A multi-model coupled wind power prediction method, device, equipment and storage medium

Abstract

The present invention relates to the technical field of wind power, and discloses a multi-model coupled wind power prediction method, device, equipment and storage medium. The present invention analyzes the operating behaviors of multiple wind turbine groups in combination with the initial historical data sets, geographical location information sets and spatial topological relationships corresponding to the multiple wind turbine groups, taking into account the spatial relationships and semantic similarities between the wind turbines, and improving the non-uniformity of the spatial characteristics of the analysis results of the operating behaviors of the wind turbine groups. Further, a wind speed control model is used to select multiple target prediction models from the multiple constructed models for predicting the wind power of the multiple wind turbine groups, improving the robustness and stability of the prediction results. At the same time, the wind speed control model is combined in the prediction process for prediction, taking into account the differences in prediction effects under different wind speed conditions, and being able to effectively reduce the deviation of the prediction results at high wind speeds or low wind speeds. Therefore, by implementing the present invention, real-time ultra-short-term power prediction of multiple wind turbine groups is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power, and in particular to a multi-model coupled wind power prediction method, device, equipment and storage medium. Background Art

[0002] Wind power generation is a clean and renewable energy utilization method. However, due to the uncertainty and randomness of meteorological factors such as wind speed and wind direction, the power output of wind turbines has strong volatility and intermittency. This brings great challenges to the operation of the power grid, and accurate power prediction of wind turbines is required to improve the economic benefits of wind farms and the safety and stability of the power grid. Wind turbines are more significantly affected by meteorological changes in areas with frequent weather mutations. Currently, short-term power prediction methods have certain limitations and are difficult to handle complex and sudden weather. When the wind conditions suddenly change, due to the lag of the control system, it is easy to cause the load of the unit to be too large or even the unit to shut down, resulting in significant economic losses. Ultra-short-term power prediction is of great significance.

[0003] Currently, there are some methods based on machine learning or deep learning to predict the power of wind power generation, such as decision trees, support vector machines, recurrent neural networks, etc. However, these methods have the following defects: (1) ignoring the spatial relationship and semantic similarity between wind turbines, resulting in the non-uniformity of the spatial characteristics of the prediction results; (2) not considering the difference in prediction effects under different wind speed conditions, resulting in large deviations in the prediction results at high wind speeds or low wind speeds; (3) not fully using the prediction results of multiple models for fusion, resulting in the lack of robustness and stability of the prediction results. Summary of the Invention

[0004] In view of this, the present invention provides a multi-model coupled wind power prediction method, device, equipment and storage medium to solve the above defects existing in the method of predicting the power of wind power generation based on machine learning or deep learning.

[0005] In a first aspect, the present invention provides a multi-model coupled wind power prediction method, which includes:

[0006] Obtain the initial historical data set, geographical location information set, and spatial topological relationship corresponding to multiple wind turbine groups; analyze the operating behaviors of multiple wind turbine groups based on the initial historical data set, geographical location information set, and spatial topological relationship to obtain the analysis results of the operating behaviors of wind turbine groups; respectively establish a spatio-temporal attention network model, a LightGBM model, a GRU model, and a Local-Ensemble model based on the initial historical data set and the analysis results of the operating behaviors of wind turbine groups; establish a wind speed control model based on the initial historical data set; determine multiple target prediction models among the spatio-temporal attention network model, the LightGBM model, the GRU model, and the Local-Ensemble model based on the wind speed control model; predict the wind power of multiple wind turbine groups based on the wind speed control model and the multiple target prediction models to obtain the target prediction result of wind power.

[0007] The multi-model coupled wind power prediction method provided by the present invention analyzes the operating behaviors of multiple wind turbine groups by combining the initial historical data set, geographical location information set, and spatial topological relationship corresponding to multiple wind turbine groups, considers the spatial relationship and semantic similarity between wind turbines, and improves the non-uniformity of the spatial characteristics of the analysis results of the operating behaviors of wind turbine groups. Further, the wind speed control model is used to select multiple target prediction models from the constructed multiple models to predict the wind power of multiple wind turbine groups, which improves the robustness and stability of the prediction results. At the same time, the wind speed control model is combined in the prediction process for prediction, considering the difference in prediction effects under different wind speed conditions, and can effectively reduce the deviation of the prediction results at high wind speeds or low wind speeds. Therefore, by implementing the present invention, real-time ultra-short-term power prediction of multiple wind turbine groups is achieved.

[0008] In an optional implementation manner, analyzing the operating behaviors of multiple wind turbine groups based on the initial historical data set, geographical location information set, and spatial topological relationship to obtain the analysis results of the operating behaviors of wind turbine groups includes:

[0009] Performing similarity partitioning on multiple wind turbine groups based on the initial historical data set, geographical location information set, and spatial topological relationship to obtain a similarity wind turbine partitioning result; processing the initial historical data set based on the similarity wind turbine partitioning result to obtain a target historical data set; identifying abnormal points and analyzing the operating behaviors of the wind power of multiple wind turbine groups based on the target historical data set to obtain the analysis results of the operating behaviors of wind turbine groups.

[0010] The multi-model coupled wind power prediction method provided by the present invention can automatically divide similar wind turbines by combining the initial historical data set, geographical location information set, and spatial topological relationship. Further, by combining the results of similar wind turbine division, anomaly point identification and operation behavior analysis are performed on the wind power of multiple wind turbine groups, improving the wind turbine monitoring and management capabilities. At the same time, the spatial relationship and semantic similarity between wind turbines are considered, improving the non-uniformity of the spatial characteristics of the operation behavior analysis results of wind turbine groups.

[0011] In an alternative embodiment, based on the initial historical data set, geographical location information set, and spatial topological relationship, similar division is performed on multiple wind turbine groups to obtain the similar wind turbine division results, including:

[0012] Based on different time dimensions, the historical data set is processed and clustered to obtain the clustering results; based on the geographical location information set and spatial topological relationship, through graph construction method processing, the graph construction results are obtained; based on the clustering results and graph construction results, similar division and analysis are performed on multiple wind turbine groups to obtain the similar wind turbine division results.

