Short-term power prediction method and system based on wind power plant cluster

Abstract

The invention discloses a short-term power prediction method and system based on a wind power plant cluster. The method comprises the following steps: constructing a similarity distance matrix of a key meteorological element time-varying rule between different stations; calculating a Copula entropy between the historical power samples of each station, and constructing a spatial similarity matrix of station power according to each Copula entropy; performing normalization processing on the similarity distance matrix and the spatial similarity matrix, and summing the similarity distance matrix and the spatial similarity matrix after normalization processing to obtain a space-time similarity distance matrix; clustering the similarity distance matrix by adopting a hierarchical clustering algorithm, determining a partitioning scheme in combination with the geographic distribution of the new energy stations, and dividing each wind power plant in the target region into each region; and outputting a total power prediction result of the obtained region according to the LSTM time sequence prediction model. The overall precision and robustness of power prediction are improved, and a more reliable scheduling basis is provided for large-scale access of a new energy station to a power grid.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind energy power generation and comprehensive consumption, and particularly relates to a short-term power prediction method and system for a wind farm cluster. Background Art

[0002] The volatility and uncertainty of wind power are key factors affecting the stability of power grid dispatching and operation. During the power prediction process of wind farms, due to the influence of complex meteorological factors, geographical environment, and power grid operation characteristics on wind power, its power output shows a high degree of spatio-temporal correlation. However, traditional wind power prediction methods mostly take individual stations as units, ignoring the collaborative relationship between stations and failing to fully explore the similarity of power changes shown by wind farms in the region driven by similar meteorological backgrounds. At the same time, existing cluster prediction methods generally have problems such as rough regional division, over-reliance on geographical information or subjective experience, and it is difficult to ensure that the stations within the partition have consistent characteristics. Summary of the Invention

[0003] The present invention provides a short-term power prediction method and system for a wind farm cluster to solve the technical problem of failing to fully explore the similarity of power changes shown by wind farms in the region driven by similar meteorological backgrounds.

[0004] In a first aspect, the present invention provides a short-term power prediction method for a wind farm cluster, including:[[]] Obtaining power data at 5-minute intervals, measured wind speed data at the hub of the fan, and predicted wind speed data at the hub of the fan for N wind farm stations in a target area on M historical sample days, where: M is a three-dimensional real space, R represents the real number field, M represents M sample days, 288 represents the number of 5-minute samples in a day, N represents the number of wind farm stations, P_i is the power sequence of the i-th wind farm station, V_i is the measured wind speed sequence of the i-th wind farm station, and V_i^ is the predicted wind speed sequence of the i-th wind farm station; N wind farm stations M historical sample days of 5-minute power data , measured wind speed data at the hub of the fan and predicted wind speed data at the hub of the fan , where, M is a three-dimensional real space, R represents the real number field, M represents M sample days, 288 represents the number of 5-minute samples in a day, N represents the number of wind farm stations, P_i is the power sequence of the i-th wind farm station, V_i is the measured wind speed sequence of the i-th wind farm station, V_i^ is the predicted wind speed sequence of the i-th wind farm station; Calculating the ACF curve of the wind speed at the hub of the fan, calculating the weighted Euclidean distance between the ACF curves of different stations, and constructing a similarity distance matrix of the time-varying laws of key meteorological elements between different stations , where, S is a two-dimensional real space, Represents the number of wind farms; Calculate the Copula entropy between the historical power samples of each farm, and construct a spatial similarity matrix of the farm power according to each Copula entropy ; The similarity distance matrix and the spatial similarity matrix are normalized, and the normalized similarity distance matrix and the spatial similarity matrix are summed to obtain a spatio-temporal similarity distance matrix ; Use the hierarchical clustering algorithm to cluster the spatio-temporal similarity distance matrix , and determine the partition scheme in combination with the geographical distribution of new energy farms, and divide the wind farms in the target area into K regions; Establish an LSTM time series prediction model for each region respectively. Input the power sequence in a certain region, the measured wind speed sequence at the hub of the wind turbines representing the farms in a certain region, and the predicted wind speed sequence at the hub of the wind turbines representing the farms in a certain region into a certain LSTM time series prediction model. The certain LSTM time series prediction model outputs the total power prediction result of a certain region, where the farms representing a certain region are selected in a certain region based on the maximum correlation - minimum redundancy principle.

[0005] In a second aspect, the present invention provides a short-term power prediction system based on a wind farm cluster, including: An acquisition module configured to acquire the power data at the 5 - minute level of wind farms in the target area for historical sample days, the measured wind speed data at the hub of the wind turbines, and the predicted wind speed data at the hub of the wind turbines, where , is a three - dimensional real - number space, represents the real - number field, represents sample days, 288 represents the number of 5 - minute - level samples in a day, represents the number of wind farms, is the power sequence of the th wind farm, is the measured wind speed sequence of the th wind farm, is the predicted wind speed sequence of the th wind farm; A calculation module, configured to calculate the ACF curve of the wind speed at the hub of the wind turbine, calculate the weighted Euclidean distance between the ACF curves of different stations, and construct a similarity distance matrix of the time-varying laws of key meteorological elements between different stations , where is a two-dimensional real number space, represents the number of wind farm stations; A construction module, configured to calculate the Copula entropy between the historical power samples of each station, and construct a spatial similarity matrix of the station power according to each Copula entropy ; A summation module, configured to normalize the similarity distance matrix and the spatial similarity matrix , and sum the normalized similarity distance matrix and the spatial similarity matrix to obtain a spatio-temporal similarity distance matrix ; A partitioning module, configured to cluster the spatio-temporal similarity distance matrix using a hierarchical clustering algorithm, determine a partitioning scheme in combination with the geographical distribution of new energy stations, and divide the wind farms in the target area into K regions; An output module, configured to establish an LSTM time series prediction model for each region respectively, input the power sequence in a certain region, the measured wind speed sequence at the hub of the wind turbine of the representative station in a certain region, and the predicted wind speed sequence at the hub of the wind turbine of the representative station in a certain region into a certain LSTM time series prediction model, and the certain LSTM time series prediction model outputs the total power prediction result of a certain region, where the representative station in a certain region is selected in a certain region based on the maximum correlation - minimum redundancy principle.

