Power prediction method and system for distributed new energy cluster

Abstract

The application provides a power prediction method and system for a distributed new energy cluster, and relates to the technical field of power prediction. The method comprises the following steps: based on the energy types, operating characteristics and component structures of multiple new energy units in the distributed new energy cluster, a new energy unit third cluster is constructed; based on the power data of the new energy unit third cluster, importance calculation is performed, and according to the importance, a screening threshold is set to obtain a representative unit; according to the representative unit, spatio-temporal characteristic power prediction is performed to obtain a spatio-temporal prediction result; and energy scheduling and energy storage control are performed through the spatio-temporal prediction result. The application can solve the technical problem of low power prediction accuracy of the distributed new energy cluster in the prior art, and achieve the technical effect of improving the power prediction accuracy.

Description

Technical Field

[0001] This application relates to the field of power prediction technology, and in particular to a power prediction method and system for distributed new energy clusters. Background Technology

[0002] With the large-scale integration of distributed energy technologies and new energy power generation devices, energy systems are gradually evolving from centralized structures to distributed systems characterized by multiple sources, heterogeneity, and regional interconnection. Especially in scenarios where multiple types of new energy units, such as photovoltaics, wind power, energy storage, and microgrids, operate in synergy, the volatility, randomness, and spatiotemporal differences in power output are significantly enhanced. This makes the prediction and regulation of the overall power change trend of the system a key link in achieving efficient energy utilization and stable grid operation.

[0003] Currently, existing prediction methods often focus only on power variation patterns over time, neglecting the coupling effects caused by spatial structure and inter-node interactions. For example, multiple wind farms or photovoltaic power plants in the same region exhibit significant spatial correlations in power output due to differences in meteorological conditions, geographical location, and equipment type. However, traditional methods typically treat these as independent samples, failing to effectively reflect the interactions between nodes. Furthermore, while some methods incorporate spatial clustering or feature dimensionality reduction techniques, they often employ fixed thresholds or static weights, making it difficult to maintain the model's generalization ability in dynamic operating environments.

[0004] In summary, existing technologies suffer from a lack of a joint modeling mechanism for the spatial coupling characteristics and temporal evolution patterns in distributed new energy clusters. This makes it difficult for power prediction models to accurately depict the dynamic relationships and mutual influences between different nodes, further affecting the overall power prediction accuracy, energy dispatch efficiency, and response reliability of the energy storage system. Summary of the Invention

[0005] The purpose of this application is to provide a power prediction method and system for distributed renewable energy clusters, in order to solve the technical problem in the prior art that the lack of a joint modeling mechanism for the spatial coupling characteristics and temporal evolution law in distributed renewable energy clusters makes it difficult for the power prediction model to accurately depict the dynamic correlation and mutual influence between different nodes, which further affects the power prediction accuracy, energy dispatch efficiency and energy storage system response reliability of the overall energy system.

[0006] In view of the above problems, this application provides a power prediction method and system for distributed new energy clusters.

[0007] In a first aspect, this application provides a power prediction method for distributed renewable energy clusters, implemented through a power prediction system for distributed renewable energy clusters, comprising: constructing a tertiary cluster of renewable energy units based on the energy type, operating characteristics, and composition structure of multiple renewable energy units in the distributed renewable energy cluster; calculating the importance of the power data of the tertiary cluster of renewable energy units, and selecting representative units by setting a screening threshold according to the importance; performing spatiotemporal characteristic power prediction based on the representative units to obtain spatiotemporal prediction results; and performing energy scheduling and energy storage control based on the spatiotemporal prediction results.

[0008] Preferably, the power prediction method for distributed new energy clusters further includes: constructing an energy similarity granularity based on the energy type of the multiple new energy units, dividing similar energy based on the energy similarity granularity to obtain an initial cluster of new energy units; constructing a power similarity granularity based on the operating characteristics and similar power output characteristics, dividing the initial cluster of new energy units based on the power similarity granularity to obtain a secondary cluster of new energy units; and constructing a structural similarity granularity based on the composition structure, dividing the secondary cluster of new energy units based on the structural similarity granularity to obtain a tertiary cluster of new energy units.

[0009] Preferably, the power prediction method for distributed new energy clusters further includes: the operating characteristics include the shape of the power output curve, the power output fluctuation rate, the rated power of the equipment, and the meteorological response sensitivity.