[0013] The multi-model coupled wind power prediction method provided by the present invention can first obtain the clustering results under different time dimensions by processing and clustering the historical data set in different time dimensions. Secondly, graph construction can be performed by combining the geographical location information set and spatial topological relationship, considering the spatial relationship and semantic similarity between wind turbines. Finally, by combining the clustering results and graph construction results, similar division of multiple wind turbine groups can be achieved, providing support for subsequent operation behavior analysis of multiple wind turbine groups.

[0014] In an alternative embodiment, based on the initial historical data set and the operation behavior analysis results of wind turbine groups, a spatio-temporal attention network model, a LightGBM model, a GRU model, and a Local-Ensemble model are respectively established, including:

[0015] Construct three-dimensional tensor data based on the initial historical data set; based on the three-dimensional tensor data and the operation behavior analysis results of wind turbine groups, establish a spatio-temporal attention network model; based on the initial historical data set and the operation behavior analysis results of wind turbine groups, respectively establish a LightGBM model, a GRU model, and a Local-Ensemble model.

[0016] The multi-model coupled wind power prediction method provided by the present invention can construct three-dimensional tensor data through the initial historical data set, and establish the corresponding spatio-temporal attention network model by combining the operation behavior analysis results of wind turbine groups. Further, by combining the initial historical data set and the operation behavior analysis results of wind turbine groups, multiple corresponding prediction models can be respectively established, providing support for subsequent fusion of multi-model prediction results.

[0017] In an alternative embodiment, the wind power of multiple wind turbine groups is predicted based on a wind speed control model and multiple target prediction models to obtain a wind power target prediction result, including:

[0018] Based on the wind speed control model, determine the wind speed condition; use multiple target prediction models to predict multiple wind turbine groups to obtain multiple initial wind power prediction results; based on the wind speed condition and multiple initial wind power prediction results, determine the wind power target prediction result.

[0019] The multi-model coupled wind power prediction method provided by the present invention adopts a multi-model coupling method, makes full use of the prediction results of multiple models for complementarity and optimization, and improves the robustness and stability of the prediction results. At the same time, combining the wind speed control model for prediction takes into account the prediction effect differences under different wind speed conditions, and can effectively reduce the deviation of the prediction results at high wind speeds or low wind speeds.

[0020] In an alternative embodiment, the method further includes:

[0021] Obtain the measured power data set of multiple wind turbine groups; update multiple target prediction models based on the wind power target prediction result and the measured power data set.

[0022] The multi-model coupled wind power prediction method provided by the present invention can perform real-time analysis, management, and update on the prediction results of the multi-model coupled wind power prediction model by combining the wind power target prediction result and the measured power data set, can effectively reflect the changes in the current data distribution, and then update multiple target prediction models to improve the real-time performance of the prediction results.

[0023] In a second aspect, the present invention provides a multi-model coupled wind power prediction device, which includes:

[0024] A first acquisition module for acquiring an initial historical data set, a geographical location information set, and a spatial topological relationship corresponding to multiple wind turbine groups; an analysis module for analyzing the operating behaviors of multiple wind turbine groups based on the initial historical data set, the geographical location information set, and the spatial topological relationship to obtain an analysis result of the operating behaviors of the wind turbine groups; a first establishment module for respectively establishing a spatio-temporal attention network model, a LightGBM model, a GRU model, and a Local-Ensemble model based on the initial historical data set and the analysis result of the operating behaviors of the wind turbine groups; a second establishment module for establishing a wind speed control model based on the initial historical data set; a determination module for determining multiple target prediction models from the spatio-temporal attention network model, the LightGBM model, the GRU model, and the Local-Ensemble model based on the wind speed control model; a prediction module for predicting the wind power of multiple wind turbine groups based on the wind speed control model and multiple target prediction models to obtain a wind power target prediction result.

[0025] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the multi-model coupled wind power prediction method according to the first aspect or any corresponding embodiment thereof.

[0026] In a fourth aspect, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to perform the multi-model coupled wind power prediction method according to the first aspect or any corresponding embodiment thereof.

[0027] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, and the computer instructions are used to cause a computer to perform the multi-model coupled wind power prediction method according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description 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.

[0029] Figure 1 is a schematic flowchart of a multi-model coupled wind power prediction method according to an embodiment of the present invention;

[0030] Figure 2 is a schematic flowchart of another multi-model coupled wind power prediction method according to an embodiment of the present invention;

[0031] Figure 3 is a schematic flowchart of yet another multi-model coupled wind power prediction method according to an embodiment of the present invention;

[0032] Figure 4 is a structural block diagram of a multi-model coupled wind power prediction device according to an embodiment of the present invention;

[0033] Figure 5 is a schematic hardware structure diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] 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.

[0035] The embodiments of the present invention provide a multi-model coupled wind power prediction method, which improves the non-uniformity of the spatial characteristics of the prediction results by considering the spatial relationship and semantic similarity between wind turbines, effectively reduces the deviation of the prediction results at high or low wind speeds by considering the difference in prediction effects under different wind speed conditions, and improves the robustness and stability of the prediction results by fusing multiple models. Therefore, by implementing the present invention, real-time ultra-short-term power prediction of multiple wind turbine groups is achieved.

[0036] According to the embodiments of the present invention, an embodiment of a multi-model coupled wind power prediction method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0037] In this embodiment, a multi-model coupled wind power prediction method is provided, which can be used in electronic devices such as computers, mobile phones, and tablet computers. Figure 1 is a flowchart of the multi-model coupled wind power prediction method according to the embodiments of the present invention, as Figure 1 shown, and the process includes the following steps:

[0038] Step S101, obtain the initial historical data set, geographical location information set, and spatial topological relationship corresponding to multiple wind turbine groups.