[0006] In a third aspect, an electronic device is provided, which includes: at least one processor, and a memory communicatively connected to the at least one processor, where the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the method for short-term power prediction of a wind farm cluster according to any embodiment of the present invention.

[0007] In a fourth aspect, the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the program instructions are executed by a processor, the processor is enabled to execute the steps of the method for short-term power prediction of a wind farm cluster according to any embodiment of the present invention.

[0008] The short-term power prediction method and system for a wind farm cluster according to the present application utilize the spatio-temporal coupling characteristics between wind speed and power to cluster and partition the wind farms, and predict the total power of each sub-region. By optimizing the station partition strategy and selecting representative stations for cluster prediction modeling, this method can effectively reduce random errors, improve the overall accuracy and robustness of power prediction, and provide a more reliable scheduling basis for large-scale access of new energy power stations to the power grid. Description of the Drawings

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0010] Figure 1 It is a flowchart of a short-term power prediction method for a wind farm cluster provided by an embodiment of the present invention; Figure 2 It is a flowchart of the partition aggregation of wind farm stations based on spatio-temporal correlation provided by a specific embodiment of an embodiment of the present invention; Figure 3 It is a flowchart of the selection of regional representative stations based on maximum correlation and minimum redundancy provided by a specific embodiment of an embodiment of the present invention; Figure 4 It is a flowchart of an LSTM prediction model provided by a specific embodiment of an embodiment of the present invention; Figure 5 It is a flowchart of the training and prediction of an LSTM model provided by a specific embodiment of an embodiment of the present invention; Figure 6 It is a geographical distribution map of wind farm stations provided by a specific embodiment of an embodiment of the present invention; Figure 7 It is a schematic diagram of the partition results of 68 wind farm stations and the selection results of regional representative stations provided by a specific embodiment of an embodiment of the present invention; Figure 8 It is each partition provided by a specific embodiment of an embodiment of the present invention bar chart of prediction evaluation indicators; Figure 9 It is each model provided by a specific embodiment of an embodiment of the present invention 、 、 bar chart of evaluation indicators; Figure 10 It is a power prediction curve chart of a comparative model provided by a specific embodiment of an embodiment of the present invention; Figure 11A structural block diagram of a short-term power prediction system based on a wind farm cluster provided by an embodiment of the present invention; Figure 12 It is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0012] Please refer to Figure 1 , which shows a flowchart of a short-term power prediction method based on a wind farm cluster of the present application.

[0013] As Figure 1 shown, the short-term power prediction method based on a wind farm cluster specifically includes the following steps: Step S101, obtain the 5-minute power data, the measured wind speed data at the hub of the fan, and the predicted wind speed data at the hub of the fan of wind farm stations in a target area for historical sample days. .

[0014] In this step, is the power sequence of the th wind farm station, is the measured wind speed sequence of the th wind farm station, is the predicted wind speed sequence of the th wind farm station, is a three-dimensional real number space, represents the real number field, represents sample days, 288 represents the number of 5-minute samples in a day, represents the number of wind farm stations; Step S102, calculate the ACF curve of the wind speed at the hub of the fan, calculate the weighted Euclidean distance between the ACF curves of different stations, and construct a similarity distance matrix of the time-varying laws of key meteorological elements between different stations .

[0015] In this step, is a two-dimensional real number space, Indicates the number of wind farms; for the time-varying laws of key meteorological elements, calculate the similarity between the ACF curves of key meteorological elements at each farm. First, calculate the autocorrelation function (ACF) curve of the wind speed at the hub of the wind turbine. Then, considering that as the lag time increases, the difference in the correlation coefficients on the ACF curve between different farms gradually increases, so combined with the characteristic of the "near-large and far-small" influence weight of time-series correlation, calculate the weighted Euclidean distance between the ACF curves of the key meteorological elements at the farms, and finally form a similarity distance matrix of the time-varying laws of key meteorological elements between different farms. According to Figure 2 Explain the calculation process of the similarity distance matrix as follows: Calculate the ACF curve of the wind speed at the hub of the wind turbine for each wind farm , where the expression for calculating the ACF value at the lag time t for the i-th farm is: , In the formula, is the length of the wind speed sequence, is the mean of the measured wind speed sequence of the i-th wind farm, , is the ACF value at the lag time t for the i-th wind farm, is the wind speed value of the measured wind speed sequence of the i-th wind farm at the time point , is the wind speed value of the measured wind speed sequence of the i-th wind farm at the time point ; Construct the ACF matrix, and the expression is: , In the formula, is the ACF matrix, is the ACF value at the lag time n for the m-th wind farm; The autocorrelation value at each lag point of the ACF curve represents the correlation of the wind speed data at that time lag. The time points with larger lags indicate weaker correlations, while the time points with smaller lags have stronger correlations.

[0016] When multiple wind farms are involved, the ACF curve of a single wind farm may have large fluctuations, making it difficult to draw stable conclusions. Therefore, use the average ACF curve to synthesize the autocorrelation information of multiple data sets to obtain more reliable weights. Calculate the average ACF value at the lag time t for all wind farms, and calculate the Euclidean distance weight for each lag time point according to the average ACF value. Among them, the expression for calculating the average ACF value is: , In the formula,​ is the ACF value of wind farm k at lag time t, is the number of wind farm stations; The expression for calculating the Euclidean distance weight is: , According to the Euclidean distance weight, calculate the weighted Euclidean distance matrix between the ACF curves of each station, that is, construct the similarity distance matrix of the time-varying law of key meteorological elements between different stations , and the expression is: , In the formula, is the weighted Euclidean distance between the m-th wind farm and the m-th wind farm, and the value is 0, is the weighted Euclidean distance between wind farm h and wind farm j; , In the formula, 、 are the autocorrelation coefficients of wind farm station h and wind farm station j at lag time t respectively, is the maximum lag time of ACF. The obtained The smaller the value, the higher the temporal similarity between the wind speeds at the hub heights of the two stations' wind turbines.

[0017] Step S103, calculate the Copula entropy between the historical power samples of each station, and construct the spatial similarity matrix of the station power according to each Copula entropy .