[0010] Preferably, the power prediction method for distributed new energy clusters further includes: calculating the mean of the power data of each of the three clusters of the new energy units to obtain a cluster power sequence; obtaining the total power data of the three clusters of the new energy units and calculating the Pearson correlation coefficient between the cluster power sequence and the total power data; calculating the inverse variance of the power data of each of the three clusters of the new energy units to obtain a stability coefficient; calculating the proportion of the power data of each of the three clusters of the new energy units based on the total power data to obtain a contribution rate coefficient; and performing a weighted summation of the Pearson correlation coefficient, the stability coefficient, and the contribution rate coefficient based on the coefficient of variation method to obtain the importance of each of the three clusters of the new energy units.

[0011] Preferably, the power prediction method for distributed new energy clusters further includes: constructing a spatiotemporal correlation graph based on the representative unit as a node; extracting node feature vectors and dividing them into time windows, then concatenating the divided feature vectors according to the time windows to obtain a spatiotemporal feature matrix; performing power prediction on the spatiotemporal feature matrix to obtain a power prediction result; and mapping the power prediction result to the spatiotemporal correlation graph to obtain the spatiotemporal prediction result.

[0012] Preferably, the power prediction method for distributed new energy clusters further includes: constructing a first distance matrix and a first power time-series correlation matrix; fusing the first distance matrix and the first power time-series correlation matrix using a threshold filtering method to obtain a first weighted adjacency matrix; obtaining a first node and a second node based on the nodes; and establishing a graph structure for the first node and the second node and the first edge between the first node and the second node based on the first weighted adjacency matrix to obtain the spatiotemporal correlation graph.

[0013] Preferably, the power prediction method for distributed new energy clusters further includes: obtaining the spatial coordinates of the first node, the first power time series data, the spatial coordinates of the second node, and the second power time series data based on the first node and the second node; calculating a first distance matrix between the spatial coordinates of the first node and the spatial coordinates of the second node using Euclidean distance; and calculating the time series correlation based on the first power time series data and the second power time series data to obtain the first power time series correlation matrix.

[0014] Preferably, the power prediction method for distributed new energy clusters further includes: constructing a spatiotemporal prediction model, wherein the spatiotemporal prediction model includes a spatial correlation prediction channel and a time dependence prediction channel; inputting the spatiotemporal feature matrix into the spatiotemporal prediction model, extracting power spatial correlation through the spatial correlation prediction channel, extracting power time dependence based on the time dependence prediction channel, and combining to obtain the power prediction result.

[0015] Preferably, the power prediction method for distributed new energy clusters further includes: the spatiotemporal prediction model adopts a cascaded structure of a graph convolutional network and a long short-term memory network, wherein the graph convolutional network is used to extract spatial correlation features, and the long short-term memory network is used to extract temporal dependence features.

[0016] Secondly, this application also provides a power prediction system for distributed new energy clusters, used to execute the power prediction method for distributed new energy clusters as described in the first aspect, comprising: a cluster construction module, used to construct a three-dimensional cluster of new energy units based on the energy type, operating characteristics and composition structure of multiple new energy units in the distributed new energy cluster; a unit acquisition module, used to perform importance calculation based on the power data of the three-dimensional cluster of new energy units, and to select representative units according to the importance by setting a screening threshold; a result acquisition module, used to perform spatiotemporal characteristic power prediction based on the representative units to obtain spatiotemporal prediction results; and an energy storage control module, used to perform energy scheduling and energy storage control based on the spatiotemporal prediction results.

[0017] The technical solution provided in this application has at least the following technical effects or advantages: by realizing the technical goal of high-precision dynamic prediction and intelligent optimization scheduling of the power of distributed new energy clusters based on the spatiotemporal prediction model that integrates spatial structural features and time series features, the technical effects of improving the accuracy of power prediction, reducing energy scheduling deviation, and enhancing the rapid response capability of energy storage systems are achieved.

[0018] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the power prediction method for distributed renewable energy clusters used in this application.

[0021] Figure 2 This is a schematic diagram of the power prediction system for distributed new energy clusters used in this application.

[0022] Figure labeling: Cluster construction module 1, unit acquisition module 2, result acquisition module 3, energy storage control module 4. Detailed Implementation

[0023] This application provides a power prediction method and system for distributed renewable energy clusters, addressing the technical problem in existing technologies where the lack of a joint modeling mechanism for the spatial coupling characteristics and temporal evolution of distributed renewable energy clusters makes it difficult for power prediction models to accurately depict the dynamic correlations and mutual influences between different nodes, further affecting the overall power prediction accuracy, energy dispatch efficiency, and energy storage system response reliability. The application achieves the technical goal of high-precision dynamic prediction and intelligent optimized dispatch of power in distributed renewable energy clusters using a spatiotemporal prediction model based on the fusion of spatial structural features and time series features, thereby improving power prediction accuracy, reducing energy dispatch deviations, and enhancing the rapid response capability of energy storage systems.

