Wind power prediction method, system and equipment for wind power plant in combination with wind turbine spatial-temporal characteristics and adjacency matrix optimization, and medium

Abstract

The invention discloses a wind power plant wind power prediction method, system, equipment and medium in combination with fan spatial-temporal characteristics and adjacent matrix optimization, and relates to the field of wind power plant wind power prediction.The wind power plant wind power prediction method comprises the steps that wind power plant historical wind power data and real-time wind direction and wind speed data are collected, and the collected wind power plant historical wind power data are preprocessed; dynamically generating an adjacent matrix between the fans based on the spatial position information of the fans and the real-time wind direction and wind speed data; constructing a prediction network for extracting spatial and temporal characteristics of the wind power data, wherein the prediction network is used for establishing a time dependency relationship and a spatial coupling relationship of fan power changes; and predicting the wind power in a future time period by using the prediction network, and evaluating a prediction result to verify the performance of the model. The adjacent matrix is adjusted through the real-time wind speed and wind direction data, the spatial correlation and wake effect between the fans are accurately described, the temporal and spatial features are extracted in combination with the time sequence convolution layer and the graph convolution layer, and the prediction precision is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind farm wind power prediction, in particular to a wind farm wind power prediction method, system, device and medium combining wind turbine space-time characteristics and adjacency matrix optimization. BACKGROUND

[0002] Wind power prediction is a key technology in wind farm operation and energy scheduling, and its accuracy directly affects the stability of the power grid and the economic benefits of wind power. Currently, wind power prediction methods are mainly divided into physical methods, statistical methods and machine learning methods. Physical methods are based on numerical weather prediction and fluid dynamics models, which can reflect the physical characteristics of the wind farm, but have poor adaptability to complex terrain and weather conditions; statistical methods (such as ARIMA, Kalman filter) rely on statistical analysis of historical data, which is difficult to capture the nonlinear relationship between wind speed and wind power; machine learning methods (such as support vector machine, random forest) improve the prediction accuracy through data-driven methods, but still have difficulty in effectively modeling the spatial dependence relationship between wind turbines.

[0003] With the scale and complexity of wind farms, the wake effect and spatial correlation between wind turbines have an increasingly significant impact on wind power prediction. In recent years, graph convolutional networks (GCN) have been introduced into the field of wind power prediction, which can extract spatial features using the topological structure between wind turbines. However, existing methods mostly use static adjacency matrices, which cannot adapt to the dynamic changes of wind direction and wind speed, resulting in limited prediction accuracy. In addition, existing methods have limited ability to extract spatio-temporal features, making it difficult to model both temporal and spatial dependence relationships. Therefore, the present application proposes a wind power prediction method that can dynamically adjust the adjacency matrix and combine spatio-temporal features to improve prediction accuracy and practicality. SUMMARY

[0004] In view of the above problems, the present application is proposed.

[0005] Therefore, the problem to be solved by the present application is that existing methods mostly use static adjacency matrices, which cannot adapt to the dynamic changes of wind direction and wind speed, resulting in limited prediction accuracy; in addition, existing methods have limited ability to extract spatio-temporal features, making it difficult to model both temporal and spatial dependence relationships.

[0006] To solve the above technical problems, the application provides the following technical scheme: a wind power prediction method combining wind turbine space-time characteristics and adjacency matrix optimization, which comprises the following steps: collecting historical wind power data and real-time wind direction and speed data of a wind farm, preprocessing the collected historical wind power data of the wind farm to ensure continuity and consistency of the space-time characteristics of the data; dynamically generating an adjacency matrix between wind turbines based on spatial position information of the wind turbines and real-time wind direction and speed data, which is used to represent the spatial correlation between the wind turbines; constructing a prediction network for extracting space-time characteristics of wind power data, which is used to establish a time-dependent relationship and a spatial coupling relationship of wind turbine power variation; and predicting future wind power by using the prediction network and evaluating the prediction result to verify the performance of the model.

[0007] As a preferred scheme of the wind power prediction method combining wind turbine space-time characteristics and adjacency matrix optimization, the preprocessing of the collected historical wind power data of the wind farm comprises loading historical wind power data of each wind turbine and integrating the wind power data; converting the integrated data into a time series format to generate historical input and future prediction target; and dividing the processed data into a training set, a validation set and a test set.