[0018] In this step, for the randomness law of historical power, calculate the output power sample values of each station The non-linear spatial correlation between them is calculated. By calculating the Copula entropy between the historical power samples of each station, the non-linear correlation between the power data of different stations is reflected. The smaller the Copula entropy, the stronger the correlation. Finally, a spatial correlation matrix of the station power is formed . According to Figure 2 The calculation process of the similarity distance matrix is explained as follows: Calculate the power time series samples of wind farms h and j and The Copula entropy between them , where T is the number of samples, and the calculation process is: , , In the formula, is the empirical marginal distribution function of the power time series sample of the h-th wind farm, is the number of power time series samples, is an indicator function. If holds, the function value is 1. If does not hold, the function value is 0. is the s-th sample in the power time series of the h-th wind farm, is the s-th sample in the power time series of the j-th wind farm, is the power time series of the wind farm; Map the power time series samples of wind farm and wind farm to marginal probabilities through the empirical distribution function. The expression is: , , In the formula, is the t-th sample in the power sequence of wind farm h , is the t-th sample in the power sequence of wind farm j , is the empirical marginal distribution function of the power sequence of wind farm j , is the empirical marginal distribution function of the power sequence of wind farm h , is the marginal probability of the t-th sample in the power sequence of wind farm h , is the marginal probability of the t-th sample in the power sequence of wind farm j ; Construct the joint sample pair based on the marginal probabilities, and calculate the distance between the sample pairs according to the Chebyshev distance. The expression is: , , In the formula, is the Chebyshev distance between the s-th sample pair and the v-th sample pair , , is the marginal probability of the s-th sample in wind farm h in the sample pair , is the marginal probability of the v-th sample in wind farm h in the sample pair , is the marginal probability of the s-th sample in wind farm j in the sample pair , is the marginal probability of the v-th sample in wind farm j in the sample pair , is the s-th power sample, is the v-th power sample; For each sample pair, find the k-nearest neighbor distance of each sample pair, where k is the number of nearest neighbors. In this method, k = 3 is set to balance the bias and variance of the estimation, and estimate the Copula entropy between the historical power samples of each station based on the k-nearest neighbor distance. The expression is: , In the formula, is the volume of the two-dimensional unit sphere. Under the Chebyshev distance = 1, is the Chebyshev distance of the k-th nearest neighbor of sample pair s, is the Copula entropy between station h and station j, is the Digamma function; consists of to form the spatial correlation matrix representing the nonlinearity between two stations, that is, the spatial similarity matrix of the station power is constructed , and the expression is: , In the formula, is the Copula entropy between station h and station j, is the Copula entropy between station m and station m. The smaller the value, the higher the nonlinear similarity between the two stations.

[0019] Step S104, normalize the similarity distance matrix and the spatial similarity matrix , and sum the normalized similarity distance matrix and the spatial similarity matrix to obtain the spatio-temporal similarity distance matrix .

[0020] In this step, the maximum-minimum normalization method is adopted. Let the matrix elements before and after normalization be and , and the maximum and minimum values of the matrix elements are , respectively. The specific normalization formula is: . Based on the obtained normalized matrix, sum and to form the spatio-temporal similarity distance matrix , . Among them, the element represents the similarity distance between the i-th and j-th stations. The smaller the value, the higher the similarity between the two stations.

[0021] Spatio-temporal similarity distance matrix The expression of is as follows: , In the formula, is the normalized Copula entropy matrix, is the weighted Euclidean distance matrix between the ACF curves of each station after normalization, is the spatio-temporal similarity distance between station h and station j, is the spatio-temporal similarity distance between the m-th wind farm and the m-th wind farm.

[0022] Step S105: Use the hierarchical clustering algorithm to cluster the similarity distance matrix and determine the partitioning scheme in combination with the geographical distribution of new energy stations, and divide the wind farms in the target area into K regions.

[0023] In this step, input the spatio-temporal similarity distance matrix obtained in step S104 . Initially, each wind farm station is regarded as a separate cluster, that is, there are a total of m clusters .

[0024] The initial distance between two clusters . The clustering process uses the Ward minimum variance method (Ward linkage), aiming to minimize the total squared error of the merged clusters. Let the cluster and the cluster be any two clusters to be merged, |C| be the number of wind farm stations in the clusters to be merged, be the spatio-temporal similarity distance between wind farms h and j, then and The Ward distance calculation formula between them is: , Find the two clusters with the smallest d distance and , merge and into a new cluster . Then recalculate the distance between the new cluster and other clusters: , Repeat calculating the distance between the new cluster and other clusters until the number of clusters is reduced to the preset value K, and determine the partitioning result in combination with the geographical distribution, denoted as , where is the set of wind farms included in the k-th region, i is the wind farm number, and denote the number of stations in as .

[0025] Step S106: Establish an LSTM time series prediction model for each region. Input the historical total power time series in a certain region, the measured wind speed time series at the hub of the wind turbines representing the power station in a certain region, and the predicted wind speed time series at the hub of the wind turbines representing the power station in a certain region into a certain LSTM time series prediction model. The certain LSTM time series prediction model outputs the total power prediction result of a certain region, where the power station representing the region in a certain region is selected based on the maximum correlation - minimum redundancy principle.