[0024] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0025] Example 1, please refer to the appendix. Figure 1 This application provides a power prediction method for distributed renewable energy clusters, applied to a power prediction system for distributed renewable energy clusters, specifically including the following steps: S1: Based on the energy type, operating characteristics and composition structure of multiple new energy units in the distributed new energy cluster, construct a three-dimensional cluster of new energy units.

[0026] Furthermore, this application also includes: constructing an energy similarity granularity based on the energy type of the plurality of new energy units, dividing similar energy based on the energy similarity granularity, and obtaining an initial cluster of new energy units; constructing a power similarity granularity based on the operating characteristics and similar power output characteristics, dividing the initial cluster of new energy units based on the power similarity granularity, and obtaining a secondary cluster of new energy units; constructing a structural similarity granularity based on the composition structure, dividing the secondary cluster of new energy units based on the structural similarity granularity, and obtaining a tertiary cluster of new energy units.

[0027] Furthermore, this application also includes: the operating characteristics include the power output curve shape, power output fluctuation rate, equipment rated power, and meteorological response sensitivity.

[0028] Specifically, constructing an energy similarity granularity based on the energy type of multiple new energy units refers to classifying various new energy units, such as wind energy, photovoltaic energy, bioenergy, and geothermal energy, into categories with similar energy characteristics based on the similarity in energy source, energy conversion method, and operating environment. Here, energy type refers to different types of renewable energy, while energy similarity granularity represents the scale or fineness of the similarity classification. For example, with a coarser granularity, wind power and photovoltaics can be classified as the same type of clean power source; while with a finer granularity, they can be further distinguished based on wind speed response characteristics and light intensity variation patterns. This classification establishes an initial clustering basis for new energy units, providing a basis for subsequent more precise grouping.

[0029] Secondly, constructing power similarity granularity based on similar power output characteristics according to operational features refers to further subdividing the initial cluster based on the power output patterns of each new energy unit during operation. Operational features refer to a set of performance indicators reflecting the power generation capacity and response patterns of new energy units under different operating conditions, describing the dynamic behavior of the equipment under time and environmental changes. Among these, the power output curve shape refers to the trend line shape of the equipment's output power over a certain period of time. For example, photovoltaic power generation exhibits a parabolic shape with an increase in the morning, a stable state at noon, and a decrease in the evening, while wind power shows stronger random fluctuation characteristics. Output volatility refers to the proportion of the change in power output per unit time to the rated power, used to measure the stability of the new energy unit's output power. A lower output volatility indicates a more stable output, which is beneficial for grid dispatch and control, while a higher volatility indicates a need for more energy storage or regulation methods to maintain power balance. The rated power of the equipment refers to the maximum electrical power that the new energy equipment can stably output over a long period under standard operating conditions, representing the upper limit of the equipment's design performance. For example, a 500 kW rated photovoltaic inverter can continuously output 500 kW of power under optimal sunlight conditions. This indicator not only affects the power generation capacity of a single unit but also determines the overall system's capacity planning and load allocation strategy. Meteorological response sensitivity refers to the degree to which the output power of a renewable energy unit responds to changes in meteorological conditions, primarily related to factors such as light intensity, wind speed, temperature, and humidity. For instance, the output power of photovoltaic modules increases linearly with changes in light intensity, while the output power of wind turbines is proportional to the cube of the wind speed. Therefore, the meteorological response sensitivity of wind power systems is typically higher than that of photovoltaic systems. High sensitivity means faster output changes and greater sensitivity to meteorological fluctuations; low sensitivity indicates that the equipment can maintain relatively stable power output under different environmental conditions. This, in turn, helps identify equipment with similar dynamic output characteristics, forming a secondary cluster of renewable energy units.

[0030] Next, structural similarity granularity is constructed based on the composition structure, which refers to further classification from the perspective of hardware structure and physical connection relationships. The composition structure includes the unit's support structure, energy storage configuration, grid connection method, and control module composition, while the structural similarity granularity indicates the extent to which these structural differences are distinguished. For example, two photovoltaic arrays that use the same inverter model and electrical topology can be considered structurally similar; however, if one array is equipped with an independent energy storage unit, it will be classified into a different structurally similar cluster. Based on this classification, three clusters of new energy units can be obtained, thus ensuring that each cluster has a high degree of consistency in energy type, operating characteristics, and structural configuration.

[0031] S2: Based on the power data of the three clusters of the new energy unit, the importance is calculated, and a screening threshold is set according to the importance to select representative units.