[0039] Among them, the initial historical data set may include the meteorological data (wind speed, wind direction, temperature, etc.), status data (power, excitation current, etc.) of each wind turbine group, and the data of the wind farm anemometer tower.

[0040] Exemplarily, in the past year, taking a sampling frequency of 10 minutes, historical data of each wind turbine group is obtained from the SCADA (Supervisory Control And Data Acquisition) system of the wind farm, which may include meteorological parameters (such as wind speed, wind direction, and temperature), unit status parameters (such as power and excitation current), and data of the wind farm's anemometer tower. Ensure the standardized processing of the data, integrate it in the form of a time series, and store the data independently for each wind turbine group, considering their different sampling frequencies and data lengths, as shown in Table 1 below:

[0041] Table 1

[0042] Serial Number Field Name Field Description 1 TurbID Turbine Generator ID 2 Day Sampling Date 3 Tmstamp Sampling Time 4 Wspd (m / s) Wind Speed Recorded by Anemometer 5 Wdir (°) Angle between Wind Direction and Turbine Generator Nacelle Position 6 Etmp (℃) Ambient Temperature 7 Itmp (℃) Internal Temperature of Turbine Generator Nacelle 8 Ndir (°) Nacelle Direction, i.e., Yaw Angle of Nacelle 9 Pab1 (°) Pitch Angle of Blade 1 10 Pab2 (°) Pitch Angle of Blade 2 11 Pab3 (°) Pitch Angle of Blade 3 12 Prtv (kW) Reactive Power 13 Patv (kW) Active Power

[0043] Furthermore, collect and record the geographical location information and spatial topological relationship of the wind turbine groups, which may include specific geographical locations (latitude, longitude, altitude) and spatial relative positions (distance, relative position, obstacles).

[0044] Step S102, based on the initial historical data set, geographical location information set, and spatial topological relationship, analyze the operating behaviors of multiple wind turbine groups to obtain the analysis results of the operating behaviors of the wind turbine groups.

[0045] Specifically, by combining the initial historical data sets, geographical location information sets, and spatial topological relationships corresponding to multiple wind turbine groups to analyze the operating behaviors of multiple wind turbine groups, the spatial relationship and semantic similarity between the wind turbines are considered, improving the non-uniformity of the spatial characteristics of the analysis results of the operating behaviors of the wind turbine groups.

[0046] Step S103, based on the initial historical data set and the analysis results of the operating behaviors of the wind turbine groups, establish a spatio-temporal attention network model, a LightGBM model, a GRU model, and a Local-Ensemble model respectively.

[0047] Specifically, there are already some methods based on machine learning or deep learning. When predicting the power of wind power generation, the prediction results of multiple models are not fully utilized for fusion, resulting in the lack of robustness and stability of the prediction results. Therefore, in this embodiment, by combining the initial historical data set and the initial prediction results of the wind turbine groups, different multiple models are established respectively, which may include a spatio-temporal attention network model, a LightGBM (Light Gradient Boosting Machine) model, a GRU (Gated Recurrent Unit) model, and a Local-Ensemble model.

[0048] Step S104, based on the initial historical data set, establish a wind speed control model.

[0049] Specifically, by observing the initial historical data set, representative features are extracted and a wind speed control model is constructed.

[0050] Furthermore, during the construction of the model, both the feature extraction and the model selection have fully considered the spatial information, which can further support the improvement of the accuracy of wind speed prediction.

[0051] Step S105: Based on the wind speed control model, determine multiple target prediction models from the spatio-temporal attention network model, LightGBM model, GRU model, and Local-Ensemble model.

[0052] Specifically, according to the current wind speed conditions identified by the wind speed control model, corresponding multiple models can be selected from the constructed multiple models as the final prediction models.

[0053] For example, when the wind speed control model identifies that the current prediction instance belongs to the high wind speed condition, a high wind speed optimization model is selected for prediction. The high wind speed optimization model adds a loss function of Lasso regression on the basis of the conventional model to penalize the absolute value of the model parameters to prevent overfitting and improve the fitting ability for high wind speed data.

[0054] Furthermore, according to the classification result, a corresponding prediction model is selected for the current prediction instance, that is, when the instance is judged to be in the high wind speed condition, a high wind speed model with enhanced overfitting is selected for prediction, otherwise a conventional model is selected.

[0055] Among them, both the high wind speed model and the conventional model are determined from the above-mentioned constructed spatio-temporal attention network model, LightGBM model, GRU model, and Local-Ensemble model.

[0056] Step S106: Based on the wind speed control model and multiple target prediction models, predict the wind power of multiple wind turbine groups to obtain the target prediction result of wind power.

[0057] Specifically, using multiple target prediction models to predict the wind power of multiple wind turbine groups improves the robustness and stability of the prediction results. At the same time, combining the wind speed control model in the prediction process takes into account the differences in prediction effects under different wind speed conditions, and can effectively reduce the deviation of the prediction results at high wind speed or low wind speed.

[0058] The multi-model coupled wind power prediction method provided in this embodiment analyzes the operating behaviors of multiple wind turbine groups by combining the initial historical data sets, geographical location information sets, and spatial topological relationships corresponding to the multiple wind turbine groups, taking into account the spatial relationships and semantic similarities between the wind turbines, and improving the non-uniformity of the spatial characteristics of the analysis results of the operating behaviors of the wind turbine groups. Further, a wind speed control model is used to select multiple target prediction models from the constructed multiple models to predict the wind power of the multiple wind turbine groups, improving the robustness and stability of the prediction results. At the same time, the wind speed control model is combined in the prediction process for prediction, taking into account the differences in prediction effects under different wind speed conditions, and being able to effectively reduce the deviations that occur in the prediction results at high wind speeds or low wind speeds. Therefore, by implementing the present invention, real-time ultra-short-term power prediction of multiple wind turbine groups is achieved.