[0008] As a preferred scheme of the wind power prediction method combining wind turbine space-time characteristics and adjacency matrix optimization, the dynamic generation of the adjacency matrix between the wind turbines comprises constructing a wind turbine relationship expression structure that is updated with real-time wind direction and environmental variables; and establishing an adjacency matrix reflecting the dependency relationship between the wind turbines according to the expression structure.

[0009] The preferred technical scheme has the beneficial effect that by introducing dynamic variables such as wind direction and speed, a real-time updated wind turbine relationship expression structure is constructed, which can effectively reflect the aerodynamic coupling effect and spatial dependency between the wind turbines.

[0010] As a preferred scheme of the wind power prediction method combining wind turbine space-time characteristics and adjacency matrix optimization, the construction of the prediction network comprises constructing a modeling network structure containing time dimension and space dimension; and training the network structure by using the training set, the validation set and the test set to extract the time evolution law and spatial dependency relationship of the wind power data.

[0011] The preferred technical scheme has the beneficial effect that by constructing a prediction network for joint time-space modeling, the time series trend and spatial correlation of the wind power are expressed in an integrated manner, thereby improving the recognition ability and precision of the prediction model for complex dynamic characteristics.

[0012] As a preferred scheme of the wind power prediction method of the wind farm combining the spatiotemporal characteristics of wind turbines and the adjacency matrix optimization, in the construction of the modeling network structure containing the time dimension and the space dimension, the information propagation of adjacent wind turbine nodes is performed by using the dynamically generated adjacency matrix; the training of the network structure by using the training set, the verification set and the test set comprises loading the training set, the verification set and the test set into the network structure for training, and in the training process, the mean square error is used as the loss function, and the adaptive moment estimation optimizer is used for parameter updating.

[0013] As a preferred scheme of the wind power prediction method of the wind farm combining the spatiotemporal characteristics of wind turbines and the adjacency matrix optimization, the prediction of the future time period wind power by using the prediction network comprises using the trained prediction network, combining the real-time wind speed and wind direction data, separately predicting and outputting the wind power of each wind turbine in the target time period, and generating the power output value of each wind turbine in the prediction window.

[0014] As a preferred scheme of the wind power prediction method of the wind farm combining the spatiotemporal characteristics of wind turbines and the adjacency matrix optimization, the evaluation of the prediction result comprises comparing the model prediction result with the actual observation value, generating a prediction error result for evaluating the model performance, and based on the prediction error result, iteratively optimizing the model parameters and adjusting the model prediction output.

[0015] To solve the above technical problems, the present application provides the following technical scheme: a wind power prediction system of a wind farm combining the spatiotemporal characteristics of wind turbines and the adjacency matrix optimization, comprising a data acquisition and processing module, an adjacency matrix construction module, a prediction network training module and a prediction and correction module; the data acquisition and processing module is used for acquiring historical wind power data and real-time wind direction and wind speed data of the wind farm, preprocessing the acquired historical wind power data of the wind farm, and making the spatiotemporal characteristics of the data continuous and consistent; the adjacency matrix construction module dynamically generates the adjacency matrix between wind turbines based on the spatial position information of the wind turbines and the real-time wind direction and wind speed data, and is used for representing the spatial correlation between the wind turbines; the prediction network training module constructs a prediction network for extracting the spatiotemporal characteristics of the wind power data, and the prediction network is used for establishing the time-dependent relationship and the spatial coupling relationship of the wind turbine power change; the prediction and correction module is used for predicting the wind power of a future time period by using the prediction network, and evaluating the prediction result to verify the model performance.

[0016] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the wind power prediction method of the wind farm combining the spatiotemporal characteristics of wind turbines and the adjacency matrix optimization when executing the computer program.

[0017] A computer readable storage medium, having stored thereon a computer program, the computer program being executed by a processor to implement the steps of the wind farm wind power prediction method combining the spatiotemporal characteristics of the wind turbine and the adjacency matrix optimization as described above.

[0018] The application has the advantages that: the adjacency matrix is adjusted through real-time wind speed and wind direction data, the spatial correlation and wake effect between wind turbines are accurately described, the spatiotemporal characteristics are extracted by combining the time series convolution layer and the graph convolution layer, the prediction accuracy is improved, the method can adapt to wind direction changes, reduce prediction errors, improve the determination coefficient, and has good generalization ability and calculation efficiency, and provides reliable technical support for real-time wind power prediction and operation scheduling of the wind farm. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creating laborious work.