[0026] In this step, to accurately represent the wind speed within the region and improve the regional prediction accuracy, based on the wind farm zoning result obtained in Step 5, each zoning is taken as an object, and the maximum correlation - minimum redundancy principle (MRMR) is used to select the representative power stations in the region. The number of representative power stations in each region is not less than 1 / 3 of the total number of power stations in the region, that is , where represents rounding up. According to Figure 3 the process of selecting representative power stations is explained as follows: In the k - th region, calculate the mutual information between the measured wind speed sequence at the hub of each wind turbine in the wind farms in the region and the total regional power. The expression is: , where is the mutual information between the measured wind speed sequence at the hub of wind turbine b in the wind farm and the total regional power, is the marginal probability function of the measured wind speed at the hub of wind turbine b in the wind farm, is the marginal probability function of the total wind power, is the joint probability distribution between the marginal probability function of the measured wind speed at the hub of wind turbine b in the wind farm and the total wind power, is the total power of region k, is the measured wind speed sequence at the hub of wind turbine b in the wind farm, is a specific observed value of the measured wind speed at the hub of wind turbine b in the wind farm; Store the calculation result as the correlation vector , select the first representative power station with the largest correlation with the total regional power, is the correlation vector, add to the set of representative power stations , and the remaining set of power stations is . For each candidate power station u ∈ U, calculate the MRMR score , is the wind speed at the hub of the candidate wind power station in region k, is the wind speed at the hub of the selected representative wind farm in area k, is the mutual information between the wind speed at the hub of the candidate wind farm in area k and the total power of area k, is the mutual information between the wind speed at the hub of the candidate wind farm in area k and the wind speed at the hub of the selected representative wind farm in area k. Select the wind farm with the highest MRMR score to join the representative set. When output the representative wind farm set as the representative wind farm of area k.

[0027] It should be noted that after determining the partition and the representative wind farms of each area, an LSTM time series prediction model is established for each area to predict the total wind power generation of the area. For each area, an LSTM time series prediction model is established to output the 288-step prediction result of the area power. According to Figure 4 、 Figure 5 explain the LSTM prediction model.

[0028] LSTM (Long Short-Term Memory) is a variant of the Recurrent Neural Network (RNN) designed specifically to solve the problem of long-term dependencies in sequential data. Compared with traditional RNNs, LSTM can effectively alleviate the problem of gradient vanishing and is suitable for modeling complex time series such as the variation pattern of wind power. The LSTM network consists of multiple LSTM units connected in series, and each unit updates the input sequence at each time step t. Given the current input 、the hidden state at the previous time step 、the memory state output the current hidden state and the memory state . The specific calculation process of LSTM includes the following steps: Forget Gate : Controls the proportion of information to be forgotten in the memory state at the previous time step. The formula is:

[0029] where represents the Sigmoid activation function, is the weight matrix of the forget gate, is the bias term of the forget gate.

[0030] Input Gate and Candidate Memory Unit: The input gate determines how much new information to write, and the candidate memory unit is the candidate value of the new information: , , where is the output of the input gate, is a candidate memory cell, is the hyperbolic tangent activation function, and are both weight matrices of the input gate, and are both bias terms of the input gate.

[0031] Memory state update: Integrate the results of the forget gate and the input gate to update the current memory cell state: , wherein, is the memory cell state updated at the current time t, is the Hadamard product, that is, the corresponding elements are multiplied one by one, is the memory cell state at the previous time t - 1; Output gate and hidden state update: Determine the output hidden state according to the current memory state: , , wherein, is the output of the output gate, is the weight matrix of the output gate, is the bias term of the output gate, is the hidden state at time t - 1.

[0032] The measured wind speed of the representative wind farm station in the acquisition area k , predicted wind speed and the total regional power .

[0033] Data preprocessing: Normalize various data used as input variables and output variables of the LSTM prediction model according to their characteristics; among them, the wind power data is normalized to the interval [0, 1] based on the rated capacity of the wind farm, and the wind speed meteorological data adopts the maximum - minimum normalization method. Let the wind power before and after normalization be and , the wind speed be and , the maximum and minimum values of the wind speed samples be , , the rated capacity of the wind power be , and the specific normalization formulas are as follows: , , Divide the data set into a training set, a validation set and a test set, and construct an input sequence according to the collected and processed data, including: The measured wind speed of the representative station : The normalized value of the measured wind speed of the representative power station within the look-back time window wherein is the normalized value of the measured wind speed of the representative power station within the look-back time window, and is the number of representative power stations in area k; The total historical power of the area : The normalized value of the total power of each sub-area within the look-back time window .

[0034] represents the predicted wind speed of the representative power station : The normalized value of the predicted wind speed sequence at the hub of the representative power station within the prediction time window .

[0035] In the LSTM time series prediction model, the historical wind speed and the total historical power of the area of the representative power stations in area k within the look-back window are concatenated along the feature dimension to form the input sequence . The input sequence is input into the LSTM network. After the LSTM extracts the time series features, the hidden state of the last time step is output; for the prediction period, the predicted wind speed of the representative power stations in the prediction time period is flattened into a one-dimensional vector , which is sent to the fully connected layer to extract global features and compressed into a feature vector , is the weight matrix of the fully connected layer, is the one-dimensional vector after flattening the predicted wind speed of the representative power stations in the prediction time period, is the bias term of the weight matrix of the fully connected layer. Then, the time series features output by the LSTM are concatenated with the feature vector mapped by NWP (numerical weather prediction data) to form a fusion vector , which is input into the fusion fully connected layer network: , is the weight matrix of the fully connected layer, is the bias term of the fully connected layer, is the first output of the fusion fully connected layer network, is the weight matrix of the fully connected layer, is the bias term of the fully connected layer. Through the fully connected layer, the predicted result of the total regional power for the next 288 steps in area k is output, and the inverse normalization is performed on the output predicted result to obtain the predicted result of the total wind power of each sub-area , ​​It is the predicted result after anti - normalization of the total regional power for the 288th step in the future of region k It is the normalized predicted result of the total regional power for the 288th step in the future of region k

[0036] In summary, the method of this application collects the historical wind speed and power data of each wind farm, quantifies the spatio - temporal similarity between stations through the autocorrelation function (ACF) and Copula entropy; secondly, uses the hierarchical clustering method to perform regional division based on the calculated spatio - temporal similarity distance matrix to ensure high consistency of stations within the same partition; then, applies the maximum relevance minimum redundancy (MRMR) principle to select representative stations in the region, screening out representative stations from each region, effectively reducing the scale of input data while ensuring prediction accuracy; finally, based on the wind speed and historical total regional power data of the selected representative stations, a time - series prediction model is constructed to achieve short - term prediction of the day - ahead wind power at the regional scale. The present invention improves the accuracy and computational efficiency of regional power prediction by integrating wind speed - power spatio - temporal correlation modeling, regional division, and selection of representative stations, providing effective support for the stable operation and scheduling of large - scale wind power clusters