[0059] In this embodiment, a multi-model coupled wind power prediction method is provided, which can be used in electronic devices such as computers, mobile phones, and tablet computers. Figure 2 It is a flowchart of the multi-model coupled wind power prediction method according to an embodiment of the present invention, as Figure 2 shown, and this process includes the following steps:

[0060] Step S201, obtain the initial historical data sets, geographical location information sets, and spatial topological relationships corresponding to multiple wind turbine groups. For details, please refer to Figure 1 step S101 of the embodiment shown, which will not be elaborated here.

[0061] Step S202, based on the initial historical data sets, geographical location information sets, and spatial topological relationships, analyze the operating behaviors of multiple wind turbine groups to obtain the analysis results of the operating behaviors of the wind turbine groups.

[0062] Specifically, the above step S202 includes:

[0063] Step S2021, based on the initial historical data sets, geographical location information sets, and spatial topological relationships, perform similarity partitioning on multiple wind turbine groups to obtain a similarity wind turbine partitioning result.

[0064] Specifically, by combining the initial historical data sets, geographical location information sets, and spatial topological relationships corresponding to multiple wind turbine groups to perform similarity partitioning on the multiple wind turbine groups, multiple similar wind turbine groups can be partitioned together.

[0065] In some alternative embodiments, the above step S2021 includes:

[0066] Step a1, process and cluster the historical data set based on different time dimensions to obtain a clustering result.

[0067] Specifically, according to different time dimensions, such as daily dimension, weekly dimension, monthly dimension, etc., the operation data of wind turbines is cleaned to improve the quality and usability of the data. At the same time, a distance analysis method based on Pearson similarity is used for feature dimensionality reduction to screen out features with high correlation with the operation behavior of wind turbines, reducing the dimension and redundancy of the data:

[0068] (1) Outlier detection: Conduct descriptive statistical analysis on the original data to observe the distribution of each attribute, such as mean, standard deviation, maximum value, minimum value, quartiles, etc. Then, according to the physical characteristics and working principle of the wind turbine, set some abnormal conditions, such as power output less than 0, mismatch between wind speed and power output, too high or too low temperature, etc. Next, traverse all data records. If a record meets any abnormal condition, mark it as an outlier and set its attributes except TurbId and Day to Nan. Finally, count and output the number and proportion of outliers for subsequent processing

[0069] (2) Feature selection and dimensionality reduction (application of Pearson correlation coefficient): After completing the outlier processing, perform feature selection on the clean data set. Use the Pearson correlation coefficient to analyze the correlation between each feature and the target variable (such as power generation efficiency or output power). Screen out features with high correlation to reduce the dimension and redundancy of the data.

[0070] Furthermore, slice the processed data above and use a clustering algorithm to perform clustering analysis on the sliced data to obtain similar wind turbines under different time dimensions.

[0071] Among them, slicing the data means splitting the data into different time periods, and the data within each time period is used as a separate data set.

[0072] Specifically, according to different clustering objectives and data characteristics, select a suitable clustering algorithm to perform clustering analysis on the operation data of wind turbines. In this embodiment, six clustering algorithms, namely K-means, K-medoids, K-shape, hierarchical clustering, spectral clustering, and Gaussian mixture model, are used to perform clustering on the operation data of wind turbines respectively, and the advantages, disadvantages, and applicability of each clustering algorithm are compared and analyzed.

[0073] Furthermore, the analysis and comparison of the clustering results of different clustering algorithms are shown in Table 2 below:

[0074] Table 2. Comparison of clustering results of different algorithms

[0075] Clustering Algorithm Number of Clusters Silhouette Coefficient CH Index DB Index K - means Clustering 5 0.36 47.48 0.58 K - medoids Clustering 4 0.56 44.55 0.68 K - shape Clustering 3 0.31 1.61 4.15 Hierarchical Clustering 5 0.56 47.48 0.58 Spectral Clustering 2 0.027 1.92 3.46 Gaussian Mixture Model Clustering 3 0.53 50.02 0.67

[0076] Step a2, based on the geographical location information set and spatial topological relationship, through the graph construction method, the graph construction result is obtained.

[0077] Specifically, the spatial relationship and semantic similarity between wind turbines are processed through a graph construction method to construct two different graph structures as the input for graph convolution operations, including the following operations:

[0078] Generate a geographical distance graph: According to the relative coordinates of the wind turbines provided by the dataset, calculate the Euclidean distance between each pair of wind turbines to obtain a geographical distance matrix. Then, based on a threshold, the geographical distance graph can reflect the upstream and downstream relationship between wind turbines and has an important impact on wind power generation.

[0079] Generate a semantic distance graph: According to the historical power time series of the wind turbines, calculate the dynamic time warping (DTW) distance between each pair of wind turbines to obtain a semantic distance matrix. Then, select several wind turbines that are most similar to each wind turbine as its neighboring nodes, thereby constructing a semantic distance graph that reflects the similarity of the power time series between wind turbines. The semantic distance graph can reflect the similarity of the power time series between wind turbines and has a potential association with wind power generation.

[0080] Step a3, based on the clustering results and graph construction results, perform similarity partitioning and analysis on multiple wind turbine groups to obtain the similarity wind turbine partitioning results.

[0081] Specifically, according to the obtained clustering results and graph construction results, divide multiple wind turbine groups into different clusters, and the wind turbines in each cluster have similar operating behavior characteristics. According to the number of wind turbines in different clusters and the distribution of operating data, determine the partitioning criteria for similar wind turbines and give a list of similar wind turbines.

[0082] Furthermore, after dividing multiple wind turbine groups into different similar groups, by analyzing the behavior changes and characteristics of similar wind turbines in each time dimension, it can help identify the operating behavior patterns and rules of wind turbines at different time scales, such as seasonal changes, time correlations, etc.

[0083] Step S2022, based on the similarity wind turbine partitioning results, process the initial historical dataset to obtain the target historical dataset.

[0084] Specifically, the data processing includes the following operations:

[0085] (1) Completing missing data values: According to the similarity wind turbine partitioning results, for each attribute column, fill in the Nan values according to the time stamp dimension. The filling method is to replace them with the average value of other non-Nan values in the group. The purpose of doing this is to utilize the time correlation, assuming that wind turbines within the same time period have similar data change trends. Finally, check and output whether there are still missing values or outliers in the filled data for subsequent analysis.

