Wind power cluster regulation and control method, system and equipment based on distributed computing and medium

Abstract

The invention discloses a wind power cluster regulation and control method, system and device based on distributed calculation and a medium, and belongs to the technical field of wind power cluster regulation and control, and the method comprises the steps: obtaining the historical wind speed and wind direction data of each wind driven generator in a wind power cluster; calculating a wind condition correlation coefficient between any two wind driven generators; constructing a graph model of a wind power cluster; based on the clustering number, clustering wind power generators in the wind power cluster to obtain a plurality of wind power groups; on the basis of a wind power set, an optimization objective function is constructed in combination with resource constraint conditions of the distributed computing nodes, and a distribution strategy of the distributed computing nodes is determined through an optimization algorithm; and operation regulation and control of the wind power cluster are realized. According to the method, the wind regime correlation coefficient is introduced to construct the graph model, the wind regime correlation degree between the wind driven generators can be accurately reflected, and the physical reasonability of clustering division is improved. The clustering number is determined through graph feature similarity matching, and the problem that parameter setting depends on experience in a traditional clustering method is avoided.

Description

Technical Field

[0001] This invention relates to the field of wind power cluster control technology, specifically to a wind power cluster control method, system, equipment, and medium based on distributed computing. Background Technology

[0002] Traditional wind farms typically employ a centralized architecture for control and data processing, where a central node collects operational information from each wind turbine and issues control commands to complete scheduling tasks.

[0003] In the existing centralized control architecture, data from all wind turbines must be aggregated at a central node for processing. When the wind farm is large, the central node needs to process a large amount of real-time data, which can easily lead to excessive computational pressure and control delays. Furthermore, if the central node fails, it will affect the stability of the entire wind power cluster's control. In addition, some current control methods do not adequately analyze the spatial differences in wind conditions within the wind farm, and cannot establish effective correlation models for wind speed and direction variations among wind turbines, thus limiting the fine-grained scheduling capabilities of the wind power cluster.

[0004] Some existing technologies attempt to improve computing efficiency through distributed computing structures, but during the node deployment process, there is a lack of modeling and analysis of the wind conditions inside the wind farm, and the wind condition correlation between wind turbines is not fully explored, resulting in uneven distribution of distributed computing node resources and affecting the overall control effect.

[0005] Existing distribution strategies typically rely on preset rules or average allocation methods, failing to adjust in real time according to dynamic changes in wind conditions and differences in the computational needs of each wind turbine, making it difficult to meet the resource load balancing and real-time requirements during the operation of wind power clusters. Summary of the Invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problems solved by this invention are: the lack of modeling methods for the correlation of wind conditions among wind turbines within a wind farm, the inability to quantify the impact of wind condition correlation on group control strategies, resulting in unreasonable wind power cluster division, coarse control granularity, and difficulty in adapting to complex wind condition changes; the clustering process usually relies on empirical setting of the number of clusters or static clustering methods, failing to dynamically determine a reasonable number of clusters in combination with the wind condition network structure, resulting in a lack of targeted cluster structure division and reduced subsequent scheduling efficiency; the allocation strategy of distributed computing nodes lacks systematic optimization methods, and the consideration of computing resource distribution, transmission distance, and node load balancing is not comprehensive, which can easily lead to computing bottlenecks, resource idleness, or system instability.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a wind power cluster control method based on distributed computing, comprising,

[0009] Obtain historical wind speed and direction data for each wind turbine in the wind power cluster;

[0010] Based on historical wind speed and direction data, calculate the wind condition correlation coefficient between any two wind turbines;

[0011] Based on the calculated wind condition correlation coefficient, a graph model of the wind power cluster is constructed. In the graph model of the wind power cluster, each wind turbine is a node in the graph model. An edge is established between any two nodes when the corresponding wind condition correlation coefficient is greater than a set threshold. The wind condition correlation coefficient is used as the weight of the edge to represent the wind condition relationship between the wind turbines.

[0012] Based on the graph model, graph feature parameters representing the graph structure are extracted and similarity matching is performed with the graph feature parameters of multiple sample graph models to determine the sample graph model most similar to the current graph model. The number of clusters corresponding to the sample graph model is used as the number of clusters of the current wind power cluster.

[0013] Based on the number of clusters, the wind turbines in the wind power cluster are clustered to obtain multiple wind power groups;

[0014] Based on wind turbines and considering the resource constraints of distributed computing nodes, an optimization objective function is constructed, and the distribution strategy of distributed computing nodes is determined through optimization algorithms.

[0015] Based on the distributed computing node distribution strategy and combined with the real-time operating status of the wind turbines, the node allocation is dynamically adjusted to achieve operation control of the wind power cluster.

[0016] As a preferred embodiment of the wind power cluster control method based on distributed computing of the present invention, wherein: the calculation of the wind condition correlation coefficient between any two wind turbines includes extracting the wind speed sequence and wind direction sequence under the corresponding timestamp;

[0017] The wind speed correlation coefficient between the wind speed sequences and the wind direction correlation coefficient between the wind direction sequences were calculated using the Pearson correlation coefficient, respectively.

[0018] Calculate the standard deviation of the wind speed sequence and the standard deviation of the wind direction sequence for each of the two wind turbines.

[0019] Based on the standard deviation of wind speed and the standard deviation of wind direction, a wind speed weighting factor and a wind direction weighting factor are determined. The wind speed weighting factor is the ratio of the standard deviation of wind speed to the sum of the standard deviations of wind speed and wind direction, and the wind direction weighting factor is the ratio of the standard deviation of wind direction to the sum of the standard deviations of wind speed and wind direction.

[0020] After taking the absolute values ​​of the wind speed correlation coefficient and the wind direction correlation coefficient, the wind condition correlation coefficient is obtained by weighting and summing them according to the wind speed weighting factor and the wind direction weighting factor.

[0021] As a preferred embodiment of the wind power cluster control method based on distributed computing of the present invention, wherein: the construction of the graph model of the wind power cluster includes representing each wind turbine in the wind power cluster as a node in the graph model;

[0022] When the wind condition correlation coefficient between any two wind turbines is greater than a set threshold, an edge is established between the nodes representing the two wind turbines, and the weight of the edge is set to the wind condition correlation coefficient between the two wind turbines.

[0023] As a preferred embodiment of the wind power cluster control method based on distributed computing of the present invention, wherein: the step of taking the cluster number corresponding to the sample graph model as the cluster number of the current wind power cluster includes extracting multiple graph feature parameters for characterizing the graph structure characteristics based on the current wind power cluster graph model;

[0024] The extracted graph feature parameters are compared with the graph feature parameters of multiple preset sample graph models to measure their similarity. The similarity measurement is calculated based on the Euclidean distance between the graph feature parameters.