[0020] Figure 1 The flowchart of the wind farm wind power prediction method combining the spatiotemporal characteristics of the wind turbine and the adjacency matrix optimization in Example 1.

[0021] Figure 2 The spatiotemporal graph convolution model architecture diagram of the wind farm wind power prediction method combining the spatiotemporal characteristics of the wind turbine and the adjacency matrix optimization in Example 2. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings of the specification.

[0023] In the following description, many specific details are set forth in order to provide a thorough understanding of the application, but the application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the concept of the application, therefore the application is not limited to the specific embodiments disclosed below.

[0024] Example 1, refer to Figure 1 , the first embodiment of the application, the embodiment provides a wind farm wind power prediction method combining the spatiotemporal characteristics of the wind turbine and the adjacency matrix optimization, which includes, as shown in Figure 1

[0025] S1: collect historical wind power data and real-time wind direction and wind speed data of the wind farm, and pretreat the collected historical wind power data of the wind farm, so that the spatiotemporal characteristics of the data exist continuity and consistency. ​

[0026] S2: Based on the wind turbine spatial position information and real-time wind direction and wind speed data, a neighbor matrix between wind turbines is dynamically generated to represent the spatial correlation between wind turbines.

[0027] S3: A prediction network for extracting the spatio-temporal features of wind power data is constructed, and the prediction network is used to establish the time-dependent relationship and spatial coupling relationship of wind turbine power changes.

[0028] S4: The prediction network is used to predict the future period wind power, and the prediction result is evaluated to verify the model performance.

[0029] It should be noted that with the large-scale and complex of wind farms, the influence of wake effect and spatial correlation between wind turbines on wind power prediction is increasingly significant. The existing wind power prediction methods mostly use static neighbor matrix, which cannot adapt to the dynamic changes of wind direction and wind speed, resulting in limited prediction accuracy. In addition, the existing methods have limited ability to extract spatio-temporal features, making it difficult to model both time-dependent relationship and spatial-dependent relationship.

[0030] Therefore, in order to solve the above problems, through S1-S4 steps, first, the historical wind power data of the wind farm is preprocessed to ensure good continuity and consistency of the data; second, the spatial position information of the wind turbine and real-time wind direction and wind speed and other meteorological factors are combined to dynamically construct the neighbor matrix between wind turbines to accurately reflect the spatial dependence relationship between wind turbines; third, a prediction network model that can extract both time-dependent and spatial coupling features is constructed; finally, the trained network model is used to predict the wind power, and the performance of the prediction result is evaluated to verify the accuracy and stability of the model; solve the problem that the existing method cannot adapt to the dynamic changes of wind direction and wind speed.

[0031] Embodiment 2, refer to Figure 2 The second embodiment of the present application is different from the first embodiment in that the wind farm wind power prediction method combining wind turbine spatio-temporal features and neighbor matrix optimization further comprises,

[0032] In step S1, the historical wind power data and real-time wind direction and wind speed data of the wind farm are collected, and the collected historical wind power data of the wind farm is preprocessed, including the following steps A1-A3:

[0033] A1: Load the historical wind power data of each wind turbine, and integrate the wind power data.

[0034] A2: Convert the integrated data into time series format to generate historical input and future prediction target for model training.

[0035] A3: Divide the processed data into training set, validation set and test set to provide data support for model training and evaluation.

[0036] In the embodiments of the present application, in step A1, the way of integrating wind power data is to load wind power data files of each wind turbine from a specified path, perform data cleaning and normalization processing, including the following steps A111-A114:

[0037] A111: Load the adjacency matrix from the adjacency matrix file, return a sparse matrix, and represent the static connection relationship between wind turbines.

[0038] A112: Load single wind power data files from a specified path, perform data cleaning and normalization processing.

[0039] A113: For data points with missing values, fill them with the previous valid value to maintain the continuity of the time series.

[0040] A114: For abnormal data (such as negative wind power or values outside the reasonable range), set it to 0 to avoid data errors.

[0041] In an optional embodiment, the way of integrating wind power data can also be to directly extract historical wind power data of each wind turbine from a database interface, and aggregate and integrate the original data according to the wind turbine number and sampling time, including the following steps A121-A123:

[0042] A121: For missing values, use the sliding window average method to fill in (e.g. fill in the missing point with the average of the previous 3 non-empty sampling points).

[0043] A121: For abnormal values, remove data points outside the statistical reasonable interval by the interquartile range (IQR) method, and replace them with local average values.