[0037] In a specific embodiment, in this embodiment, 68 wind farms in a certain area are used, and the 5 - minute - level wind power data in 2024 are divided into a training set, a validation set, and a test set with a ratio of 8:1:1 for day - ahead wind farm cluster power prediction. The geographical distribution of each wind farm is as Figure 6 shown. The input data for partition aggregation and selection of representative stations include the 5 - minute - level power data and hub - height wind speed data of 68 wind farms in the region throughout 2024; the input data for the regional power prediction stage include the total power of the corresponding partition within the look - back window and the measured wind speed of the representative stations, as well as the predicted wind speed of the representative stations during the prediction period. This embodiment uses the mean absolute percentage error and the mean absolute error and the root mean square error as the evaluation criteria for the prediction accuracy of each model. The calculation formulas for each evaluation index are as follows: , , , wherein, is the number of test samples; and are respectively the actual value and the predicted value of the wind power at the th sampling point at the prediction moment; is the total rated capacity of each wind farm in the corresponding region

[0038] Applying the method of this application, partition aggregation is performed on 68 stationsFigure 7 It is a schematic diagram of the selection results of sub - regions and regional representative stations. It is divided into 5 regions in total. Considering that the dominant wind direction in this region is from northeast to southwest, it can be seen from the figure that the wind farms in each sub - region also show a northeast - to - southwest trend, which proves the rationality of the sub - regions. Observing the regional representative stations, they are evenly distributed within the region and are not limited to a certain area, which proves the rationality of the selection of regional representative stations. Table 1 and Figure 8 show the evaluation results of the prediction performance of the method proposed in this invention on the wind power sub - regions of each region. By statistically comparing the prediction error indexes of each sub - region, it can be seen that the proposed method shows high prediction accuracy in each sub - region. The differences between the evaluation indexes are small and the volatility is low. The average MAPE of each sub - region is 14.19%, indicating that this invention has good regional adaptability and stability. At the same time, there are no obvious deviations or local failures in the prediction results of each sub - region, verifying the generality and reliability of the method of this invention under different spatial distributions. Generally speaking, through the collaborative design of representative station selection, feature fusion and modeling methods, this invention can effectively improve the accuracy and robustness of regional wind power prediction, meeting the requirements of high efficiency and generalization ability for regional wind power prediction in practical applications.

[0039] Table 1 Evaluation indexes of power prediction in each region , To verify the application effect of the prediction model of this invention in practical engineering, taking sub - region 2 as an example for experimental verification, the following comparison methods are selected: MRMR - LSTM: The method proposed in this paper, which selects regional representative stations using the maximum relevance - minimum redundancy principle and uses the LSTM prediction model for regional power prediction.

[0040] All - LSTM: Without selecting regional representative stations, taking the historical wind speed data and power data of all stations in the sub - region as input and using the LSTM prediction model for regional power prediction.

[0041] MRMR - TCN: Selects regional representative stations using the maximum relevance - minimum redundancy principle, takes the predicted and measured wind speed data and power data of the representative stations as input, and uses the TCN prediction model for regional power prediction.

[0042] MRMR - CNN - LSTM: Selects regional representative stations using the maximum relevance - minimum redundancy principle, takes the predicted and measured wind speed data and power data of the representative stations as input, and uses the CNN - LSTM prediction model for regional power prediction.

[0043] MRMR-Bi-LSTM: Select representative stations for the region using the maximum relevance - minimum redundancy principle. Use the predicted and measured wind speed data and power data of the representative stations as inputs, and use the Bi-LSTM prediction model for regional power prediction.

[0044] MRMR-LSTM-OnlyWS: Select representative stations for the region using the maximum relevance - minimum redundancy principle. Only use the measured wind speed data and power data of the representative stations as inputs, and use the LSTM prediction model for regional power prediction.

[0045] In the above comparison models, the comparison between MRMR-LSTM and All-LSTM aims to verify the effectiveness of selecting representative stations for the region. Through comparison, from Table 3 and Figure 9 、 Figure 10 it can be seen that the method proposed in the present invention performs equivalently to it in terms of MAPE and MAE indicators, and performs better in terms of RMSE indicator. At the same time, it can be seen from Table 2 that compared with all comparison models, MRMR-LSTM improves the prediction time by 14.95, 31.04, 28.35, 25.01, -4.05 percentage points respectively. Since MRMR-LSTM-OnlyWS does not combine the predicted wind speed information of the prediction period and cannot guarantee the prediction accuracy, therefore, it can be verified that the method proposed in this paper is significantly superior to each comparison model in terms of computational time consumption. Combining the above results, it shows that selecting representative wind farms using the maximum relevance - minimum redundancy (MRMR) principle can not only effectively reduce the input dimension of the model, reduce computational resource consumption and modeling complexity, but also guarantee the prediction accuracy. Compared with directly using all station data, this method can effectively avoid prediction errors caused by redundant information and noise interference, and provides a more efficient and robust solution for regional wind power modeling.

[0046] Table 2 Training time of each model in Partition 2 , In the above comparison models, the comparison between All-LSTM and MRMR-TCN, MRMR-CNN-LSTM, MRMR-Bi-LSTM aims to verify the rationality of the prediction model design. From Table 2, Figure 9 、 Figure 10 it can be seen that the MRMR-LSTM model that selects representative stations based on the maximum relevance - minimum redundancy (MRMR) principle, in the comparison with the three models of MRMR-TCN, MRMR-CNN-LSTM, and MRMR-Bi-LSTM, the prediction error of the model of the present invention is improved by 3.81 percentage points compared with the comparison model, is improved by 17.23 percentage points, It has increased by 18.1 percentage points. In comparison with other methods, the model proposed in this paper demonstrates better prediction performance. This is mainly due to the model's ability to effectively model the complex and causal mapping relationship between wind speed and power, while avoiding the introduction of invalid information during the feature extraction process. Compared with structures such as convolutional neural networks (CNNs) or bidirectional long short-term memory networks (Bi-LSTMs), the time series modeling strategy adopted by the model in this paper is more in line with the physical characteristics of the wind power prediction task, capable of fully exploiting the long-term dependence characteristics in the time series to ensure the stability and accuracy of the prediction results. In addition, the model structure is relatively simple, with high training efficiency. While maintaining high prediction accuracy, it has good potential for engineering applications, verifying the rationality and practicality of its use as a regional power prediction tool.