[0025] The sample graph model with the minimum Euclidean distance is identified, and the number of clusters corresponding to the sample graph model with the minimum Euclidean distance is determined as the number of clusters of the current wind power cluster.

[0026] As a preferred embodiment of the wind power cluster control method based on distributed computing of the present invention, wherein: the clustering of wind turbines in the wind power cluster includes, based on the wind condition correlation coefficient, selecting a number of wind turbines whose wind condition correlation coefficients between each pair are less than a first threshold as initial cluster centers.

[0027] For each wind turbine that is not a cluster center, calculate the wind condition correlation coefficient with each cluster center and assign it to the cluster group corresponding to the cluster center with the smallest distance;

[0028] For each cluster, calculate the average wind condition correlation coefficient between all wind turbines in the cluster and other wind turbines in the cluster, and select the wind turbine with the smallest average distance as the new cluster center.

[0029] Repeat the wind turbine allocation and cluster center update operation until the cluster centers no longer change;

[0030] When a cluster does not contain any wind turbines, a wind turbine is randomly selected from the assigned wind turbine clusters and reassigned to the empty cluster.

[0031] As a preferred embodiment of the wind power cluster control method based on distributed computing of the present invention, wherein: the construction of the optimization objective function and the determination of the distribution strategy of the distributed computing nodes through the optimization algorithm include obtaining historical wind condition data of the wind power cluster, wherein the historical wind condition data of the wind power cluster includes wind condition characteristics of the locations of multiple wind turbines in multiple historical time periods.

[0032] Based on historical wind data from wind power clusters, multiple wind scenario scenarios are constructed.

[0033] Generate multiple initial computing node distribution strategies that satisfy resource allocation constraints;

[0034] For each wind condition scenario, the resource utilization, task latency, and computing load balancing metrics of each initial computing node distribution strategy are calculated using the fitness function, thus obtaining the fitness value of each initial computing node distribution strategy in the wind condition scenario.

[0035] Determine the weighting coefficients for each wind condition scenario;

[0036] For each initial computing node distribution strategy, the fitness values ​​of the initial computing node distribution strategy in each wind condition scenario are weighted and summed based on the weighting coefficients corresponding to each wind condition scenario to obtain the comprehensive fitness value of the initial computing node distribution strategy.

[0037] Using the comprehensive fitness value as the optimization objective, a strategy search is performed based on the particle swarm optimization algorithm to determine the distribution strategy of the distributed computing nodes.

[0038] As a preferred embodiment of the wind power cluster control method based on distributed computing of the present invention, the dynamic adjustment of node allocation to realize the operation control of the wind power cluster includes extracting the current wind condition parameters based on the real-time wind speed and wind direction data of the wind turbines.

[0039] Input wind condition parameters into the computing power prediction model to predict the computing resource requirements of each wind turbine.

[0040] Based on the prediction results, a fitness function is constructed by combining the constraints of the number of computing nodes, node load limits, and scheduling response time. The fitness function aims to maximize resource utilization, minimize task latency, achieve balanced node load distribution, and minimize node redistribution time.

[0041] When the computing power requirement of the wind turbine changes and the triggering condition is met, the computing node allocation strategy is updated based on the fitness function.

[0042] Based on the updated allocation strategy, wind power cluster operation and control operations are executed, including implementing speed control strategy, pitch angle control strategy, power smoothing control strategy, and grid frequency response control strategy under different wind speed ranges.

[0043] This invention provides a wind power cluster control system based on distributed computing.

[0044] To solve the above technical problems, the present invention provides the following technical solution: a wind power cluster control system based on distributed computing, comprising: a data acquisition module, used to acquire historical wind speed and wind direction data of each wind turbine in the wind power cluster;

[0045] The correlation calculation module is used to calculate the wind condition correlation coefficient between any two wind turbines based on the historical wind speed and wind direction data.

[0046] The graph model construction module is used to construct a graph model of the wind power cluster based on the wind condition correlation coefficient. In the graph model, each wind turbine is a node in the graph model. An edge is established between any two nodes when the corresponding wind condition correlation coefficient is greater than a set threshold, and the wind condition correlation coefficient is used as the weight of the edge.

[0047] The cluster number determination module is used to extract graph feature parameters based on the graph model, perform similarity matching with the graph feature parameters of multiple sample graph models, and determine the cluster number of the current wind power cluster.

[0048] The clustering module is used to cluster the wind turbines in the wind power cluster based on the clustering number to obtain multiple wind power groups.

[0049] The distribution strategy determination module is used to construct an optimization objective function based on the wind turbine and the resource constraints of the distributed computing nodes, and to determine the distribution strategy of the distributed computing nodes through an optimization algorithm.

[0050] The control and execution module is used to dynamically adjust the node allocation based on the distributed computing node distribution strategy and the real-time operating status of the wind turbine, and to execute the operation control operations of the wind power cluster.

[0051] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of a wind power cluster control method based on distributed computing.

[0052] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of a wind power cluster control method based on distributed computing.

[0053] The beneficial effects of this invention are as follows: By introducing wind condition correlation coefficients to construct a graph model, this invention can accurately reflect the degree of wind condition correlation between wind turbines, improving the physical rationality of clustering. Using graph feature similarity matching to determine the number of clusters avoids the problem of parameter setting relying on experience in traditional clustering methods, enhancing the method's adaptability. Introducing wind condition scenario modeling and multi-objective optimization strategies into distributed computing resource allocation can improve computing power utilization and allocation balance while ensuring resource constraints. Combining real-time wind condition information with dynamic adjustment of control strategies enhances the system's ability to cope with sudden wind speed changes and load fluctuations, contributing to the refined and highly reliable control of wind power clusters. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 The above is a flowchart of a wind power cluster control method based on distributed computing, which is provided as an embodiment of the present invention.

[0056] Figure 2 This is a schematic diagram of a graph model of a wind power cluster control method based on distributed computing, provided as an embodiment of the present invention.

[0057] Figure 3 This is a computer equipment diagram of a wind power cluster control method based on distributed computing, provided as an embodiment of the present invention. Detailed Implementation

[0058] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0059] Example 1, referring to Figures 1-2 This is one embodiment of the present invention, which provides a wind power cluster control method based on distributed computing, including:

[0060] Step 1: Obtain historical wind speed and direction data for each wind turbine in the wind power cluster;

[0061] In step 1, the historical operating data of the wind power cluster can include wind condition information for each wind turbine at multiple historical time points, such as wind speed and wind direction.

[0062] Specifically, this includes: wind speed data: instantaneous wind speed (unit: m / s) measured at the blades of each wind turbine, collected by an anemometer on the top of the wind turbine nacelle.

[0063] Wind direction data: The wind direction angle directly in front of the wind turbine (unit: °, range 0°~360°, north is 0°, increasing clockwise), measured by a wind vane.