[0032] Furthermore, this application also includes: calculating the mean of the power data of each of the three clusters of the new energy units to obtain a cluster power sequence; obtaining the total power data of the three clusters of the new energy units and calculating the Pearson correlation coefficient between the cluster power sequence and the total power data; calculating the inverse variance of the power data of each three cluster of the new energy units to obtain a stability coefficient; calculating the proportion of the power data of each three cluster of the new energy units based on the total power data to obtain a contribution rate coefficient; and performing a weighted summation of the Pearson correlation coefficient, the stability coefficient, and the contribution rate coefficient based on the coefficient of variation method to obtain the importance of each three cluster of the new energy units.

[0033] Specifically, averaging the power data of each of the three new energy units within the three-stage clusters refers to calculating the average power output data of all new energy units in the cluster over a certain time period. This average value reflects the overall power generation level of the cluster. Power data refers to the amount of electrical energy output per unit time, and averaging can eliminate the impact of short-term fluctuations, making the power change trend smoother. The average value is a point in time that constitutes the cluster's power sequence; multiple time points forming a time series can be used for trend analysis and predictive modeling.

[0034] Obtaining the total power data of the three clusters of renewable energy units refers to summing the power of all three clusters along the same time dimension to obtain the total output power of the entire distributed renewable energy system at that point in time. By calculating the Pearson correlation coefficient between the cluster power sequence and the total power data, the linear influence of a particular cluster on the overall power change can be measured. The Pearson correlation coefficient ranges from -1 to 1; the closer the value is to 1, the more consistent the trend of the cluster's power change with the overall system's change trend.

[0035] Calculating the inverse variance of the power data for each renewable energy unit's three clusters involves first calculating the variance of the power series to measure its volatility, and then taking its reciprocal as a stability coefficient. A smaller variance indicates more stable power changes, and a larger stability coefficient obtained after taking the reciprocal indicates more stable power output for the cluster. The stability coefficient helps identify clusters that maintain stable output under different meteorological conditions.

[0036] The power data of each renewable energy unit's three clusters is used to calculate the proportion of power in the total system power, reflecting the cluster's contribution to overall power generation. This is known as the contribution rate coefficient; a higher value indicates a larger share of power generation for that cluster in the system. This indicator can be used to measure the weight of different clusters in the overall operation, which is helpful for subsequent resource scheduling and priority allocation.

[0037] The weighted summation of Pearson correlation coefficient, stability coefficient, and contribution rate coefficient using the coefficient of variation method is a comprehensive evaluation method used to measure the importance of each renewable energy unit's tertiary cluster. The coefficient of variation method determines the weights by calculating the dispersion of each indicator, giving higher weights to indicators with smaller fluctuations and greater impact. Finally, a comprehensive score is obtained through weighted summation, representing the cluster's importance to power prediction and regulation. For example, when a cluster has a high correlation coefficient, strong stability, and a large contribution rate, its comprehensive score may be significantly higher than other clusters, thus being marked as a key cluster.

[0038] Setting a screening threshold based on importance refers to determining a critical value among the importance values ​​of all clusters to distinguish between highly representative and weakly representative new energy units. The screening threshold is typically set based on statistical distribution patterns, model performance requirements, or computational resource constraints. For example, the average importance value of all clusters can be used as the threshold, or the cumulative contribution rate reaching the importance threshold can be selected as the screening criterion. A higher screening threshold results in fewer representative units being selected, but with stronger representativeness; conversely, a lower screening threshold results in more units being selected, providing broader overall coverage.

[0039] Selecting representative units means retaining new energy units whose importance is higher than the threshold by comparing them with a threshold. These units serve as the core objects for subsequent spatiotemporal feature prediction and energy scheduling. This can significantly reduce data redundancy, improve computational efficiency, and ensure prediction accuracy.

[0040] S3: Based on the representative unit, perform spatiotemporal characteristic power prediction to obtain the spatiotemporal prediction result.

[0041] Furthermore, this application also includes: constructing a spatiotemporal correlation graph based on the representative unit as a node; extracting node feature vectors to divide the time window, and concatenating the divided feature vectors according to the time window to obtain a spatiotemporal feature matrix; performing power prediction on the spatiotemporal feature matrix to obtain a power prediction result; and mapping the power prediction result to the spatiotemporal correlation graph to obtain the spatiotemporal prediction result.

[0042] Furthermore, this application also includes: constructing a first distance matrix and a first power time-series correlation matrix; fusing the first distance matrix and the first power time-series correlation matrix using a threshold filtering method to obtain a first weighted adjacency matrix; obtaining a first node and a second node based on the nodes; and establishing a graph structure for the first node and the second node and the first edge between the first node and the second node based on the first weighted adjacency matrix to obtain the spatiotemporal correlation graph.