[0086] (2) Outlier filtering: The method based on the masked loss function is adopted to ignore the influence of irregular data. The masked loss function is an adaptive loss function. It can dynamically adjust the weight of the loss function according to the difference between the true value and the predicted value of the data, so that the contribution of outliers or extreme values to the loss function is reduced, thereby reducing the interference of outliers on model training.

[0087] Specifically, the masked loss function can be expressed as the following relational expression:

[0088]

[0089] In the formula: ||P atvij -P atvdij || r represents the error between the predicted value and the true value of the i-th wind turbine at the j-th time point. If this error exceeds a threshold, this item is ignored or given a smaller weight; r represents the exponent of the error, which can be adjusted according to the data distribution; WN represents the number of wind turbines; YB represents the prediction step.

[0090] Step S2023, based on the target historical data set, identify outliers and analyze the operating behavior of multiple wind turbine groups to obtain the analysis results of the operating behavior of the wind turbine groups.

[0091] Specifically, the K-means clustering algorithm is used to identify outliers in the operating data of wind turbines, that is, the data points far from the cluster center are regarded as outliers, and the proportion of outliers in each cluster is calculated to reflect the abnormality degree of the operating data of wind turbines.

[0092] Furthermore, according to the outlier identification results, analyze the abnormal operating behavior of the wind turbines, that is, infer the possible abnormal causes and abnormal types corresponding to the outliers according to the cluster and eigenvalue where the outliers are located, such as faults, noise interference, extreme weather, etc., and give corresponding treatment suggestions.

[0093] Step S203, based on the initial historical data set and the analysis results of the operating behavior of the wind turbine groups, establish a spatio-temporal attention network model, a LightGBM model, a GRU model, and a Local-Ensemble model respectively.

[0094] Specifically, the above step S203 includes:

[0095] Step S2031, construct three-dimensional tensor data based on the initial historical data set.

[0096] Specifically, the three-dimensional tensor data is constructed in the form of a sliding window as the input of the network model, such as [batch, sequence length, number of features].

[0097] Step S2032, establishing a spatiotemporal attention network model based on the three-dimensional tensor data and the wind turbine operation behavior analysis results.

[0098] Specifically, a sequence-to-sequence model based on the spatiotemporal attention network is trained, which mainly consists of three parts: spatial information fusion module, encoder module and decoder module:

[0099] Spatial information fusion module: Through the K nearest neighbor algorithm, according to the geographic distance map and semantic distance map, the K neighboring wind turbines closest to the target wind turbine are found, and the environmental characteristics of the neighboring wind turbines are used as a supplement to the characteristics of the target wind turbine; the connection or spatial attention mechanism is used to fuse the characteristics of the target wind turbine and its neighboring wind turbines to obtain the feature vector after spatial information fusion.

[0100] Encoder module: Process the feature vector after spatial information fusion, extract the temporal dependency information in the historical data, and generate a context vector as the initial input of the decoder module; use TCN or GRU as the main structure of the encoder module. TCN can capture longer-range temporal dependency information, and GRU can reduce the number of parameters and computational complexity.

[0101] Decoder module: Based on the context vector and other auxiliary features, it gradually outputs the power prediction results of the target wind turbine in the future period of time; GRU is used as the main structure of the decoder module, and a multi-layer perceptron network is introduced to coordinate the dimensional matching between the encoder module and the decoder module, and enhance the nonlinear ability of the model; at the same time, a temporal attention mechanism is introduced, using all the moment vectors output by the encoder module and the hidden vectors of the decoder module to calculate the attention weight of each moment, and obtain a weighted average temporal information vector as the auxiliary input of the decoder module.

[0102] Step S2033, based on the initial historical data set and the wind turbine group operation behavior analysis results, a LightGBM model, a GRU model and a Local-Ensemble model are established respectively.

[0103] Specifically, the LightGBM model, GRU model and Local-Ensemble model can be established respectively by combining the initial historical data set and the analysis results of the wind turbine operation behavior:

[0104] LightGBM module: Use the decision tree algorithm based on gradient boosting to train the feature-engineered data, use the multi-step output strategy to predict the power output of the next 288 data points in sequence, and generate 288 LightGBM models. For each model, use the current data and the past 6 data as input X, and use the power output of a future data point as the label Y.

[0105] GRU module: Use the recurrent neural network algorithm based on the PaddlePaddle framework to train the data after feature engineering. Use the multi-step input and multi-step output mode for multi-step prediction and generate GRU models with different step lengths. For each model, use a number of past data as input X and a number of future data as output Y. Considering the impact of data integrity and quality, this module trains each wind turbine separately and outputs an independent model.

[0106] Local-Ensemble module: Design a local ensemble method, including three units: LinearRegression (linear regression), PolynomialFeatures (polynomial regression) + LinearRegression, and Local_GRU. LinearRegression and PolynomialFeatures + LinearRegression use the algorithm based on linear regression to train the data after feature engineering. Use the current data and the past 12 data as input X and the future 288 data as label Y. Local_GRU uses the recurrent neural network algorithm based on the PaddlePaddle framework to train the data after feature engineering. Use the current data and a number of past data as input X and a number of future data as output Y. This module uses the whole station data for training and outputs a whole station model.

[0107] Furthermore, by combining the analysis results of the operation behavior of the wind turbine groups, the prediction accuracy of each constructed model can be improved.

[0108] Furthermore, by constructing multiple models, the potential limitations of the sequence-to-sequence model based on the spatio-temporal attention network constructed in step S2032 are supplemented.

[0109] Step S204, based on the initial historical data set, establish a wind speed control model. For details, please refer to Figure 1 Step S104 of the illustrated embodiment, which will not be elaborated here.