[0064] Timestamp: The precise acquisition time of each data point (must be synchronized to a unified time zone, such as UTC+8) to ensure time alignment of data from multiple wind turbines.

[0065] Data frequency: Select the sampling interval (e.g., 1 minute, 10 minutes or 1 hour) according to the analysis requirements.

[0066] The collected historical data is shown in Table 1.

[0067] Table 1 Data Collection Details

[0068] Wind Turbine ID Timestamp Wind speed (m / s) Wind direction (°) WT001 2023-01-0100:00 5.2 30 WT002 2023-01-0100:10 6.1 35 WT003 2023-01-0100:00 4.8 28 WT004 2023-01-0100:10 5.5 32

[0069] Furthermore, to ensure the accuracy of subsequent correlation analysis, the raw data needs to be preprocessed, specifically including the following steps:

[0070] Missing value handling includes: Deletion: If the missing percentage is low (e.g., <5%), directly delete the data for the missing time points. Interpolation: Use linear interpolation, spline interpolation, or spatial interpolation based on adjacent wind turbine data to fill in missing values. Model prediction: Use historical data to train a regression model (e.g., LSTM) to predict missing values.

[0071] Outlier correction includes: physical threshold filtering: removing invalid data with wind speed > 50 m / s (overcut-off wind speed) or wind direction > 360°. Statistical detection: identifying outliers based on the 3σ principle or box plots and replacing them with the median or mean.

[0072] Time synchronization includes: unifying the timestamps of all wind turbines to the same time zone, and correcting time misalignments caused by network latency or sensor clock deviations (such as using the NTP protocol for synchronization).

[0073] Data denoising includes applying moving averages or low-pass filters to high-frequency wind speed data to eliminate sensor noise (such as short-term fluctuations caused by turbulence).

[0074] Step 2: Based on historical wind speed and direction data, calculate the wind condition correlation coefficient between any two wind turbines;

[0075] In step 2, calculating the wind condition correlation coefficient between any two wind turbines includes extracting the wind speed sequence and wind direction sequence at the corresponding timestamp;

[0076] The wind speed correlation coefficient between the wind speed sequences and the wind direction correlation coefficient between the wind direction sequences were calculated using the Pearson correlation coefficient, respectively.

[0077] Calculate the standard deviation of the wind speed sequence and the standard deviation of the wind direction sequence for each of the two wind turbines.

[0078] Based on the standard deviation of wind speed and the standard deviation of wind direction, a wind speed weighting factor and a wind direction weighting factor are determined. The wind speed weighting factor is the ratio of the standard deviation of wind speed to the sum of the standard deviations of wind speed and wind direction, and the wind direction weighting factor is the ratio of the standard deviation of wind direction to the sum of the standard deviations of wind speed and wind direction.

[0079] After taking the absolute values ​​of the wind speed correlation coefficient and the wind direction correlation coefficient, the wind condition correlation coefficient is obtained by weighting and summing them according to the wind speed weighting factor and the wind direction weighting factor.

[0080] Specifically, for any two wind turbines, the wind speed correlation coefficient is calculated based on the wind speeds at multiple historical time points, according to the correlation coefficient calculation formula (e.g., Pearson correlation coefficient, Spearman rank correlation coefficient, etc.). Similarly, the wind direction correlation coefficient is calculated based on the wind direction at multiple historical time points, according to the correlation coefficient calculation formula. The absolute values ​​of the wind speed and wind direction correlation coefficients are then taken and weighted summed to obtain the wind condition correlation coefficient between the two wind turbines. Assume the data for the two wind turbines at five time points are shown in Table 2.

[0081] Table 2 Comparison of Fan Data

[0082]

[0083] Specifically, the method in this embodiment calculates the wind speed correlation coefficient in Table 2 using the Pearson correlation coefficient method, which is expressed as follows:

[0084] r speed ≈0.988

[0085] Where, r speed This represents the wind speed correlation coefficient calculated using the Pearson coefficient.

[0086] By substituting the wind speeds of the two wind turbines as variables X and Y into the formula for calculating the Pearson correlation coefficient, the wind speed correlation coefficient between the two wind turbines can be obtained.

[0087] Calculate the wind direction correlation coefficient (angle difference conversion):

[0088] r dir =0.9918

[0089] Where, r dir This represents the wind direction correlation coefficient calculated using the Pearson coefficient.

[0090] By substituting the wind directions of the two wind turbines as X and Y variables into the formula for calculating the Pearson correlation coefficient, the wind direction correlation coefficient between the two wind turbines can be obtained.

[0091] The correlation coefficient between the wind conditions of the two wind turbines is: weighted summation (weight: w) speed =0.8, W dir =0.2)

[0092] R wind =0.8×0.988+0.2×0.9918≈0.98876

[0093] Based on the fluctuations in wind speed and direction of the wind power cluster, the aforementioned weights are determined. For example, the standard deviations of wind speed and wind direction at multiple historical time points are calculated, and the sum of these standard deviations is used as w. speed The ratio of the standard deviation of wind direction to the sum of the standard deviations of wind speed and wind direction is taken as W. dir .

[0094] Step 3: Based on the calculated wind condition correlation coefficient, construct a graph model of the wind power cluster. In the graph model of the wind power cluster, each wind turbine is a node. An edge is established between any two nodes when the corresponding wind condition correlation coefficient is greater than a set first threshold (e.g., 0.5). The wind condition correlation coefficient is used as the weight of the edge to represent the wind condition association between the wind turbines. As an example, a histogram or cumulative distribution function of the correlation coefficient can be plotted based on the corresponding wind condition correlation coefficient between any two nodes. The first threshold is selected based on the distribution characteristics. For example, if the correlation coefficient is concentrated in the range of 0.2 to 0.8, and the area around 0.5 is a dense distribution area, 0.5 can be selected as the first threshold to distinguish between strong and weak associations.

[0095] In step 3, the construction of the graph model of the wind power cluster includes representing each wind turbine in the wind power cluster as a node in the graph model;

[0096] When the wind condition correlation coefficient between any two wind turbines is greater than a set threshold, an edge is established between the nodes representing the two wind turbines, and the weight of the edge is set to the wind condition correlation coefficient between the two wind turbines.

[0097] Step 4: Based on the graph model, extract the graph feature parameters that represent the graph structure, and perform similarity matching with the graph feature parameters of multiple sample graph models to determine the sample graph model that is most similar to the current graph model. The number of clusters corresponding to the sample graph model is then used as the number of clusters for the current wind power cluster.

[0098] In step 4, the step of using the number of clusters corresponding to the sample graph model as the number of clusters of the current wind power cluster includes extracting multiple graph feature parameters to characterize the graph structure characteristics based on the current wind power cluster graph model.

[0099] The extracted graph feature parameters are compared with the graph feature parameters of multiple preset sample graph models to measure their similarity. The similarity measurement is calculated based on the Euclidean distance between the graph feature parameters.