[0047] The comparison between All-LSTM and MRMR-LSTM-OnlyWS in the above comparison models aims to verify the effectiveness of fusing predicted wind speed data for the prediction period. As can be seen from Table 3, MRMR-LSTM is significantly superior to MRMR-LSTM-OnlyWS in all prediction evaluation metrics. This is because the All-LSTM model introduces features containing wind speed information for future prediction time steps in the input, effectively enhancing the model's perception ability of future power changes and making the prediction results more accurate. In contrast, the MRMR-LSTM-OnlyWS model only relies on the measured wind speed data representing the power station and does not consider future meteorological information, making it difficult to capture the wind speed change trend and resulting in limited prediction accuracy. Through the above comparison, the importance of introducing NWP information for improving the accuracy of regional wind power prediction is verified.

[0048] Table 3 Evaluation Metrics of Each Model in Sub-region 2 。

[0049] Please refer to Figure 11 , which shows a structural block diagram of a short-term power prediction system for a wind farm cluster according to the present application.

[0050] As Figure 11 shown, the short-term power prediction system 200 for a wind farm cluster includes an acquisition module 210, a calculation module 220, a construction module 230, a summation module 240, a partitioning module 250, and an output module 260.

[0051] Among them, the acquisition module 210 is configured to acquire the power data at the 5-minute level of wind farm stations in the target area for historical sample days, the measured wind speed data at the hub of the wind turbines , where , is a three-dimensional real number space, represents the real number field, denote sample days, 288 represents the number of 5 - minute - level samples in a day, represents the number of wind farm stations, is the power sequence of the th wind farm station, is the measured wind speed sequence of the th wind farm station, is the predicted wind speed sequence of the th wind farm station; , where, is a two - dimensional real - number space, represents the number of wind farm stations; Construction module 230, configured to calculate the Copula entropy between the historical power samples of each station and construct a spatial similarity matrix of the station power according to each Copula entropy ; Summation module 240, configured to normalize the similarity distance matrix and the spatial similarity matrix , and sum the normalized similarity distance matrix and the spatial similarity matrix to obtain a spatio - temporal similarity distance matrix ; Partition module 250, configured to cluster the spatio - temporal similarity distance matrix using a hierarchical clustering algorithm, determine a partitioning scheme in combination with the geographical distribution of new - energy stations, and divide the wind farms in the target area into K regions; Output module 260, configured to establish an LSTM time - series prediction model for each region respectively, input the power sequence in a certain region, the measured wind speed sequence at the hub of the fan of the representative station in a certain region, and the predicted wind speed sequence at the hub of the fan of the representative station in a certain region into a certain LSTM time - series prediction model, and the certain LSTM time - series prediction model outputs the total power prediction result of a certain region, where the representative station in a certain region is selected in a certain region based on the maximum correlation - minimum redundancy principle.

[0052] It should be understood that Figure 11 the modules described in Figure 1 correspond to the respective steps in the method described in the reference Figure 11The modules therein will not be elaborated here.

[0053] In some other embodiments, the embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored. When the program instructions are executed by a processor, the processor is enabled to execute the short-term power prediction method based on a wind farm cluster in any of the above method embodiments. As an implementation manner, the computer-readable storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are set as follows: Obtain the power data of wind farm stations at the 5-minute level for historical sample days, the measured wind speed data at the hub of the fan and the predicted wind speed data at the hub of the fan, wherein, is a three-dimensional real number space, represents the real number field, represents sample days, 288 represents the number of 5-minute-level samples in a day, represents the number of wind farm stations, is the power sequence of the th wind farm station, is the measured wind speed sequence of the th wind farm station, is the predicted wind speed sequence of the th wind farm station; Calculate the ACF curve of the wind speed at the hub of the fan, calculate the weighted Euclidean distance between the ACF curves of different stations, and construct a similarity distance matrix of the time-varying law of key meteorological elements between different stations , wherein, is a two-dimensional real number space, represents the number of wind farm stations; Calculate the Copula entropy between the historical power samples of each station, and construct a spatial similarity matrix of the station power according to each Copula entropy ; Normalize the similarity distance matrix and the spatial similarity matrix , and sum the normalized similarity distance matrix and the spatial similarity matrix to obtain a spatio-temporal similarity distance matrix ; Use the hierarchical clustering algorithm to cluster the spatio-temporal similarity distance matrix , and determine the zoning scheme in combination with the geographical distribution of new energy stations, and the target area A wind farm is divided into K regions; An LSTM time series prediction model is established for each region. The power sequence in a certain region, the measured wind speed sequence at the hub of the wind turbines representing the power station in a certain region, and the predicted wind speed sequence at the hub of the wind turbines representing the power station in a certain region are input into a certain LSTM time series prediction model. The certain LSTM time series prediction model outputs the total power prediction result of a certain region, where the wind turbines representing the power station in a certain region are selected based on the maximum correlation - minimum redundancy principle in a certain region.

[0054] A computer - readable storage medium may include a storage program area and a storage data area. Among them, the storage program area can store an operating system and application programs required for at least one function; the storage data area can store data created according to the use of the short - term power prediction system for a wind farm cluster, etc. In addition, the computer - readable storage medium may include high - speed random - access memory, and may also include memories, such as at least one magnetic disk storage device, a flash memory device, or other non - volatile solid - state storage devices. In some embodiments, the computer - readable storage medium may optionally include a memory remotely set relative to the processor, and these remote memories can be connected to the short - term power prediction system for a wind farm cluster through a network. Examples of the above - mentioned network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and their combinations.