[0044] A123: Finally output the normalized time series data for model training.

[0045] In another optional embodiment, the way of integrating wind power data can also be to read the wind power log data automatically synchronized uploaded by the distributed edge device, and perform time alignment and completion on the asynchronous collection results through the data lake platform, including the following steps A131-A133:

[0046] A131: For missing points, use the nearest neighbor interpolation method (KNN Imputation) to complete.

[0047] A131: For detected abnormal points, use a machine learning anomaly detection model to identify and correct them.

[0048] A131: Finally, reconstruct all wind turbine data into time series data sets with fixed sampling intervals.

[0049] It should be noted that by cleaning, normalizing and structuring the wind power data, the integrity and quality of the original data are improved, which helps to eliminate the interference of missing values and outliers on model training, so that the subsequent model can more fully learn the time evolution and spatial distribution characteristics of wind turbine power on the basis of high-quality input, thereby improving the modeling stability and prediction reliability.

[0050] Further, in step S2, based on the spatial position information of the wind turbines and the real-time wind direction and speed data, a neighbor matrix between the wind turbines is dynamically generated, including the following steps B1-B2:

[0051] B1: Construct a wind turbine relationship expression structure that is updated with real-time wind direction and environmental variable changes.

[0052] B2: Establish a neighbor matrix reflecting the dependency relationship between wind turbines according to the expression structure.

[0053] In the embodiments of the present application, in step B2, for the neighbor matrix, the neighbor matrix is constructed based on the wake loss to reflect the dependency relationship between wind turbines, including the following steps B211-B213:

[0054] B211: According to the real-time wind direction data, determine the wind direction angle, and calculate the relative wind direction angle between wind turbine i and wind turbine j, denoted as:

[0055]

[0056] wherein, is the relative wind direction angle; (x i ,y i ) and (x j ,y j ) are the coordinates of wind turbine i and wind turbine j, respectively; θ is the wind direction angle.

[0057] B212: Calculate the wake loss between the wind turbines, which is related to the distance and the relative wind direction angle between the wind turbines, and the calculation formula of the wake loss is denoted as:

[0058]

[0059] wherein, V ij is the wake loss; C is the wake loss coefficient, used to adjust the strength of the wake loss; σ is the distance decay coefficient, which controls the speed of the wake loss decay with distance; D ij is the distance between wind turbine i and wind turbine j.

[0060] B213: Calculate the influence weight between the wind turbines based on the wake loss, generate the wind turbine neighbor matrix, and reflect the wake effect and spatial correlation between the wind turbines.

[0061] W ij = 1 - V ij

[0062] wherein, W ij is the adjacency matrix of the wind turbines.

[0063] In an alternative embodiment, the adjacency matrix can also be constructed based on a dynamic adjacency matrix construction method based on wind speed correlation, comprising the following steps B221-B225:

[0064] B221: The system periodically collects real-time wind speed time series data of each wind turbine, and calculates the Pearson correlation coefficient matrix between the wind turbines, which is used as a measure of the mutual influence degree between the wind turbines.

[0065] B222: For each time, the wind speed correlation in a sliding time window in the past is calculated in real time.

[0066] B223: The wind turbine pair with a correlation coefficient higher than a set threshold is set as a connected edge.

[0067] B224: The absolute value of the correlation coefficient is used as the weight in the adjacency matrix, reflecting the degree of cooperative change.

[0068] B225: The adjacency matrix is updated at each time step, dynamically reflecting the wind speed propagation or coupling relationship between the wind turbines.

[0069] In another alternative embodiment, the adjacency matrix can also be constructed based on a joint weight factor based on terrain-wind direction, comprising the following steps B231-B234:

[0070] B231: Collect terrain data and real-time wind direction information of the area where the wind farm is located, and evaluate the wind direction downwind channel property according to the included angle between the relative azimuth angle between the wind turbines and the real-time dominant wind direction, to obtain the wind direction compliance factor.

[0071] B232: Calculate the terrain shielding conditions such as altitude mutation or wind channel contraction on the path between the wind turbines, to obtain the terrain shielding factor.

[0072] B233: Weight and fuse the wind direction compliance factor and the terrain shielding factor to form an influence score matrix between the wind turbines.

[0073] B234: The score matrix is used to construct the adjacency matrix, and the edge weight represents the spatial influence strength, which is updated in real time with the change of wind direction; wherein the edge weight represents the spatial influence degree or action strength of one wind turbine on another wind turbine under specific wind direction and terrain conditions.