[0110] Step S205, based on the wind speed control model, determine multiple target prediction models among the spatio-temporal attention network model, LightGBM model, GRU model, and Local-Ensemble model. For details, please refer to Figure 1 Step S105 of the illustrated embodiment, which will not be elaborated here.

[0111] Step S206, based on the wind speed control model and multiple target prediction models, predict the wind power of multiple wind turbine groups to obtain the target prediction result of the wind power. For details, please refer toFigure 1 Step S106 of the illustrated embodiment will not be elaborated herein.

[0112] For the multi-model coupled wind power prediction method provided in this embodiment, firstly, by processing and clustering the historical data set in different time dimensions, clustering results in different time dimensions can be obtained. Secondly, graph construction can be performed by combining the geographical location information set and the spatial topological relationship, taking into account the spatial relationship and semantic similarity between wind turbines. Finally, similarity division of multiple wind turbine groups can be achieved by combining the clustering results and the graph construction results. Further, by combining the similarity wind turbine division results, outlier identification and operation behavior analysis of the wind power of multiple wind turbine groups are carried out, improving the wind turbine monitoring and management ability. At the same time, considering the spatial relationship and semantic similarity between wind turbines, the non-uniformity of the spatial characteristics of the operation behavior analysis results of wind turbine groups is improved. Further, a wind speed control model is used to select multiple target prediction models from the constructed multiple models to predict the wind power of multiple wind turbine groups, improving the robustness and stability of the prediction results. At the same time, the wind speed control model is combined in the prediction process for prediction, taking into account the prediction effect differences under different wind speed conditions, and being able to effectively reduce the deviation of the prediction results at high wind speeds or low wind speeds. Therefore, by implementing the present invention, real-time ultra-short-term power prediction of multiple wind turbine groups is achieved.

[0113] In this embodiment, a multi-model coupled wind power prediction method is provided, which can be used in electronic devices such as computers, mobile phones, and tablet computers. Figure 3 is a flowchart of the multi-model coupled wind power prediction method according to an embodiment of the present invention, as Figure 3 shown, and this process includes the following steps:

[0114] Step S301, obtain the initial historical data set, geographical location information set, and spatial topological relationship corresponding to multiple wind turbine groups. For details, please refer to Figure 1 Step S101 of the illustrated embodiment will not be elaborated herein.

[0115] Step S302, based on the initial historical data set, geographical location information set, and spatial topological relationship, analyze the operation behavior of multiple wind turbine groups to obtain the wind turbine group operation behavior analysis results. For details, please refer to Figure 2 Step S202 of the illustrated embodiment will not be elaborated herein.

[0116] Step S303, based on the initial historical data set and the wind turbine group operation behavior analysis results, establish a spatio-temporal attention network model, a LightGBM model, a GRU model, and a Local-Ensemble model respectively. For details, please refer to Figure 2 Step S203 of the illustrated embodiment will not be elaborated herein.

[0117] Step S304: Establish a wind speed control model based on the initial historical data set. For details, please refer to Figure 1 Step S104 of the embodiment shown, which will not be elaborated here.

[0118] Step S305: Determine multiple target prediction models from the spatio-temporal attention network model, LightGBM model, GRU model, and Local-Ensemble model based on the wind speed control model. For details, please refer to Figure 1 Step S105 of the embodiment shown, which will not be elaborated here.

[0119] Step S306: Predict the wind power of multiple wind turbine groups based on the wind speed control model and multiple target prediction models to obtain the target prediction result of wind power.

[0120] Specifically, the above Step S306 includes:

[0121] Step S3061: Determine the wind speed condition based on the wind speed control model.

[0122] Specifically, the current wind speed condition can be identified according to the wind speed control model.

[0123] Step S3062: Use multiple target prediction models to predict multiple wind turbine groups to obtain multiple initial wind power prediction results.

[0124] Specifically, by obtaining multiple data of each wind turbine group in real time and inputting them into each target prediction model, the prediction of multiple wind turbine groups can be realized, and the corresponding multiple initial wind power prediction results can be obtained.

[0125] Step S3063: Determine the target prediction result of wind power based on the wind speed condition and multiple initial wind power prediction results.

[0126] Specifically, considering the current wind speed condition, the multiple obtained initial wind power prediction results are fused to obtain the final target prediction result of wind power.

[0127] In an example, when the multiple target prediction models are the LightGBM model, GRU model, and Local-Ensemble model, the fusion process includes:

[0128] Weight assignment: Different weight coefficients are assigned according to the prediction effects of each model. Specifically, for each wind turbine, the root mean square error (RMSE) is calculated using the prediction results and the true value of the three models, and the reciprocal of the RMSE is used as the weight coefficient. For example, if the RMSE of the LightGBM model is 0.1, the RMSE of the GRU model is 0.2, and the RMSE of the Local-Ensemble module is 0.3, the corresponding weight coefficients are 10, 5, and 3.33.

[0129] Weight fusion: According to the weight coefficients of each model, the prediction results of the three models are weighted and averaged to obtain the final prediction result. For example, if the prediction result of the LightGBM model is 1.5, the prediction result of the GRU module is 1.8, and the prediction result of the Local-Ensemble model is 2.1, then the final prediction result is (10×1.5 + 5×1.8 + 3.33×2.1) / (10 + 5 + 3.33) = 1.69.

[0130] Weight fine-tuning: According to the experimental results, the weight coefficients are fine-tuned to achieve better prediction performance. Specifically, for each wind turbine, the grid search method is used to find the optimal combination of weight coefficients within a certain range, and the original weight coefficients are replaced with this combination. For example, if the original weight coefficients are 10, 5, 3.33, then the optimal combination of weight coefficients is searched within [9, 11]×[4, 6]×[2.33, 4.33], and the original weight coefficients are replaced with this combination.

[0131] Step S307, obtain the measured power data sets of multiple wind turbine groups.

[0132] Specifically, the power data of multiple wind turbine groups can be obtained by real-time measurement at a preset time interval.

[0133] Step S308, update multiple target prediction models based on the wind power target prediction results and the measured power data sets.