[0043] Furthermore, this application also includes: obtaining the spatial coordinates of the first node, the first power time series data, the spatial coordinates of the second node, and the second power time series data based on the first node and the second node; calculating a first distance matrix between the spatial coordinates of the first node and the spatial coordinates of the second node using Euclidean distance; and calculating the time series correlation based on the first power time series data and the second power time series data to obtain the first power time series correlation matrix.

[0044] Furthermore, this application also includes: constructing a spatiotemporal prediction model, wherein the spatiotemporal prediction model includes a spatial correlation prediction channel and a time dependence prediction channel; inputting the spatiotemporal feature matrix into the spatiotemporal prediction model, extracting power spatial correlation through the spatial correlation prediction channel, extracting power time dependence based on the time dependence prediction channel, and combining to obtain the power prediction result.

[0045] Furthermore, this application also includes: the spatiotemporal prediction model adopts a cascaded structure of a graph convolutional network and a long short-term memory network, wherein the graph convolutional network is used to extract spatial correlation features, and the long short-term memory network is used to extract temporal dependence features.

[0046] Specifically, based on representative units as nodes, a first node and a second node are randomly extracted. Obtaining the spatial coordinates of the first node, the first power time-series data, and the spatial coordinates and second power time-series data of the second node from the first and second nodes involves selecting any two representative units from the node set of the new energy cluster and extracting data on their spatial location and power output, respectively. Spatial coordinates refer to the fixed location of each node in geographic space. Power time-series data refers to the power output sequence recorded within a certain time interval. By extracting the spatial coordinates and power sequences from the two nodes respectively, a foundation is provided for subsequent spatial distance calculations and temporal correlation analysis.

[0047] The first distance matrix, calculated using Euclidean distance to determine the spatial coordinates of the first and second nodes, measures the spatial difference between two nodes using the Euclidean distance formula, thus forming a matrix describing the spatial proximity of all nodes. Euclidean distance is a geometric distance calculation method, obtained by taking the square root of the sum of the squares of the differences between two coordinate points. By calculating the distance between any two nodes in the cluster sequentially, a first distance matrix containing all node pairs can be obtained. The smaller the value of the matrix, the closer the spatial distance between the nodes, and the more likely their geographical locations are similar in terms of meteorological conditions.

[0048] The time-series correlation is calculated based on the first and second power time-series data to obtain the first power time-series correlation matrix. This matrix compares the power trends of two nodes over time to measure the degree of synchronization in their power output behavior. The time-series correlation is represented by the Pearson correlation coefficient, which ranges from -1 to 1. A value closer to 1 indicates more consistent power changes, while a value closer to 0 indicates no significant correlation in the trends. For example, when the power increases or decreases at the same time by roughly the same magnitude, the correlation coefficient may be above 0.9; if the directions of change are opposite, it may be negative. By calculating the correlation for all node pairs, a symmetrical first power time-series correlation matrix can be formed to reflect the power coupling relationship in the time dimension.

[0049] Furthermore, the first distance matrix and the first power temporal correlation matrix are fused using a threshold screening method to obtain the first weighted adjacency matrix. This refers to filtering and integrating the two types of matrices in both spatial and temporal dimensions by setting reasonable thresholds, thereby obtaining a weighted matrix that comprehensively reflects the degree of node correlation. The core of the threshold screening method is to limit a critical value. When the distance is too large or the power correlation is too low, the connection between two nodes is considered weak and they are eliminated; when the distance is close and the correlation is high, their connection is retained and assigned a higher weight. The fused first weighted adjacency matrix reflects both spatial proximity and retains the temporal characteristics of synchronous power changes.

[0050] Obtaining the first and second nodes based on nodes refers to selecting paired node combinations from all representative units in the new energy cluster. The selection of the first and second nodes usually follows the matrix index correspondence, that is, there exists a pair of nodes with non-zero weights in the first weighted adjacency matrix.

[0051] Based on the first weighted adjacency matrix, a graph structure is constructed for the first node, the second node, and the first edge between them, resulting in a spatiotemporal correlation graph. This involves transforming the node and edge relationships in the weighted adjacency matrix into a specific graph data structure. Each node corresponds to a new energy unit, and each edge represents the spatiotemporal connection between two units. The weight of each edge is derived from the numerical values ​​in the weighted adjacency matrix. The resulting spatiotemporal correlation graph visually reflects the coupling strength between different nodes in terms of geographical distance and power variation. For example, if the edge weight between nodes is large, it indicates that the two new energy units significantly influence each other in terms of power output and should be given priority in subsequent predictions.

[0052] Furthermore, extracting node feature vectors for time window segmentation refers to extracting features from the multidimensional data of each node, transforming information such as power, meteorology, and equipment operating parameters into computable feature vectors. The purpose of time window segmentation is to divide continuous time-series data into segments according to fixed time intervals, thereby capturing the dynamic trends of short-term power changes. The length of the time window is usually determined based on the time scale of the prediction target.