[0074] It should be noted that the scheme breaks through the limitations of the traditional static adjacency matrix, and by introducing dynamic variables such as wind direction and wind speed, a real-time updated fan relationship expression structure is constructed, which can effectively reflect the aerodynamic coupling effect and spatial dependence between fans, enhance the adaptability of the model to complex wind field changes, and improve the spatial modeling precision and generalization effect.

[0075] Further, in step S3, a prediction network for extracting the spatiotemporal characteristics of wind power data is constructed, including the following steps C1-C2:

[0076] C1: Construct a modeling network structure containing time dimension and space dimension.

[0077] C2: Train the network structure using the preprocessed data to extract the temporal evolution law and spatial dependence relationship of the wind power data.

[0078] Specifically, when constructing the modeling network structure containing time dimension and space dimension, the dynamic generated adjacency matrix is used for information propagation between adjacent fan nodes to reduce the computational complexity.

[0079] In the embodiments of the present application, in step C1, for the modeling network structure, the modeling network structure is constructed by using a graph convolutional network (GCN, Graph Convolutional Network) for modeling, including the following steps C111-C114:

[0080] C111: Construct a time series convolution layer, use one-dimensional causal convolution to extract features of wind power data, ensure that the model only uses past information for prediction, and avoid future information leakage.

[0081] C112: Construct a graph convolution layer, use a dynamic adjacency matrix to propagate information between adjacent fan nodes to reduce computational complexity.

[0082] C113: Combine the time series convolution layer and the graph convolution layer to form a spatiotemporal convolution module, and then integrate multiple spatiotemporal convolution modules to capture the spatiotemporal dependence relationship of fan data. Finally, use a fully connected layer to map the features to the output dimension, output the single-machine wind power prediction value in the future period of time, and the specific spatiotemporal graph convolution model architecture is as shown in Figure 2 .

[0083] C114: Load the divided training set, validation set and test set into the spatiotemporal graph convolution model for training. In the training process, the mean square error is used as the loss function, the Adam (Adaptive Moment Estimation, Adaptive Moment Estimation) optimizer is used for parameter update, and the wake correlation parameters can be optimized synchronously.

[0084] In an optional embodiment, the modeling network structure can also be constructed based on a spatio-temporal attention mechanism network (STAM) to construct a prediction model, including the following steps C121-C125:

[0085] C121: Introduce a self-attention mechanism in the time dimension to capture key fluctuations in the historical time series.

[0086] C122: Introduce an attention score matrix in the spatial dimension to the relationship between each fan node, and adjust the information flow weight between the fans according to the real-time feature data.

[0087] C123: The entire model does not rely on static or graph structure, and automatically constructs fan dependency relationship expression through learnable attention weight.

[0088] C124: The output end combines residual connection and normalization mechanism to output the predicted value, improving the stability and generalization ability of the model.

[0089] C125: Load the divided training set, validation set and test set into the prediction model for training, use mean square error or mean absolute error as loss function during training, use Adam optimizer for parameter iteration, and introduce attention sparse regularization or entropy regularization to suppress overfitting. Monitor the validation error during the training process and apply the early stopping mechanism.

[0090] In another optional embodiment, the modeling network structure can also be constructed based on a convolutional gated recurrent unit network (ConvGRU) to construct a prediction model, including the following steps C131-C135:

[0091] C131: In the time dimension, use the gated recurrent unit to model the historical changes of wind power.

[0092] C132: In the spatial dimension, divide the wind farm area into grids, and consider the wind turbines in each grid as a spatial unit. Use convolution operation to capture local spatial structure.

[0093] C133: The ConvGRU structure embeds convolution operation into the gated recurrent unit, so that it can process the current spatial image at each time step while preserving the temporal context information.

[0094] C134: The model gets the prediction result through end-to-end training without relying on external graphing or explicit adjacency matrix.

[0095] C135: In the training process, the mean square error is used as the objective function, the model parameters are optimized using the Adam optimizer, and the overfitting is suppressed through Batch Normalization and Dropout, and the stability and effect of long sequence training are guaranteed by combining gradient clipping and early stopping strategy.

[0096] It should be noted that by constructing the prediction network of spatio-temporal joint modeling, the integration expression of the time series trend of wind power and spatial correlation is realized, which can effectively capture the deep patterns of wind turbine power evolution and wind farm structure changes, thereby improving the recognition ability and accuracy of the prediction model for complex dynamic characteristics.