[0100] The sample graph model with the minimum Euclidean distance is identified, and the number of clusters corresponding to the sample graph model with the minimum Euclidean distance is determined as the number of clusters of the current wind power cluster.

[0101] Step 5: Based on the cluster count, cluster the wind turbines in the wind power cluster to obtain multiple wind power groups;

[0102] In step 5, the clustering of wind turbines in the wind power cluster includes selecting several wind turbines with wind condition correlation coefficients less than a second threshold (e.g., 0.3, 0.4, etc.) as initial cluster centers based on wind condition correlation coefficients. The number of initial cluster centers is the number of clusters. As an example, a histogram or cumulative distribution function of the correlation coefficients can be plotted based on the corresponding wind condition correlation coefficients between any two nodes. The second threshold can be selected based on the distribution characteristics. For example, if the proportion of correlation coefficients below 0.3 is less than 5% and the distribution drops sharply at this point, 0.3 can be selected as the second threshold. More preferably, for each wind turbine, the standard deviation of the wind condition correlation coefficient between the wind turbine and any other wind turbine can be calculated as the standard deviation of the correlation coefficient corresponding to the wind turbine. The wind turbines in the wind power cluster are sorted from largest to smallest according to the standard deviation of their correlation coefficients. The first-ranked wind turbine is used as an initial cluster center. The second-ranked wind turbine is then selected, and its correlation coefficient with each initial cluster center is checked against a first threshold. If the correlation coefficient is less than a first threshold, it is selected as an initial cluster center. Otherwise, the third-ranked wind turbine is selected, and the correlation coefficient with each initial cluster center is checked against the first threshold. This process is repeated until the number of initial cluster centers equals the number of clusters. This approach selects wind turbines with larger correlation coefficient standard deviations as initial centers to ensure that wind turbines with significant differences in correlation are selected, preventing over-concentration of cluster centers and improving the diversity of clustering results. The threshold constraint (correlation coefficient < first threshold) ensures the wind condition independence between initial cluster centers, reducing the risk of similar or related wind turbines being repeatedly selected as centers, thereby optimizing clustering efficiency.

[0103] For each wind turbine that is not a cluster center, calculate the wind condition correlation coefficient with each cluster center and assign it to the cluster group corresponding to the cluster center with the smallest wind condition correlation coefficient;

[0104] For each cluster, calculate the average wind condition correlation coefficient between all wind turbines in the cluster and other wind turbines in the cluster, and select the wind turbine with the smallest average wind condition correlation coefficient as the new cluster center.

[0105] Repeat the wind turbine allocation and cluster center update operations until the cluster centers no longer change.

[0106] Multiple sample graph models are obtained, each corresponding to a sample wind power cluster. The optimal K value for the sample wind power cluster can be determined using the silhouette coefficient. Specifically, a range of K values ​​is set (e.g., [2, 10]). A K value is sampled from this range, and the K-MEANS algorithm is used to cluster the wind power cluster based on the wind condition correlation coefficient between any two wind turbines. The clustering results are then calculated, and the global silhouette coefficient of the sampled K value is calculated. The global silhouette coefficients of the multiple sampled K values ​​are compared, and the sampled K value with the largest global silhouette coefficient is taken as the optimal K value for the sample wind power cluster, which is also the optimal K value for the sample graph model.

[0107] Determine various candidate graph model features, such as: number of nodes: the total number of nodes in the graph; number of edges: the total number of edges in the graph; average degree Q: the average number of connections between nodes; number of connected components: the number of non-connected subgraphs in the graph; maximum connected component size: the proportion of nodes in the largest subgraph; graph density D: the ratio of the actual number of edges to the total number of edges in the complete graph.

[0108] Key graph model features are selected from multiple candidate graph model features. For each candidate graph model feature, the feature value of the candidate graph model feature in each sample graph model and the optimal K value corresponding to the sample graph model are used as X and Y variables, respectively, and substituted into the formula for calculating the correlation coefficient (e.g., Pearson correlation coefficient, Spearman rank correlation coefficient, etc.). The correlation coefficient between the candidate graph model feature and the K value is calculated, and the candidate graph model features whose absolute value of the correlation coefficient is greater than the absolute value threshold (e.g., 0.5) are selected as key graph model features. As an example only, a histogram or cumulative distribution function of the correlation coefficient between the candidate graph model features and the K value can be plotted. The absolute value threshold can be selected according to the distribution characteristics. For example, if the correlation coefficient is concentrated in the range of 0.2 to 0.8, and the area around 0.5 is a dense distribution area, 0.5 can be selected as the absolute value threshold to distinguish between strong and weak associations.

[0109] For example, in 30 days of data from a wind farm: the correlation coefficient rQ between the average degree Q and the K value is 0.72 (strong positive correlation); the correlation coefficient rD between the graph density D and the K value is -0.45 (weak negative correlation). Screening results: Q is retained as the key feature, and D is removed. Average degree is the average number of connections between nodes, i.e., the average number of edges for each node in the graph model. Graph density is the ratio of the actual number of edges to the total number of edges in the complete graph, where the total number of edges in the complete graph = n(n-1) / 2, and n is the number of nodes in the graph model.

[0110] Based on key graph model features, key graph model features of the graph model and key graph model features of the sample graph model are extracted. The key graph model features can include the feature values ​​of each key graph model feature.

[0111] The K value is determined based on the key graph model features of the graph model and the key graph model features of the sample graph model.

[0112] For example, the Euclidean distance between the key graph features of the graph model and the key graph features of the sample graph models is calculated. The sample graph model with the smallest Euclidean distance is taken as the similar sample graph model, and the optimal K value of the similar sample graph model is taken as the K value of the current wind power cluster.

[0113] As an example, the Euclidean distance between the key graph features of the graph model and the key graph features of the sample graph model can be calculated as follows:

[0114] For the key feature vector f of the current graph current =[f1,f2,…,f n ] and the key feature vector f of the sample image sample =[f1 ' f2 ' ,…,f n ' [, Euclidean distance is:

[0115]

[0116] Among them, f i f is the feature value of the i-th key graph model feature corresponding to the current graph model. i ' Let be the feature value of the i-th key graph model feature corresponding to the sample graph model, and n be the total number of key graph model features.

[0117] Understandably, traditional methods (such as modularity optimization and profile coefficient) require recalculating the optimal K value for each new graph model, resulting in high computational complexity. By directly reusing the K value of similar samples, the computation time is reduced from minutes to milliseconds (e.g., matching time for a 90-day sample library is <10ms), supporting real-time wind farm cluster partitioning. The optimal K value of similar samples has been validated using historical data (e.g., maximizing modularity or optimizing the profile coefficient), ensuring the practical rationality of the reused K value. Wind farm conditions exhibit daily periodicity and seasonal variations. This method automatically adapts to wind condition evolution by dynamically updating the sample library (e.g., retaining data from the most recent 30 days), avoiding interference from outdated samples. When the distance between the current graph and all samples exceeds a threshold (e.g., 1.5 times the average distance), a recalculation of the optimal K value is triggered. For multiple samples with close proximity, a weighted average is used and rounded to obtain the K value, reducing the bias of a single sample (e.g., samples A and B have K values ​​of 4 and 5 respectively, with a distance of 0.05, resulting in a weighted average of 4.5; after rounding, the current K = 4).