[0134] Specifically, calculate the root mean square error (RMSE) and mean absolute percentage error (MAPE) between the predicted value and the actual value, as well as the error rate consistency index (ERCI) between different wind turbine groups. Set the acceptable range of the error rate. For example, RMSE is less than 0.1, MAPE is less than 5%, and ERCI is greater than 0.8. When any index exceeds this range, update multiple target prediction models to adapt to the current data distribution.

[0135] The multi-model coupled wind power prediction method provided in this embodiment analyzes the operating behaviors of multiple wind turbine groups by combining the initial historical data sets, geographical location information sets, and spatial topological relationships corresponding to the multiple wind turbine groups, taking into account the spatial relationships and semantic similarities between the wind turbines, and improving the non-uniformity of the spatial characteristics of the analysis results of the operating behaviors of the wind turbine groups. Further, the wind speed control model is used to select multiple target prediction models from the multiple constructed models to predict the wind power of the multiple wind turbine groups, improving the robustness and stability of the prediction results. At the same time, the wind speed control model is combined during the prediction process, considering the differences in prediction effects under different wind speed conditions, and being able to effectively reduce the deviations that occur in the prediction results at high wind speeds or low wind speeds. Further, by combining the wind power target prediction results and the measured power data sets, the prediction results of the multi-model coupled wind power prediction model can be analyzed, managed, and updated in real time, being able to effectively reflect the changes in the current data distribution, and then updating the multiple target prediction models to improve the real-time performance of the prediction results.

[0136] In this embodiment, a multi-model coupled wind power prediction device is also provided. This device is used to implement the above-mentioned embodiment and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0137] This embodiment provides a multi-model coupled wind power prediction device, as Figure 4 shown, this device includes:

[0138] The first acquisition module 401 is used to acquire the initial historical data sets, geographical location information sets, and spatial topological relationships corresponding to multiple wind turbine groups.

[0139] The analysis module 402 is used to analyze the operating behaviors of multiple wind turbine groups based on the initial historical data sets, geographical location information sets, and spatial topological relationships to obtain the analysis results of the operating behaviors of the wind turbine groups.

[0140] The first establishment module 403 is used to respectively establish a spatio-temporal attention network model, a LightGBM model, a GRU model, and a Local-Ensemble model based on the initial historical data sets and the analysis results of the operating behaviors of the wind turbine groups.

[0141] The second establishment module 404 is used to establish a wind speed control model based on the initial historical data sets.

[0142] A determination module 405, configured to determine multiple target prediction models from a spatio-temporal attention network model, a LightGBM model, a GRU model, and a Local-Ensemble model based on a wind speed control model.

[0143] A prediction module 406, configured to predict the wind power of multiple wind turbine groups based on the wind speed control model and the multiple target prediction models to obtain a target prediction result of the wind power.

[0144] In some alternative embodiments, the analysis module 402 includes:

[0145] A division sub-module, configured to perform similarity division on multiple wind turbine groups based on an initial historical data set, a geographical location information set, and a spatial topological relationship to obtain a similarity wind turbine division result.

[0146] A processing sub-module, configured to process the initial historical data set based on the similarity wind turbine division result to obtain a target historical data set.

[0147] An analysis sub-module, configured to perform outlier identification and operating behavior analysis on the wind power of multiple wind turbine groups based on the target historical data set to obtain an operating behavior analysis result of the wind turbine groups.

[0148] In some alternative embodiments, the division sub-module includes:

[0149] A clustering unit, configured to process and cluster the historical data set based on different time dimensions to obtain a clustering result.

[0150] A processing unit, configured to obtain a graph construction result through graph construction method processing based on the geographical location information set and the spatial topological relationship.

[0151] A division and analysis unit, configured to perform similarity division and analysis on multiple wind turbine groups based on the clustering result and the graph construction result to obtain a similarity wind turbine division result.

[0152] In some alternative embodiments, the first establishment module 403 includes:

[0153] A construction sub-module, configured to construct three-dimensional tensor data based on the initial historical data set.

[0154] A first establishment sub-module, configured to establish a spatio-temporal attention network model based on the three-dimensional tensor data and the operating behavior analysis result of the wind turbine groups.

[0155] A second establishment sub-module, configured to establish a LightGBM model, a GRU model, and a Local-Ensemble model respectively based on the initial historical data set and the operating behavior analysis result of the wind turbine groups.

[0156] In some alternative embodiments, the prediction module 406 includes:

[0157] A first determination sub-module, configured to determine a wind speed condition based on a wind speed control model.

[0158] A prediction sub-module, configured to perform predictions on multiple wind turbine sets by using a plurality of target prediction models to obtain a plurality of initial wind power prediction results.

[0159] A second determination sub-module, configured to determine a target wind power prediction result based on the wind speed condition and the plurality of initial wind power prediction results.

[0160] In some alternative embodiments, the device further includes:

[0161] A second acquisition module, configured to acquire a measured power data set of multiple wind turbine sets.

[0162] An update module, configured to update the plurality of target prediction models based on the target wind power prediction result and the measured power data set.

[0163] The further function descriptions of the above-mentioned respective modules and units are the same as those in the corresponding foregoing embodiments, and will not be elaborated herein.

[0164] The multi-model coupled wind power prediction device in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0165] An embodiment of the present invention further provides a computer device having the above-mentioned Figure 4 multi-model coupled wind power prediction device shown.

[0166] Please refer to Figure 5 , Figure 5 which is a schematic structural diagram of a computer device provided by an alternative embodiment of the present invention. As shown in Figure 5As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting the components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (such as an array of servers, a set of blade servers, or a multi-processor system). Figure 5 In the figure, one processor 10 is taken as an example.

[0167] The processor 10 can be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 can further include a hardware chip. The above-mentioned hardware chip can be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above-mentioned programmable logic device can be a complex programmable logic device, a field programmable gate array, a generic array logic, or any combination thereof.

[0168] Among them, the memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiments.