[0097] Further, in step S4, predicting the wind power of the future time period using the prediction network includes using the trained prediction network to combine real-time wind speed and wind direction data to individually predict the wind power of each wind turbine in the target time period and output the power output value of each wind turbine in the prediction window.

[0098] Further, in step S4, the prediction results are evaluated to verify the model performance, including steps D1-D2:

[0099] D1: Compare the model prediction results with the actual observation values to generate prediction error results for evaluating the model performance.

[0100] D2: Based on the prediction error results, the model parameters are optimized and iterated to adjust the model prediction output.

[0101] In the embodiments of the present application, in step D1, the method for evaluating the model performance is to introduce an abnormality detection trigger evaluation mechanism for model performance monitoring, including steps D111-D114:

[0102] D111: In the prediction phase, the system detects whether the deviation between the predicted value and the actual value exceeds a certain tolerance threshold at intervals.

[0103] D112: If the tolerance is exceeded for a plurality of times, the evaluation is started, and a model performance diagnosis is automatically performed.

[0104] D113: The diagnosis process is based on the maximum absolute error (MaxAE) and the rate-of-change error (RCE) in the sliding time window for focused analysis.

[0105] D114: If the analysis shows that the model prediction performance degradation exceeds the minimum allowed threshold, a model retraining or parameter fine-tuning instruction is sent.

[0106] In an optional embodiment, the method for evaluating the performance of the model can also be evaluated by using the overall load trend consistency comparison method of the wind farm, including the following steps D121-D123:

[0107] D121: First, compare the prediction results with the actual wind power curve in multiple time windows (such as 5 minutes, 30 minutes, 1 hour);

[0108] D121: Calculate the trend matching degree (such as the sign change direction, the inflection point coincidence rate) of the prediction curve and the actual curve in each time window.

[0109] D121: Build a trend matching accuracy index as a quantitative evaluation of the model's response ability to future trend changes.

[0110] In another optional embodiment, the method for evaluating the performance of the model can also calculate the root mean square error (RMSE), the mean absolute error (MAE), and the determination coefficient (R 2 ) and other indicators to measure the prediction accuracy and generalization ability of the model.

[0111] Embodiment 3 is the third embodiment of the present application, which is different from the first two embodiments: a wind farm wind power prediction system combining the spatio-temporal characteristics of wind turbines and the optimization of adjacency matrix, including a data acquisition and processing module, an adjacency matrix construction module, a prediction network training module, and a prediction and correction module; the data acquisition and processing module is used to acquire historical wind power data and real-time wind direction and speed data of the wind farm, and preprocess the acquired historical wind power data of the wind farm to ensure the continuity and consistency of the spatio-temporal characteristics of the data; the adjacency matrix construction module dynamically generates the adjacency matrix between wind turbines based on the spatial position information of wind turbines and real-time wind direction and speed data, which is used to represent the spatial correlation between wind turbines; the prediction network training module constructs a prediction network for extracting the spatio-temporal characteristics of wind power data, which is used to establish the time-dependent relationship and spatial coupling relationship of wind turbine power changes; the prediction and correction module is used to predict the wind power in the future time period by using the prediction network, and evaluate the prediction results to verify the performance of the model.

[0112] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the technical solutions that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0113] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a list of executable instructions for implementing logic functions, which can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor or other system that can fetch the instructions from the instruction execution system, apparatus or device and execute the instructions, or in conjunction with these instructions execution systems, apparatus or devices. For the purpose of this specification, the "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by or in connection with an instruction execution system, apparatus or device, or in conjunction with these instruction execution systems, apparatus or devices.

[0114] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electrical devices), a portable computer diskette (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or other suitable medium on which the program can be printed, because the program can be electronically obtained, for example, by optical scanning of the paper or other medium, followed by editing, interpreting or otherwise processing, if necessary, in other suitable ways, to be electronically obtained and then stored in the computer memory.