[0118] As a preferred approach, weights are automatically assigned based on the correlation between key graph model features and the optimal K value (e.g., the average degree Q is given a higher weight), and Euclidean distance is calculated to highlight features that have a greater impact on clustering. For example, the correlation coefficient between key graph model features and the optimal K value can be directly used as the weight.

[0119] Pearson correlation coefficient or Spearman correlation coefficient are usually used alone to measure linear or monotonic relationships between variables, but in the wind power scenario: the coupling between wind speed and wind direction is not taken into account: the effects of wind speed and wind direction on wind turbine operation are interrelated (for example, a sudden change in wind direction at high wind speed may lead to a sharp drop in power), but existing methods only calculate the correlation coefficients of wind speed and wind direction separately, without considering the combined effect of the two on the consistency of wind turbine operation.

[0120] In existing technologies, the weights of the correlation coefficients between wind speed and wind direction are usually fixed (e.g., 0.5 and 0.5). However, the wind speed / wind direction fluctuation characteristics of different wind farms vary significantly (e.g., wind speed fluctuates greatly in coastal wind farms, while wind direction changes frequently in inland wind farms). Fixed weights cannot adapt to the dynamic needs of actual scenarios.

[0121] Traditional K-MEANS algorithms randomly initialize cluster centers, which may lead to local optima, especially in wind farm clusters where turbine distribution can exhibit spatial clustering (e.g., turbines on the same ridge are more correlated). Random initialization may disrupt this natural grouping. Typically, the K-value is determined using the silhouette coefficient or the elbow rule. However, metrics such as the silhouette coefficient and intra-cluster variance are based solely on the geometric characteristics of data distribution (e.g., proximity), failing to consider the spatiotemporal coupling of wind conditions in the wind farm. This makes K-value selection prone to getting stuck in local optima or becoming disconnected from actual needs. Existing graph models are usually based on objective functions such as modularity optimization, but they do not explicitly establish a quantitative relationship between graph features (e.g., average degree, number of connected components) and the optimal K-value. This method applies the graph model to determine the optimal K-value. By explicitly encoding the wind farm's topology (e.g., edge weight = wind condition correlation coefficient) and operational consistency through the graph model, and combining feature engineering (screening key graph features) and transfer learning (matching with a sample graph model library), it achieves a combination of physical constraints and data-driven K-value selection. Furthermore, traditional methods require optimizing the K-value individually for each wind farm, resulting in high computational costs and difficulty in adapting to dynamic changes (such as seasonal wind direction shifts). This method constructs a sample graph model library covering wind farm characteristics and optimal K-values ​​under different geographical environments (e.g., coastal, inland, mountainous) and climatic conditions (e.g., monsoon regions, typhoon regions). By calculating the Euclidean distance of key features between the current graph model and the sample graph models, the optimal K-value of similar samples is quickly matched, enabling knowledge transfer across wind farms.

[0122] Traditional K-MEANS initial cluster centers are random. This method prioritizes wind turbines with wind condition correlation coefficients less than a first threshold as initial cluster centers to avoid local optima. When updating centers, the wind turbine with the smallest mean correlation coefficient to other wind turbines within the group is selected as the new center, rather than the traditional mean point, to adapt to the non-uniformity of wind condition distribution in wind farms.

[0123] Understandably, wind condition variations among wind turbines directly impact computational load (e.g., high-wind-speed groups require more frequent data sampling and power prediction calculations). Grouping highly correlated wind turbines into the same group could lead to a surge in computational load for that group, while the distributed computing nodes in other groups remain idle. Grouping by wind condition variations can balance the load on each distributed computing node, improving overall resource utilization.

[0124] Step 6: Based on the wind turbine and the resource constraints of the distributed computing nodes, construct an optimization objective function and determine the distribution strategy of the distributed computing nodes through an optimization algorithm;

[0125] Step 6, which involves constructing the optimization objective function and determining the distribution strategy of distributed computing nodes through an optimization algorithm, includes:

[0126] Acquire historical wind data of the wind power cluster, which includes wind characteristics (e.g., 10-minute average wind speed, standard deviation of wind direction, turbulence intensity) of multiple wind turbine locations over multiple historical time periods.

[0127] Based on historical wind data from wind power clusters, multiple wind scenario scenarios are constructed.

[0128] Generate multiple initial computing node distribution strategies that satisfy resource allocation constraints;

[0129] For each wind condition scenario, the resource utilization, task latency, and computing load balancing metrics of each initial computing node distribution strategy are calculated using the fitness function, thus obtaining the fitness value of each initial computing node distribution strategy in the wind condition scenario.

[0130] Determine the weighting coefficients for each wind condition scenario;

[0131] For each initial computing node distribution strategy, the fitness values ​​of the initial computing node distribution strategy in each wind condition scenario are weighted and summed based on the weighting coefficients corresponding to each wind condition scenario to obtain the comprehensive fitness value of the initial computing node distribution strategy.

[0132] Using the comprehensive fitness value as the optimization objective, a strategy search is performed based on the particle swarm optimization algorithm to determine the distribution strategy of the distributed computing nodes.

[0133] Specifically, determine the set of constraints used for policy generation, for example:

[0134] The minimum number of distributed computing nodes n min The maximum number of distributed computing nodes n max ;

[0135] Maximum constraints on the computing load of distributed computing nodes (the CPU / memory / GPU utilization of a single node shall not exceed the maximum utilization threshold Umax (e.g., 80%) to avoid overload leading to task failure); minimum constraints on the computing load of distributed computing nodes (the utilization of a single node shall not be lower than the minimum utilization Umin (e.g., 20%) to avoid resource idleness and waste).

[0136] Geographic isolation constraint: The computing nodes of critical wind turbine units need to be deployed in different physical locations (e.g., the distance between them is greater than a distance threshold) to avoid a single point of failure causing a complete shutdown;

[0137] Transmission distance constraint: The transmission distance between the wind turbine and at least two distributed computing nodes is less than the transmission distance threshold;

[0138] Node isolation constraint: Different wind turbines are assigned different distributed computing nodes.

[0139] Multiple distributed computing node distribution strategies can be generated in various ways.