[0169] The memory 20 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the computer device, etc. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 can optionally include a memory remotely set relative to the processor 10, and these remote memories can be connected to the computer device 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 combinations thereof.

[0170] The memory 20 can include a volatile memory, such as a random access memory; the memory can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 can also include a combination of the above types of memories.

[0171] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or communication networks.

[0172] Embodiments of the present invention also provide a computer-readable storage medium. The method according to the embodiments of the present invention can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored as such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0173] A part of the present invention can be applied as a computer program product, for example, computer program instructions, which when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the forms of existence of computer program instructions in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Herein, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to the computer.

[0174] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A multi-model coupling wind power prediction method, characterized in that: The method comprises: Acquire an initial historical data set, a geographical location information set, and a spatial topological relationship corresponding to a plurality of wind turbine groups, wherein the initial historical data set includes meteorological data, status data of each wind turbine group, and data of a wind farm wind tower; Based on the initial historical data set, the geographic location information set and the spatial topological relationship, analyzing the operation behaviors of the multiple wind turbine groups to obtain analysis results of the wind turbine group operation behaviors; Based on the initial historical data set and the analysis results of the wind turbine group operation behavior, a spatiotemporal attention network model, a LightGBM model, a GRU model and a Local-Ensemble model are established respectively; Based on the initial historical data set, establishing a wind speed control model; Based on the wind speed control model, determining multiple target prediction models in the spatiotemporal attention network model, the LightGBM model, the GRU model, and the Local-Ensemble model; Predicting the wind power of the multiple wind turbine groups based on the wind speed control model and the multiple target prediction models to obtain a wind power target prediction result; Wherein, based on the initial historical data set, the geographical location information set and the spatial topological relationship, the operation behaviors of the plurality of wind turbine groups are analyzed to obtain the analysis results of the operation behaviors of the wind turbine groups, including: Based on the initial historical data set, the geographical location information set and the spatial topological relationship, the plurality of wind turbine groups are similarly divided to obtain similar wind turbine division results; Based on the similar wind turbine division result, the initial historical data set is processed to obtain a target historical data set; Based on the target historical data set, abnormal point identification and operation behavior analysis are performed on the wind power of the multiple wind turbine groups to obtain the operation behavior analysis results of the wind turbine groups; Wherein, based on the initial historical data set, the geographical location information set and the spatial topological relationship, the plurality of wind turbine groups are similarly divided to obtain similar wind turbine division results, including: Based on different time dimensions, the historical data set is processed and clustered to obtain a clustering result; Based on the geographic location information set and the spatial topological relationship, a graph construction result is obtained through processing by a graph construction method, wherein the graph construction result includes a geographic distance graph and a semantic distance graph, the geographic distance graph is used to characterize the spatial relationship between wind turbines, and the semantic distance graph is used to characterize the semantic similarity between wind turbines; Based on the clustering result and the graph construction result, the plurality of wind turbine groups are similarly divided and analyzed to obtain the similar wind turbine division result; Wherein, based on the wind speed control model, multiple target prediction models are determined in the spatiotemporal attention network model, the LightGBM model, the GRU model and the Local-Ensemble model, including: identifying and determining wind speed conditions according to the wind speed control model, and determining multiple target prediction models in the spatiotemporal attention network model, the LightGBM model, the GRU model and the Local-Ensemble model according to the wind speed conditions.

2. The method according to claim 1, characterized in that Based on the initial historical data set and the analysis results of the wind turbine operation behavior, a spatiotemporal attention network model, a LightGBM model, a GRU model and a Local-Ensemble model are established respectively, including: constructing three-dimensional tensor data based on the initial historical data set; Based on the three-dimensional tensor data and the wind turbine group operation behavior analysis results, establishing the spatiotemporal attention network model; Based on the initial historical data set and the analysis results of the wind turbine group operation behavior, the LightGBM model, the GRU model and the Local-Ensemble model are established respectively.

3. The method according to claim 1, characterized in that Predicting the wind power of the multiple wind turbine groups based on the wind speed control model and the multiple target prediction models to obtain wind power target prediction results includes: Based on the wind speed control model, determining a wind speed condition; Using the multiple target prediction models to predict the multiple wind turbine groups, obtaining multiple initial wind power prediction results; The wind power target prediction result is determined based on the wind speed condition and the plurality of initial wind power prediction results.

4. The method according to claim 1, characterized in that: The method further comprises: Obtaining a measured power data set of the plurality of wind turbine groups; Based on the wind power target prediction result and the measured power data set, the multiple target prediction models are updated.

5. A multi-model coupled wind power prediction device, characterized in that: Used to execute the multi-model coupling wind power prediction method according to any one of claims 1 to 4; the device comprises: A first acquisition module is used to acquire an initial historical data set, a geographical location information set and a spatial topological relationship corresponding to a plurality of wind turbine groups; An analysis module, configured to analyze the operation behaviors of the plurality of wind turbine groups based on the initial historical data set, the geographic location information set and the spatial topological relationship, and obtain an analysis result of the operation behaviors of the wind turbine groups; A first establishment module is used to establish a spatiotemporal attention network model, a LightGBM model, a GRU model and a Local-Ensemble model respectively based on the initial historical data set and the wind turbine group operation behavior analysis results; A second establishing module is used to establish a wind speed control model based on the initial historical data set; A determination module, used to determine multiple target prediction models in the spatiotemporal attention network model, the LightGBM model, the GRU model and the Local-Ensemble model based on the wind speed control model; A prediction module is used to predict the wind power of the multiple wind turbine groups based on the wind speed control model and the multiple target prediction models to obtain a wind power target prediction result.

6. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the multi-model coupling wind power prediction method according to any one of claims 1 to 4 by executing the computer instructions.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the multi-model coupling wind power prediction method according to any one of claims 1 to 4.

8. A computer program product, characterized in that The method comprises computer instructions, wherein the computer instructions are used to enable a computer to execute the multi-model coupling wind power prediction method according to any one of claims 1 to 4.

Images