[0053] The spatiotemporal feature matrix is ​​obtained by concatenating the feature vectors of each node under different time windows after time segmentation. This involves sequentially concatenating the feature vectors of each node under different time windows to form a high-dimensional matrix. The rows of the matrix typically correspond to different nodes, and the columns correspond to time series features, thereby integrating spatial distribution and temporal evolution information.

[0054] Furthermore, the process of constructing a spatiotemporal prediction model aims to simultaneously capture the patterns of data variation across both spatial and temporal dimensions. A spatiotemporal prediction model is a predictive model capable of simultaneously analyzing the interplay between geographic spatial location and time series. It consists of a spatial correlation prediction channel and a temporal dependence prediction channel. The spatial correlation prediction channel analyzes the geographic or topological relationships between different nodes, while the temporal dependence prediction channel analyzes the power trend of the same node over time.

[0055] Furthermore, the spatiotemporal prediction model employs a cascaded structure of a graph convolutional network and a long short-term memory (LSTM) network. A graph convolutional network is a neural network specifically designed for processing graph-structured data, capable of effectively extracting spatial features when complex connections exist between nodes, and is used for modeling power transfer relationships between distributed renewable energy nodes. An LSM network, on the other hand, is an improved recurrent neural network structure with mechanisms such as forget gates, input gates, and output gates, enabling it to retain key temporal information over a longer time span, and is used to extract time-dependent features. The cascaded combination of these two networks forms a joint representation of spatiotemporal features.

[0056] Next, after inputting the constructed spatiotemporal feature matrix into the spatiotemporal prediction model, the spatial correlation prediction channel extracts the spatial power correlation between each node, describing the degree of mutual influence between power changes at different nodes. Simultaneously, the temporal dependence prediction channel extracts the power dependence of each node over time, i.e., the impact of power at one time moment on power at the next, such as the gradual increase in photovoltaic output during daytime when sunlight intensity increases. By combining spatiotemporal features, more accurate power prediction results can be obtained, enabling the spatiotemporal prediction model to simultaneously consider the dual constraints of spatial linkage and temporal dynamics.

[0057] Furthermore, mapping the power prediction results to a spatiotemporal correlation graph to obtain spatiotemporal prediction results means re-mapping the power prediction values ​​output by the spatiotemporal prediction model to each node of the spatiotemporal graph structure, so that the prediction results are consistent with the actual geographic spatial location and node relationships, which facilitates subsequent regional power assessment and energy storage scheduling.

[0058] S4: Energy scheduling and energy storage control are performed based on the spatiotemporal prediction results.

[0059] Specifically, energy dispatch and energy storage control based on spatiotemporal prediction results refer to coordinating and optimizing the generation, load distribution, and charging / discharging behavior of energy storage units in an energy system by utilizing the future power change trends output by prediction models. Spatiotemporal prediction results refer to power prediction data obtained by combining spatial distribution characteristics and temporal evolution characteristics, reflecting the output power changes of various distributed new energy units in different time periods in the future. Energy dispatch refers to dynamically allocating the operating power of generation units, energy storage devices, and loads based on the current and predicted energy supply and demand situation of the system to ensure supply and demand balance and improve energy utilization efficiency. Energy storage control, based on energy dispatch, formulates charging and discharging strategies for energy storage devices such as batteries, compressed air energy storage devices, or flywheel energy storage systems to achieve peak shaving and valley filling, improve system stability, and cope with power fluctuations.

[0060] In summary, the power prediction method for distributed renewable energy clusters provided in this application has the following technical effects: by achieving the technical goal of high-precision dynamic prediction and intelligent optimized scheduling of the power of distributed renewable energy clusters through a spatiotemporal prediction model based on the fusion of spatial structural features and time series features, the method improves the accuracy of power prediction, reduces energy scheduling deviation, and enhances the rapid response capability of energy storage systems.

[0061] Example 2: Based on the same inventive concept as the power prediction method for distributed renewable energy clusters in the foregoing examples, this application also provides a power prediction system for distributed renewable energy clusters. Please refer to the appendix. Figure 2 The system includes: a cluster construction module 1, used to construct a three-dimensional cluster of new energy units based on the energy type, operating characteristics, and composition structure of multiple new energy units in a distributed new energy cluster; a unit acquisition module 2, used to calculate the importance of the power data of the three-dimensional cluster of new energy units, and to select representative units based on the importance by setting a screening threshold; a result acquisition module 3, used to perform spatiotemporal characteristic power prediction based on the representative units to obtain spatiotemporal prediction results; and an energy storage control module 4, used to perform energy scheduling and energy storage control based on the spatiotemporal prediction results.