[0055] Figure 12 is a schematic structural diagram of the electronic device provided by an embodiment of the present invention, as Figure 12 shown. The device includes: a processor 310 and a memory 320. The electronic device may further include: an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330, and the output device 340 can be connected through a bus or other means, Figure 12 taking connection through the bus as an example. The memory 320 is the above - mentioned computer - readable storage medium. The processor 310 executes various functional applications and data processing of the server by running non - volatile software programs, instructions, and modules stored in the memory 320, that is, implements the short - term power prediction method for a wind farm cluster based on the above - mentioned method embodiments. The input device 330 can receive input digital or character information, and generate key signal inputs related to user settings and function controls of the short - term power prediction system for a wind farm cluster. The output device 340 may include display devices such as a display screen.

[0056] The above - mentioned electronic device can execute the method provided by the embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be referred to the method provided by the embodiment of the present invention.

[0057] As an implementation manner, the above-mentioned electronic device is applied to a short-term power prediction system for a wind farm cluster and is used for a client, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Obtain the power data of wind farm stations at the 5-minute level for historical sample days, the measured wind speed data at the hub of the fan and the predicted wind speed data at the hub of the fan , where is a three-dimensional real number space, represents the real number field, represents sample days, 288 represents the number of 5-minute samples in a day, represents the number of wind farm stations, is the power sequence of the th wind farm station, is the measured wind speed sequence of the th wind farm station, is the predicted wind speed sequence of the th wind farm station; Calculate the ACF curve of the wind speed at the hub of the fan, calculate the weighted Euclidean distance between the ACF curves of different stations, and construct a similarity distance matrix of the time-varying laws of key meteorological elements between different stations , where is a two-dimensional real number space, represents the number of wind farm stations; Calculate the Copula entropy between the historical power samples of each station, and construct a spatial similarity matrix of the station power according to each Copula entropy ; Normalize the similarity distance matrix and the spatial similarity matrix , and sum the normalized similarity distance matrix and the spatial similarity matrix to obtain a spatio-temporal similarity distance matrix ; Use the hierarchical clustering algorithm to cluster the spatio-temporal similarity distance matrix , and combine the geographical distribution of new energy stations to determine the zoning scheme, and divide the wind farms in the target area into K regions; An LSTM time series prediction model is established for each region. The power sequence in a certain region, the measured wind speed sequence at the hub of the wind turbines representing the power station in a certain region, and the predicted wind speed sequence at the hub of the wind turbines representing the power station in a certain region are input into a certain LSTM time series prediction model. The certain LSTM time series prediction model outputs the total power prediction result of a certain region, where the power station represented in a certain region is selected in a certain region based on the principle of maximum correlation - minimum redundancy.

[0058] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general - purpose hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above - mentioned technical solution, 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 ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiments.

[0059] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. A short-term power prediction method for a wind farm cluster, characterized in that, Including: Obtain the target area wind farm stations historical sample daily 5-minute power data , measured wind speed data at the hub of the wind turbine and predicted wind speed data at the hub of the wind turbine , where is a three-dimensional real number space represents the real number field represents sample days, 288 represents the number of 5-minute samples in a day represents the number of wind farm stations is the power sequence of the th wind farm station is the measured wind speed sequence of the th wind farm station is the predicted wind speed sequence of the th wind farm station; Calculate the ACF curve of the wind speed at the hub of the wind turbine, calculate the weighted Euclidean distance between the ACF curves of different sites, and construct a similarity distance matrix for the time-varying laws of key meteorological elements between different sites , where is a two-dimensional real number space, represents the number of wind farms; Calculate the Copula entropy between the historical power samples of each station, and construct a spatial similarity matrix of the station power based on each Copula entropy ; Normalize the similarity distance matrix and the spatial similarity matrix , and sum the normalized similarity distance matrix and the spatial similarity matrix to obtain the spatio-temporal similarity distance matrix ; Use the hierarchical clustering algorithm for the spatio-temporal similarity distance matrix to perform clustering, determine the partitioning scheme in combination with the geographical distribution of new energy power stations, and divide the wind farms in the target area into K regions; Establish an LSTM time series prediction model for each region, input the power sequence in a certain region, the measured wind speed sequence at the hub of the wind turbines representing the power station in a certain region, and the predicted wind speed sequence at the hub of the wind turbines representing the power station in a certain region into a certain LSTM time series prediction model, and the certain LSTM time series prediction model outputs the total power prediction result of a certain region, wherein the power station represented in a certain region is selected in a certain region based on the maximum correlation - minimum redundancy principle.

2. The short-term power prediction method for a wind farm cluster according to claim 1, wherein Calculate the ACF curve for the wind speed at the hub of the wind turbine, calculate the weighted Euclidean distance between the ACF curves of different sites, and construct a similarity distance matrix for the time-varying laws of key meteorological elements between different sites including: Calculate the wind speed at the hub of the wind turbines in each wind farm of the ACF curve, where the ACF value at the lag time t for the i-th station is calculated The expression is as follows: ， Wherein, is the length of the wind speed sequence, is the mean value of the measured wind speed sequence of the i-th wind farm station, , is the ACF value of the i-th wind farm station at the lag time t, is the wind speed value of the measured wind speed sequence of the i-th wind farm station at the time point , is the wind speed value of the measured wind speed sequence of the i-th wind farm station at the time point . Construct an ACF matrix, and the expression is: ， In the formula, is the ACF matrix, is the ACF value of the m-th wind farm at the lag time n; Calculate the average ACF value of all wind farms at the lag time t , and calculate the Euclidean distance weights at each lag time point based on the average ACF value. Among them, the expression for calculating the average ACF value is: ， wherein, is the ACF value of wind farm k at lag time t, is the number of wind farms; The expression for calculating the Euclidean distance weight is: ， Calculate the weighted Euclidean distance matrix between the ACF curves of each station according to the Euclidean distance weight, that is, construct the similarity distance matrix of the time-varying laws of key meteorological elements between different stations , and the expression is: ， In the formula, is the weighted Euclidean distance between the m-th wind farm and the m-th wind farm, and the value is 0. is the weighted Euclidean distance between wind farm h and wind farm j. ， Wherein, and are the autocorrelation coefficients of wind farm h and wind farm j at lag time t, respectively, is the maximum lag time of the ACF.