[0115] It should be understood that portions of the present application can be implemented in hardware, software, firmware, or combinations thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, implementation can be with a combination of any of the following technologies, which are all well known in the art: discrete logic circuitry having logic gates for implementing logic functions upon an application of data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0116] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A wind farm wind power prediction method combining the spatiotemporal characteristics of wind turbines and adjacency matrix optimization, characterized in that: include, Historical wind power data and real-time wind direction and speed data of wind farms are collected. The collected historical wind power data of wind farms are preprocessed to ensure the continuity and consistency of the spatiotemporal characteristics of the data. Based on the spatial location information of wind turbines and real-time wind direction and speed data, an adjacency matrix between wind turbines is dynamically generated to characterize the spatial correlation between wind turbines. A prediction network is constructed to extract the spatiotemporal features of wind power data. The prediction network is used to establish the temporal dependence and spatial coupling relationship of wind turbine power changes. The prediction network is used to predict wind power in the future time period, and the prediction results are evaluated to verify the model performance.

2. The wind power prediction method for wind farms that combines the spatiotemporal characteristics of wind turbines with adjacency matrix optimization as described in claim 1, characterized in that: The preprocessing of the collected historical wind power data from wind farms includes... Load the historical wind power data of each wind turbine and integrate the wind power data; The integrated data is converted into a time series format to generate historical inputs and future prediction targets. The processed data is divided into training set, validation set and test set.

3. The wind power prediction method for wind farms that combines the spatiotemporal characteristics of wind turbines with adjacency matrix optimization as described in claim 2, characterized in that: The dynamically generated adjacency matrix between wind turbines includes... Construct a wind turbine relationship expression structure that is updated in real time with changes in wind direction and environmental variables; Based on the aforementioned expression structure, an adjacency matrix reflecting the dependencies between wind turbines is established.

4. The wind power prediction method for wind farms that combines the spatiotemporal characteristics of wind turbines with adjacency matrix optimization as described in claim 3, characterized in that: The prediction network is constructed in the following ways: Construct a modeling network structure that includes time and spatial dimensions; The network structure is trained using training, validation, and test sets to extract the temporal evolution and spatial dependence of wind power data.

5. The wind power prediction method for wind farms that combines the spatiotemporal characteristics of wind turbines with adjacency matrix optimization as described in claim 4, characterized in that: When constructing the modeling network structure that includes time and space dimensions, a dynamically generated adjacency matrix is ​​used to propagate information between adjacent wind turbine nodes. Training the network structure using the training set, validation set, and test set includes loading the training set, validation set, and test set into the network structure for training. During the training process, the mean squared error is used as the loss function, and an adaptive moment estimation optimizer is used to update the parameters.

6. The wind power prediction method for wind farms that combines the spatiotemporal characteristics of wind turbines with adjacency matrix optimization as described in claim 5, characterized in that: Predicting wind power for a future time period using the prediction network includes using the trained prediction network, combined with real-time wind speed and direction data, to individually predict and output the wind power of each wind turbine within the target time period, generating the power output value of each wind turbine within the prediction window.

7. The wind power prediction method for wind farms that combines the spatiotemporal characteristics of wind turbines with adjacency matrix optimization as described in claim 6, characterized in that: The evaluation of the prediction results includes comparing the model prediction results with the actual observations to generate prediction error results, which are used to evaluate the model performance. Based on the prediction error results, the model parameters are optimized and iterated to adjust the model prediction output.

8. A wind farm wind power prediction system combining wind turbine spatiotemporal characteristics and adjacency matrix optimization, employing the wind farm wind power prediction method combining wind turbine spatiotemporal characteristics and adjacency matrix optimization as described in any one of claims 1 to 7, characterized in that: It includes a data acquisition and processing module, an adjacency matrix construction module, a prediction network training module, and a prediction and correction module; The data acquisition and processing module is used to collect historical wind power data and real-time wind direction and speed data of wind farms, and to preprocess the collected historical wind power data of wind farms to ensure that the spatiotemporal characteristics of the data are continuous and consistent. The adjacency matrix construction module dynamically generates an adjacency matrix between wind turbines based on the wind turbine spatial location information and real-time wind direction and speed data, which is used to characterize the spatial correlation between wind turbines. The prediction network training module constructs a prediction network for extracting the spatiotemporal features of wind power data. The prediction network is used to establish the temporal dependence and spatial coupling relationship of wind turbine power changes. The prediction and correction module is used to predict wind power for future time periods using the prediction network and to evaluate the prediction results to verify the model performance.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the wind farm wind power prediction method according to any one of claims 1 to 7, which combines the spatiotemporal characteristics of wind turbines with adjacency matrix optimization.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the wind farm wind power prediction method according to any one of claims 1 to 7, which combines the spatiotemporal characteristics of wind turbines with adjacency matrix optimization.

Images