[0140] For example, generating multiple distributed computing node distribution strategies using random uniform initialization includes the following steps:

[0141] For each wind farm i, randomly generate the number n of distributed computing nodes. i ∈[n min ,n max Randomly generate node coordinates (e.g., uniformly distributed within the area comprised of wind turbines in a wind farm), where n min n is the minimum number of distributed computing nodes (e.g., 2). max This represents the maximum number of distributed computing nodes (e.g., 20).

[0142] Check constraints: If the computing load constraint is violated (e.g., the estimated task load causes the initial utilization rate to be greater than Umax, where Umax is the maximum utilization threshold), regenerate n. i ;

[0143] If geographical isolation constraints or transmission distance constraints are violated, the node location will be adjusted or redistributed.

[0144] For example, nodes can be pre-allocated based on the historical workload of wind turbines to reduce the randomness of the initial solution. This includes: calculating the historical average workload T of wind turbine i. i ;

[0145] Calculate the required number of nodes n i =[T i [ / Chode], where Chode is the computing power of a single distributed computing node: Adjust n i Satisfy [n min ,n max ];

[0146] random perturbation n i (e.g., add or subtract 1) to increase diversity.

[0147] Based on historical wind data from wind power clusters, several typical wind scenario scenarios are identified, including the following steps:

[0148] For any two historical time periods, calculate the Euclidean distance between the wind condition feature vectors of the location of each wind turbine in the two historical time periods, and take the mean of the Euclidean distance between the wind condition feature vectors of the location of each wind turbine in the two historical time periods as the clustering distance between the two historical time periods. The wind condition feature vector is composed of multiple wind condition features.

[0149] By using a clustering algorithm, multiple historical time periods are clustered based on the clustering distance between any two historical time periods to determine multiple wind condition clusters;

[0150] For each wind cluster, a corresponding wind scene is generated. Specifically, for the location of each wind turbine, the mean of the wind feature vectors of the wind turbine's location over the historical time period included in the wind cluster can be calculated and used as the wind feature vector of that wind turbine's location in the corresponding wind scene. For example, wind cluster 1: high wind speed (12-15 m / s), low turbulence (TI < 0.1, where TI represents the ratio of wind speed fluctuations caused by turbulence to the average wind speed);

[0151] Wind cluster 2: Low wind speed (3-6 m / s), high turbulence (TI>0.2);

[0152] Wind condition cluster 3: Sudden change in wind direction (wind direction change rate > 5° / min).

[0153] Establish a fitness function, which aims to maximize resource utilization, minimize task latency, and maximize balanced computing power allocation. Resource utilization can be the ratio of the actual computing power utilization of all distributed computing nodes to the theoretical maximum utilization. Task latency can be based on the maximum deviation between the actual time taken by all computing tasks from initiation to completion and the ideal time. The load balancing index can be determined based on the reciprocal of the variance of node utilization (the smaller the variance, the higher the balance, and the higher the load balancing index score).

[0154] For example, the fitness function is:

[0155] f=U+L-Δt

[0156] Where f is the fitness value, U is the normalized resource utilization rate, L is the score of the normalized load balancing index, and Δt is the normalized task latency.

[0157] For each wind condition scenario, computational task data of the wind power cluster within a specific historical time period in the corresponding wind condition cluster can be obtained. This includes the type of each computational task (such as data acquisition, status monitoring, fault prediction, etc.), the computational load (such as CPU cycles, memory usage, etc.), and the task initiation time. Based on the initial computational node distribution strategy, the computational task data of the wind power cluster within that historical time period is distributed to the corresponding distributed computational nodes. The running results of each distributed computational node are obtained, including task execution status, computing load status, and the occupancy of each computing resource. Based on the running results of the wind condition scenario corresponding to the initial computational node distribution strategy, the resource utilization rate, task latency, and computational load balancing indicators of the initial computational node distribution strategy are determined. These are then substituted into the fitness function formula to calculate the fitness value of the initial computational node distribution strategy in the wind condition scenario.

[0158] The weighting coefficient corresponding to a wind condition scenario can be the ratio of the total number of historical time periods included in the wind condition cluster corresponding to the wind condition scenario to the total number of all historical time periods.

[0159] The optimal distributed computing node distribution strategy is obtained by using the particle swarm optimization algorithm based on the comprehensive fitness value of the distributed computing node distribution strategy. The optimal distributed computing node distribution strategy can be the computing node distribution strategy with the largest comprehensive fitness value.

[0160] While meeting the computing needs of wind farms for real-time monitoring, data analysis, and cluster control, the allocation strategy of distributed computing nodes is optimized to maximize resource utilization, minimize task latency, and reduce deployment and maintenance costs.

[0161] Step 7: Based on the distributed computing node distribution strategy and combined with the real-time operating status of the wind turbine, dynamically adjust the node allocation to realize the operation and control of the wind power cluster.

[0162] In step 7, the dynamic adjustment of node allocation to realize the operation and control of the wind power cluster includes extracting the current wind condition parameters based on the real-time wind speed and wind direction data of the wind turbines.

[0163] Input wind condition parameters into the computing power prediction model to predict the computing resource requirements of each wind turbine.

[0164] Based on the prediction results, a fitness function is constructed by combining the constraints of the number of computing nodes, node load limits, and scheduling response time. The fitness function aims to maximize resource utilization, minimize task latency, achieve balanced node load distribution, and minimize node redistribution time.

[0165] When the computing power requirement of the wind turbine changes and the triggering condition is met, the computing node allocation strategy is updated based on the fitness function.

[0166] Based on the updated allocation strategy, wind power cluster operation and control operations are executed, including implementing speed control strategy, pitch angle control strategy, power smoothing control strategy, and grid frequency response control strategy under different wind speed ranges.

[0167] Specifically, based on the real-time wind conditions of the wind turbines, the computing resources required for each wind turbine are predicted.

[0168] Input features: current wind speed v, wind direction θ; historical wind speed sequence (mean and variance of the past 10 minutes).

[0169] Output target: predict the computing resources required for each wind turbine.

[0170] Model selection: Lightweight model: LSTM network (suitable for real-time inference on edge devices).

[0171] The model structure is detailed below, including the input layer, where the current wind speed v is a scalar input that is directly passed to the feature embedding layer. Wind direction θ is converted to sin(θ) and cos(θ) encoding to avoid abrupt angle changes (such as jumps between 0° and 360°). The historical wind speed sequence consists of the mean and variance over the past 10 minutes, forming a time series of shape (10,2) (10 time steps, 2 features per step).

[0172] Feature embedding layer, wind speed / direction branch: A fully connected layer (FC, 16D) is used to map scalar features to a 16-dimensional space, enhancing expressive power. BatchNorm is used to accelerate training and improve stability, followed by a ReLU activation function.

[0173] Historical sequence branch: Use Conv1D(16,3) to extract local time patterns (such as sudden changes in wind speed), followed by MaxPool1D(2) downsampling.