[0062] Furthermore, the power prediction system for distributed new energy clusters is also used for: constructing energy similarity granularity based on the energy types of the multiple new energy units, dividing similar energy based on the energy similarity granularity, and obtaining an initial cluster of new energy units; constructing power similarity granularity based on the operating characteristics and similar power output characteristics, dividing the initial cluster of new energy units based on the power similarity granularity, and obtaining a secondary cluster of new energy units; constructing structural similarity granularity based on the composition structure, dividing the secondary cluster of new energy units based on the structural similarity granularity, and obtaining a tertiary cluster of new energy units.

[0063] Furthermore, the power prediction system for distributed new energy clusters is also used for: the operating characteristics include the shape of the power output curve, the power output fluctuation rate, the rated power of the equipment, and the meteorological response sensitivity.

[0064] Furthermore, the power prediction system for distributed renewable energy clusters is also used for: calculating the mean of the power data of each renewable energy unit's three-stage cluster in the renewable energy unit to obtain a cluster power sequence; obtaining the total power data of the renewable energy unit's three-stage cluster and calculating the Pearson correlation coefficient between the cluster power sequence and the total power data; calculating the inverse variance of the power data of each renewable energy unit's three-stage cluster to obtain a stability coefficient; calculating the proportion of the power data of each renewable energy unit's three-stage cluster based on the total power data to obtain a contribution rate coefficient; and performing a weighted summation calculation of the Pearson correlation coefficient, the stability coefficient, and the contribution rate coefficient based on the coefficient of variation method to obtain the importance of each renewable energy unit's three-stage cluster in the renewable energy unit's three-stage cluster.

[0065] Furthermore, the power prediction system for distributed new energy clusters is also used for: constructing a spatiotemporal correlation graph based on the representative unit as a node; extracting node feature vectors and dividing them into time windows, then concatenating the divided feature vectors according to the time windows to obtain a spatiotemporal feature matrix; performing power prediction on the spatiotemporal feature matrix to obtain the power prediction result; and mapping the power prediction result to the spatiotemporal correlation graph to obtain the spatiotemporal prediction result.

[0066] Furthermore, the power prediction system for distributed new energy clusters is also used for: constructing a first distance matrix and a first power time-series correlation matrix; fusing the first distance matrix and the first power time-series correlation matrix through a threshold filtering method to obtain a first weighted adjacency matrix; obtaining a first node and a second node based on the nodes; and establishing a graph structure for the first node and the second node and the first edge between the first node and the second node based on the first weighted adjacency matrix to obtain the spatiotemporal correlation graph.

[0067] Furthermore, the power prediction system for distributed new energy clusters is also used to: obtain the spatial coordinates of the first node, the first power time series data, the spatial coordinates of the second node, and the second power time series data based on the first node and the second node; calculate a first distance matrix between the spatial coordinates of the first node and the spatial coordinates of the second node using Euclidean distance; and calculate the time series correlation based on the first power time series data and the second power time series data to obtain the first power time series correlation matrix.

[0068] Furthermore, the power prediction system for distributed new energy clusters is also used to: construct a spatiotemporal prediction model, wherein the spatiotemporal prediction model includes a spatial correlation prediction channel and a time dependence prediction channel; input the spatiotemporal feature matrix into the spatiotemporal prediction model, extract power spatial correlation through the spatial correlation prediction channel, extract power time dependence based on the time dependence prediction channel, and combine to obtain the power prediction result.

[0069] Furthermore, the power prediction system for distributed new energy clusters is also used in the following way: the spatiotemporal prediction model adopts a cascaded structure of a graph convolutional network and a long short-term memory network, wherein the graph convolutional network is used to extract spatial correlation features, and the long short-term memory network is used to extract temporal dependence features.

[0070] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The power prediction method and specific examples for distributed new energy clusters in the foregoing embodiment one are also applicable to the power prediction system for distributed new energy clusters in this embodiment. Through the foregoing detailed description of the power prediction method for distributed new energy clusters, those skilled in the art can clearly understand the power prediction system for distributed new energy clusters in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0071] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0072] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A power prediction method for distributed renewable energy clusters, characterized in that, include: Based on the energy types, operational characteristics, and composition of multiple new energy units in a distributed new energy cluster, a three-dimensional cluster of new energy units is constructed. Importance is calculated based on the power data of the three clusters of the new energy unit, and representative units are obtained by setting a screening threshold according to the importance. Spatiotemporal characteristic power prediction is performed based on the representative unit to obtain the spatiotemporal prediction result; Energy scheduling and energy storage control are performed based on the spatiotemporal prediction results.

