of geographical location information includes: Enumerating the plurality of photovoltaic power generation units by a combination algorithm to obtain H groups of photovoltaic power generation units, where H is a natural number greater than or equal to 1; Combining the plurality of geographic location information according to the H groups of photovoltaic power generation units to obtain H groups of geographic location information; wherein the geographic location information includes the longitude and latitude of the photovoltaic unit, the altitude of the photovoltaic unit, the inclination angle of the photovoltaic panel, and the azimuth angle of the photovoltaic panel; After processing the H groups of geographic location information based on preset geographic normalization weights, performing Euclidean distance calculations on the processed results to output H spatial correlation coefficients; constructing a benchmark correlation matrix based on the plurality of photovoltaic power generation units; Using the H groups of photovoltaic power generation units as a filling guide, the H spatial correlation coefficients are filled into the reference correlation matrix to complete the construction of the spatial correlation matrix.

9. The method for predicting distributed photovoltaic power by clustering based on joint optimization according to claim 1, wherein: Locally call multi-source data for distributed photovoltaic power plants and perform time series alignment preprocessing on the call results to construct a multi-dimensional feature association table, including: Local data calls are performed by multiple edge collection terminals deployed at the multiple photovoltaic power generation units to obtain multiple meteorological data records, multiple output data records and the multiple geographic location information; Preset an output sliding window, and perform sliding mean calculation on the multiple output data records according to the output sliding window to obtain multiple downsampled output data; Presetting a meteorological interpolation granularity, and performing linear interpolation processing on the plurality of meteorological data records using the meteorological interpolation granularity to obtain a plurality of upsampled meteorological data; After time-series alignment of the multiple downsampled output data and the multiple upsampled meteorological data, the multiple downsampled output data, the multiple upsampled meteorological data and the multiple geographic location information are stored in association with the multiple photovoltaic power generation units to complete the construction of the multidimensional feature association table.

10. A jointly optimized clustered distributed photovoltaic power prediction system, characterized in that: The system comprises: The timing alignment module is used to locally call multi-source data for distributed photovoltaic power plants and perform timing alignment preprocessing on the call results to construct a multi-dimensional feature association table; a power generation unit spectrum clustering module, configured to extract multiple geographic location information and multiple historical output data of multiple photovoltaic power generation units in the distributed photovoltaic power station from the multidimensional feature association number table, perform similarity analysis, and perform power generation unit spectrum clustering on the multiple photovoltaic power generation units based on the analysis results, and output K photovoltaic clusters, where K is a natural number greater than or equal to 1; a cluster power prediction module, configured to input meteorological time series data collected according to the power prediction scale into K cluster power prediction models of the K photovoltaic clusters, perform cluster power prediction, and output K cluster-level power prediction values; An energy storage charge and discharge scheduling strategy output module is used to perform spatiotemporal grid-connected optimization based on the K cluster-level power prediction values, output an energy storage charge and discharge scheduling strategy, and perform cross-cluster power complementarity among the K photovoltaic clusters; An output deviation identification module is used to perform single-station fitting conversion based on the K cluster-level power prediction values ​​and output a distributed power prediction value, and then perform output deviation identification based on the distributed power prediction value to locate the fault node of the photovoltaic power station; The energy storage charge and discharge scheduling strategy update module is used to dynamically update the energy storage charge and discharge scheduling strategy according to the fault node of the photovoltaic power station.

Images