[0174] After being flattened, it is spliced ​​with the wind speed / wind direction branch to form an eigenvector of (1,32+16)=(1,48).

[0175] Sequence modeling layer, bidirectional LSTM: 32 hidden units, captures contextual dependencies (such as the correlation between wind speed change trends and computing power requirements). Only the output of the last time step (not the complete sequence) is returned to reduce computational cost. Dropout (rate=0.2): randomly discards 20% of neurons to prevent overfitting.

[0176] Output layer, fully connected layer (FC, 16-dimensional): further extracts higher-order features, followed by ReLU. Output layer (FC, 1-dimensional): directly predicts computing resource requirements (unit: GFLOPs).

[0177] Loss function, mean squared error (MSE): measures the squared error between the predicted and actual values. L2 regularization (λ = 1e-4): constrains the weights to prevent overfitting.

[0178] The training process includes the following steps:

[0179] Data preprocessing, normalization, and normalization of wind speed v and historical series mean / variance to [0,1] (based on training set statistics). The sin / cos encoding of wind direction θ is naturally in the range [-1,1].

[0180] Sliding window: Samples are generated by sliding along the time step. Each sample contains the wind speed sequence and wind direction sequence of the past 10 minutes, as well as the wind speed v and wind direction θ at the current moment, and is labeled with the computing power requirement at the next moment.

[0181] Model training, optimizer: Adam (lr = 0.001, decay rate = 0.9 / 0.999, where lr is the learning rate). Batch size: 32 (adapted to edge device memory). Early stopping mechanism: Training stops when the validation set MSE does not decrease for 5 consecutive rounds. Learning rate scheduling: If the validation loss stagnates, the learning rate is multiplied by 0.1.

[0182] The operation of wind power clusters is regulated based on a distributed computing node allocation strategy.

[0183] Specifically, the distributed computing nodes are used to perform the following adjustments: in low wind speed areas (cut-in wind speed ≤ v < rated wind speed vrated), the goal is to maximize wind energy capture efficiency (Cp close to the Bates limit of 0.593). Control strategies include: pitch angle control: fixed β = 0° (pitch angle not adjusted, maintaining optimal angle of attack); speed control: tracking the optimal tip speed ratio λ. opt (usually λ) opt ≈6~8), achieved through pitch control or generator torque regulation. Dynamic response: For every 1m / s change in wind speed, the rotational speed must be adjusted to ω within 2 seconds. opt ±5%; closed-loop speed control is achieved through a PID controller, with overshoot <10%.

[0184] In high wind speed areas (v ≥ rated wind speed vrated), the objective is to limit power output to the rated value Prated and protect mechanical components (gearbox, generator). Control strategies include: Pitch angle control: dynamically increasing β (usually β ∈ [0°, 30°]) to reduce wind energy captured by the rotor; Speed ​​control: maintaining the rated speed ωrated to avoid overspeeding.

[0185] Power smoothing under sudden wind speed changes: Sudden increases / decreases in wind speed can cause sudden changes in power output (e.g., changes of >20% × Prated within 10 seconds), leading to grid frequency deviations.

[0186] The method in this embodiment adds a power rate of change limit (RoCoF): dt / dP≤Rmax; Rmax represents the maximum allowable power rate of change (Rmax=0.1·Prated / s); it is achieved through converter control: when dt / dP>Rmax, the power reference value is temporarily reduced, where dt / dP is the power rate of change and s is the sampling time.

[0187] Energy storage system synergy: Configure supercapacitors or flywheel energy storage; during power surges, the energy storage system absorbs / releases energy to smooth out output fluctuations.

[0188] Grid frequency regulation support requirements: Adjust power output according to grid frequency deviation Δf (participate in primary frequency regulation). Control strategy: Droop control.

[0189] ΔP=-Kf·Δf

[0190] Where ΔP is the power adjustment value, Kf represents the frequency modulation coefficient (usually Kf = 10%·Prated / 0.1Hz); Δf represents the frequency deviation range, Δf∈[-0.5Hz,+0.5Hz).

[0191] Energy recovery: After frequency modulation, power output is restored to the rated value by reducing the pitch angle or releasing the stored energy (recovery time < 30 seconds).

[0192] Example 2, refer to Figure 3 As one embodiment of the present invention, this embodiment provides a wind power cluster control system based on distributed computing, comprising:

[0193] The data acquisition module is used to acquire historical wind speed and direction data of each wind turbine in the wind power cluster;

[0194] The correlation calculation module is used to calculate the wind condition correlation coefficient between any two wind turbines based on the historical wind speed and wind direction data.

[0195] The graph model construction module is used to construct a graph model of the wind power cluster based on the wind condition correlation coefficient. In the graph model, each wind turbine is a node in the graph model. An edge is established between any two nodes when the corresponding wind condition correlation coefficient is greater than a set threshold, and the wind condition correlation coefficient is used as the weight of the edge.

[0196] The cluster number determination module is used to extract graph feature parameters based on the graph model, perform similarity matching with the graph feature parameters of multiple sample graph models, and determine the cluster number of the current wind power cluster.

[0197] The clustering module is used to cluster the wind turbines in the wind power cluster based on the clustering number to obtain multiple wind power groups.

[0198] The distribution strategy determination module is used to construct an optimization objective function based on the wind turbine and the resource constraints of the distributed computing nodes, and to determine the distribution strategy of the distributed computing nodes through an optimization algorithm.

[0199] The control and execution module is used to dynamically adjust the node allocation based on the distributed computing node distribution strategy and the real-time operating status of the wind turbine, and to execute the operation control operations of the wind power cluster.

[0200] This embodiment also provides an electronic device applicable to a wind power cluster control method based on distributed computing, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the wind power cluster control method based on distributed computing as proposed in the above embodiment.

[0201] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a wind power cluster control method based on distributed computing as proposed in the above embodiments.

[0202] The storage medium proposed in this embodiment and the wind power cluster control method based on distributed computing proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0203] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0204] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A wind power cluster control method based on distributed computing, characterized in that: include, Obtain historical wind speed and direction data for each wind turbine in the wind power cluster; Based on historical wind speed and direction data, calculate the wind condition correlation coefficient between any two wind turbines; Based on the calculated wind condition correlation coefficient, a graph model of the wind power cluster is constructed. In the graph model of the wind power cluster, each wind turbine is a node in the graph model. An edge is established between any two nodes when the corresponding wind condition correlation coefficient is greater than a set threshold. The wind condition correlation coefficient is used as the weight of the edge to represent the wind condition relationship between the wind turbines. Based on the graph model, graph feature parameters representing the graph structure are extracted and similarity matching is performed with the graph feature parameters of multiple sample graph models to determine the sample graph model most similar to the current graph model. The number of clusters corresponding to the sample graph model is used as the number of clusters of the current wind power cluster. Based on the number of clusters, the wind turbines in the wind power cluster are clustered to obtain multiple wind power groups; Based on wind turbines and considering the resource constraints of distributed computing nodes, an optimization objective function is constructed, and the distribution strategy of distributed computing nodes is determined through optimization algorithms. Based on the distributed computing node distribution strategy and combined with the real-time operating status of the wind turbines, the node allocation is dynamically adjusted to achieve operation control of the wind power cluster.