3. A short-term power prediction method for a wind farm cluster according to claim 1, characterized in that Calculating the Copula entropy between historical power samples of each station, and constructing a spatial similarity matrix of station power according to each Copula entropy including: Estimate the empirical marginal distribution functions of wind farm h and wind farm j respectively through the order statistic, where the order statistic is a statistic describing the relative positions of each data point in the sequence, and the expression is: ， ， wherein, is the empirical marginal distribution function of the power time series samples of the h-th wind farm, is the number of power time series samples, is the indicator function. When holds, the function value is 1. When does not hold, the function value is 0. is the s-th sample in the power time series of the h-th wind farm, is the s-th sample in the power time series of the j-th wind farm, is the power time series of the wind farm; The wind farm and the wind farm The power time series samples are mapped to marginal probabilities through the empirical distribution function, and the expression is: ， ， Wherein, is the t-th sample in the power sequence of the h wind farm, is the t-th sample of the power sequence of the j wind farm, is the empirical marginal distribution function of the power sequence of the j wind farm, is the empirical marginal distribution function of the power sequence of the h wind farm, is the marginal probability of the t-th sample in the power sequence of the h wind farm, is the marginal probability of the t-th sample in the power sequence of the j wind farm; Construct joint sample pairs based on marginal probabilities , and calculate the distance between sample pairs according to the Chebyshev distance, and the expression is: ， ， Wherein, is the Chebyshev distance between the s-th sample pair and the v-th sample pair ; , is the marginal probability of the s-th sample in the wind farm h for the sample pair ; is the marginal probability of the v-th sample in the wind farm h for the sample pair ; is the marginal probability of the s-th sample in the wind farm j for the sample pair ; is the marginal probability of the v-th sample in the wind farm j for the sample pair ; is the s-th power sample, and is the v-th power sample;​​​​ For each sample pair, find the k - th nearest neighbor distance of each sample pair, and estimate the Copula entropy between the historical power samples of each power station based on the k - th nearest neighbor distance, and the expression is: ， In the formula, is the volume of the two-dimensional unit sphere. Under the Chebyshev distance, = 1, is the Chebyshev distance of the k-th nearest neighbor of the sample pair s, is the Copula entropy between station h and station j, is the Digamma function; Composed of a spatial correlation matrix representing non-linearity between two stations, that is, a spatial similarity matrix of the station power is constructed , and the expression is: ， Wherein, is the Copula entropy between station h and station j, is the Copula entropy between station m and station m.

4. A short-term power prediction method for a wind farm cluster according to claim 1, characterized in that, The spatio-temporal similarity distance matrix has the following expression: ， In the formula, is the normalized Copula entropy matrix, is the weighted Euclidean distance matrix between the ACF curves of each station after normalization, is the spatio-temporal similarity distance between station h and station j, is the spatio-temporal similarity distance between the m-th wind farm and the m-th wind farm.

5. A short-term power prediction method for a wind farm cluster according to claim 1, characterized in that, Before establishing the LSTM time series prediction model for each region respectively, the method further includes: In the k-th region, calculate the measured wind speed sequences at the hubs of the wind turbines in each wind farm among wind farms and the total regional power The mutual information between them is expressed as: ， wherein, is the mutual information between the measured wind speed sequence at the wind turbine hub of wind farm b and the total regional power, is the marginal probability function of the measured wind speed at the wind turbine hub of wind farm b, is the marginal probability function of the total wind power, is the joint probability distribution between the marginal probability function of the measured wind speed at the wind turbine hub of wind farm b and the total wind power, is the total power of region k, is the measured wind speed sequence at the wind turbine hub of wind farm b, is a specific observed value of the measured wind speed at the wind turbine hub of wind farm b; Store the calculation result as a correlation vector , select the first representative station with the largest correlation with the total regional power , As the correlation vector, add to the set of representative stations , and the remaining set of stations is . For each candidate station u ∈ U, calculate the MRMR score , is the wind speed at the hub of the candidate wind farm in region k, is the wind speed at the hub of the selected representative wind farm in region k, is the mutual information between the wind speed at the hub of the candidate wind farm in region k and the total regional power in region k, is the mutual information between the wind speed at the hub of the candidate wind farm in region k and the wind speed at the hub of the selected representative wind farm in region k. Select the station with the highest MRMR score to join the representative set. When , output the set of representative stations as the representative stations in region k 6. A short-term power prediction system based on a wind farm cluster, characterized in that, Including: An acquisition module, configured to acquire a target area wind farm stations historical sample-day 5-minute power data , measured wind speed data at the wind turbine hub and predicted wind speed data at the wind turbine hub , where is a three-dimensional real space, represents the real number field, represents sample days, 288 represents the number of 5-minute samples in a day, represents the number of wind farm stations, is the power sequence of the th wind farm station, is the measured wind speed sequence of the th wind farm station, is the predicted wind speed sequence of the th wind farm station; A calculation module, configured to calculate the ACF curve of the wind speed at the hub of the wind turbine, calculate the weighted Euclidean distance between the ACF curves of different sites, and construct a similarity distance matrix of the time-varying laws of key meteorological elements between different sites , where is a two-dimensional real number space, represents the number of wind farms A building block configured to calculate the Copula entropy between historical power samples of each station and construct a spatial similarity matrix of the station power according to each Copula entropy ; A summation module configured to sum the similarity distance matrix and the spatial similarity matrix after normalization, and sum the normalized similarity distance matrix and the spatial similarity matrix to obtain a spatio-temporal similarity distance matrix ; The partitioning module is configured to perform clustering on the spatio-temporal similarity distance matrix by using a hierarchical clustering algorithm to determine a partitioning scheme in combination with the geographical distribution of new energy power stations, and divide the target area wind farms into K regions; An output module, configured to establish an LSTM time series prediction model for each region respectively, input the power sequence in a certain region, the measured wind speed sequence at the hub of the wind turbines representing the power station in a certain region, and the predicted wind speed sequence at the hub of the wind turbines representing the power station in a certain region into a certain LSTM time series prediction model, and the certain LSTM time series prediction model outputs the total power prediction result of a certain region, wherein the power station represented in a certain region is selected in a certain region based on the maximum correlation - minimum redundancy principle.

7. An electronic device, characterized in that, Including: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method according to any one of claims 1 to 5.

Images