2. The power prediction method for distributed new energy clusters as described in claim 1, characterized in that, Constructing a three-tiered cluster for new energy units, including: Based on the energy types of the multiple new energy units, an energy similarity granularity is constructed, and similar energy is divided based on the energy similarity granularity to obtain an initial cluster of new energy units; Based on the operational characteristics, a power similarity granularity is constructed for similar power output characteristics. Based on the power similarity granularity, the initial cluster of the new energy unit is divided to obtain a secondary cluster of the new energy unit. Based on the composition structure, a structural similarity granularity is constructed, and the new energy unit secondary cluster is divided based on the structural similarity granularity to obtain the new energy unit tertiary cluster.

3. The power prediction method for distributed new energy clusters as described in claim 1, characterized in that, The operating characteristics include the shape of the power output curve, the power output fluctuation rate, the rated power of the equipment, and the sensitivity to weather response.

4. The power prediction method for distributed new energy clusters as described in claim 1, characterized in that, Importance calculations are performed based on the power data of the three clusters of the new energy unit, including: The power data of each of the three clusters of the new energy unit are averaged to obtain the cluster power sequence. Obtain the total power data of the three clusters of the new energy unit, and calculate the Pearson correlation coefficient between the cluster power sequence and the total power data; The stability coefficient is obtained by calculating the inverse variance of the power data of each new energy unit's cubic cluster. The contribution rate coefficient is obtained by calculating the proportion of the power data of each new energy unit's three clusters based on the total power data; The importance of each new energy unit tertiary cluster in the new energy unit tertiary cluster is obtained by weighted summation of the Pearson correlation coefficient, the stability coefficient and the contribution rate coefficient based on the coefficient of variation method.

5. The power prediction method for distributed new energy clusters as described in claim 1, characterized in that, The spatiotemporal prediction results obtained include: Based on the representative units as nodes, a spatiotemporal relationship graph is constructed; Extract node feature vectors and divide them into time windows. Then, concatenate the divided feature vectors according to the time windows to obtain a spatiotemporal feature matrix. Power prediction is performed on the spatiotemporal feature matrix to obtain the power prediction result; The power prediction results are mapped to the spatiotemporal correlation graph to obtain the spatiotemporal prediction results.

6. The power prediction method for distributed new energy clusters as described in claim 5, characterized in that, Constructing a spatiotemporal relationship graph includes: Construct the first distance matrix and the first power time-series correlation matrix; The first distance matrix and the first power time-series correlation matrix are fused using a threshold filtering method to obtain the first weighted adjacency matrix; Obtain the first node and the second node based on the node; Based on the first weighted adjacency matrix, a graph structure is established for the first node and the second node, and the first edge between the first node and the second node, to obtain the spatiotemporal association graph.

7. The power prediction method for distributed new energy clusters as described in claim 6, characterized in that, Construct the first distance matrix and the first power time-series correlation matrix, including: The spatial coordinates of the first node, the first power timing data, the spatial coordinates of the second node, and the second power timing data are obtained based on the first node and the second node. Calculate the first distance matrix between the spatial coordinates of the first node and the spatial coordinates of the second node using Euclidean distance; The timing correlation is calculated based on the first power timing data and the second power timing data to obtain the first power timing correlation matrix.

8. The power prediction method for distributed new energy clusters as described in claim 5, characterized in that, Obtain power prediction results, including: Construct a spatiotemporal prediction model, wherein the spatiotemporal prediction model includes a spatial correlation prediction channel and a time dependence prediction channel; The spatiotemporal feature matrix is ​​input into the spatiotemporal prediction model. Power spatial correlation is extracted through the spatial correlation prediction channel, and power time dependence is extracted based on the time dependence prediction channel. The power prediction results are obtained by combining these methods.

9. The power prediction method for distributed new energy clusters as described in claim 8, characterized in that, The spatiotemporal prediction model adopts a cascaded structure of a graph convolutional network and a long short-term memory network, wherein the graph convolutional network is used to extract spatial correlation features, and the long short-term memory network is used to extract temporal dependence features.

10. A power prediction system for distributed renewable energy clusters, characterized in that, The steps for implementing the power prediction method for distributed renewable energy clusters according to any one of claims 1 to 9 include: The cluster construction module is used to construct a three-dimensional cluster of new energy units based on the energy type, operating characteristics and composition structure of multiple new energy units in a distributed new energy cluster. The unit acquisition module is used to calculate the importance based on the power data of the three clusters of the new energy unit, and to select representative units by setting a screening threshold according to the importance. The result acquisition module is used to perform spatiotemporal characteristic power prediction based on the representative unit to obtain spatiotemporal prediction results. An energy storage control module is used to perform energy scheduling and energy storage control based on the spatiotemporal prediction results.

Images