2. The wind power cluster control method based on distributed computing as described in claim 1, characterized in that: The calculation of the wind condition correlation coefficient between any two wind turbines includes extracting the wind speed sequence and wind direction sequence under the corresponding timestamp; The wind speed correlation coefficient between the wind speed sequences and the wind direction correlation coefficient between the wind direction sequences were calculated using the Pearson correlation coefficient, respectively. Calculate the standard deviation of the wind speed sequence and the standard deviation of the wind direction sequence for each of the two wind turbines. Based on the standard deviation of wind speed and the standard deviation of wind direction, a wind speed weighting factor and a wind direction weighting factor are determined. The wind speed weighting factor is the ratio of the standard deviation of wind speed to the sum of the standard deviations of wind speed and wind direction, and the wind direction weighting factor is the ratio of the standard deviation of wind direction to the sum of the standard deviations of wind speed and wind direction. After taking the absolute values ​​of the wind speed correlation coefficient and the wind direction correlation coefficient, the wind condition correlation coefficient is obtained by weighting and summing them according to the wind speed weighting factor and the wind direction weighting factor.

3. A wind power cluster control method based on distributed computing as described in claim 2, characterized in that: The graph model for constructing the wind power cluster includes representing each wind turbine in the wind power cluster as a node in the graph model; When the wind condition correlation coefficient between any two wind turbines is greater than a set threshold, an edge is established between the nodes representing the two wind turbines, and the weight of the edge is set to the wind condition correlation coefficient between the two wind turbines.

4. A wind power cluster control method based on distributed computing as described in claim 3, characterized in that: The step of using the number of clusters corresponding to the sample graph model as the number of clusters of the current wind power cluster includes extracting multiple graph feature parameters to characterize the graph structure characteristics based on the current wind power cluster graph model. The extracted graph feature parameters are compared with the graph feature parameters of multiple preset sample graph models to measure their similarity. The similarity measurement is calculated based on the Euclidean distance between the graph feature parameters. The sample graph model with the minimum Euclidean distance is identified, and the number of clusters corresponding to the sample graph model with the minimum Euclidean distance is determined as the number of clusters of the current wind power cluster.

5. A wind power cluster control method based on distributed computing as described in claim 4, characterized in that: The clustering of wind turbines in the wind power cluster includes selecting several wind turbines whose wind condition correlation coefficients between each pair are less than a first threshold as initial cluster centers based on the wind condition correlation coefficient. For each wind turbine that is not a cluster center, calculate the wind condition correlation coefficient with each cluster center and assign it to the cluster group corresponding to the cluster center with the smallest distance; For each cluster, calculate the average wind condition correlation coefficient between all wind turbines in the cluster and other wind turbines in the cluster, and select the wind turbine with the smallest average distance as the new cluster center. Repeat the wind turbine allocation and cluster center update operations until the cluster centers no longer change.

6. A wind power cluster control method based on distributed computing as described in claim 5, characterized in that: The construction of the optimization objective function and the determination of the distribution strategy of the distributed computing nodes through the optimization algorithm include obtaining historical wind condition data of the wind power cluster, wherein the historical wind condition data of the wind power cluster includes wind condition characteristics of the locations of multiple wind turbines in multiple historical time periods. Based on historical wind data from wind power clusters, multiple wind scenario scenarios are constructed. Generate multiple initial computing node distribution strategies that satisfy resource allocation constraints; For each wind condition scenario, the resource utilization, task latency, and computing load balancing metrics of each initial computing node distribution strategy are calculated using the fitness function, thus obtaining the fitness value of each initial computing node distribution strategy in the wind condition scenario. Determine the weighting coefficients for each wind condition scenario; For each initial computing node distribution strategy, the fitness values ​​of the initial computing node distribution strategy in each wind condition scenario are weighted and summed based on the weighting coefficients corresponding to each wind condition scenario to obtain the comprehensive fitness value of the initial computing node distribution strategy. Using the comprehensive fitness value as the optimization objective, a strategy search is performed based on the particle swarm optimization algorithm to determine the distribution strategy of the distributed computing nodes.

7. A wind power cluster control method based on distributed computing as described in claim 6, characterized in that: The dynamic adjustment of node allocation to realize the operation and control of wind power clusters includes extracting current wind condition parameters based on real-time wind speed and wind direction data of wind turbines; Input wind condition parameters into the computing power prediction model to predict the computing resource requirements of each wind turbine. Based on the prediction results, a fitness function is constructed by combining the constraints of the number of computing nodes, node load limits, and scheduling response time. The fitness function aims to maximize resource utilization, minimize task latency, achieve balanced node load distribution, and minimize node redistribution time. When the computing power requirement of the wind turbine changes and the triggering condition is met, the computing node allocation strategy is updated based on the fitness function. Based on the updated allocation strategy, wind power cluster operation and control operations are executed, including implementing speed control strategy, pitch angle control strategy, power smoothing control strategy, and grid frequency response control strategy under different wind speed ranges.

8. A wind power cluster control system based on distributed computing, employing a wind power cluster control method based on distributed computing as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to acquire historical wind speed and direction data of each wind turbine in the wind power cluster; The correlation calculation module is used to calculate the wind condition correlation coefficient between any two wind turbines based on the historical wind speed and wind direction data. The graph model construction module is used to construct a graph model of the wind power cluster based on the wind condition correlation coefficient. In the graph model, each wind turbine is a node in the graph model. An edge is established between any two nodes when the corresponding wind condition correlation coefficient is greater than a set threshold, and the wind condition correlation coefficient is used as the weight of the edge. The cluster number determination module is used to extract graph feature parameters based on the graph model, perform similarity matching with the graph feature parameters of multiple sample graph models, and determine the cluster number of the current wind power cluster. The clustering module is used to cluster the wind turbines in the wind power cluster based on the clustering number to obtain multiple wind power groups. The distribution strategy determination module is used to construct an optimization objective function based on the wind turbine and the resource constraints of the distributed computing nodes, and to determine the distribution strategy of the distributed computing nodes through an optimization algorithm. The control and execution module is used to dynamically adjust the node allocation based on the distributed computing node distribution strategy and the real-time operating status of the wind turbine, and to execute the operation control operations of the wind power cluster.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the wind power cluster control method based on distributed computing according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When a computer program is executed by a processor, it implements the steps of a wind power cluster control method based on distributed computing, as claimed in any one of claims 1 to 7.

Images