Wind power prediction method and device based on multi-parameter fusion and storage medium
By employing a multi-parameter fusion wind power prediction method, and utilizing a hybrid neural network model and attention mechanism, the bias problem caused by neglecting factors in traditional wind power prediction is solved, achieving high-precision and real-time wind farm scheduling support.
Patent Information
- Application Number
- CN202510730358.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-03
- Publication Date
- 2025-10-28
AI Technical Summary
Existing wind power prediction methods rely on single meteorological data and do not consider multiple factors such as terrain, wind turbine status and grid load, resulting in a large deviation between the prediction results and the actual power. In particular, they are not adaptable enough under extreme weather conditions and cannot meet the real-time dispatching needs of wind farms.
By acquiring multi-source data information from wind turbines, a standardized dataset is constructed, time-series periodic features and spatial correlation matrices are extracted, a hybrid neural network model is used for prediction, and feature weights are dynamically allocated through an attention mechanism to correct model parameters in real time, thereby achieving bidirectional data transmission.
It significantly improves the accuracy and adaptability of wind power prediction, provides reliable prediction results in complex scenarios, meets scheduling needs at different time scales, and supports real-time management of wind farms and power grids.
Smart Images
Figure CN120855265A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of wind turbine grid technology, and in particular to a wind power prediction method, device and storage medium based on multi-parameter fusion. Background Technology
[0002] Wind power forecasting data is crucial for efficient grid connection of wind power and stable grid operation. Currently, traditional wind power forecasting methods rely heavily on single meteorological data, such as wind speed and direction, while neglecting multiple influencing factors such as terrain, turbine status, and grid load. Because the obstruction and acceleration effects of terrain on airflow, as well as changes in power generation efficiency caused by the aging of wind turbine equipment, are not considered, the forecast results deviate significantly from the actual power.
[0003] Existing prediction schemes based on physical or statistical models are not adaptable enough to complex and ever-changing weather conditions. In particular, when faced with extreme weather such as typhoons and cold waves, the trained models are unable to capture sudden weather changes, resulting in a significant decrease in prediction accuracy and failing to provide a reliable basis for power grid dispatch. In addition, traditional prediction methods have a low data update frequency and cannot reflect the dynamic changes in weather conditions and power grid load in real time, making it difficult to meet the real-time dispatch requirements of wind farms. Summary of the Invention
[0004] This application provides a wind power prediction method, device, and storage medium based on multi-parameter fusion to solve the above-mentioned technical problems.
[0005] The first aspect of this application provides a wind power prediction method based on multi-parameter fusion, the method comprising: The wind power multi-source data information of the wind turbine is obtained, including the wind speed of the wind turbine, the terrain where the wind turbine is located, the geographical coordinates of the wind turbine, and the grid load information of the wind turbine. A standardized dataset is constructed based on the multi-source data information; Extract the time-series periodic features from the standardized dataset; Calculate the spatial correlation matrix of the wind turbine based on its geographical coordinates; A spatiotemporal feature vector set is constructed using the periodic features of the time series and the spatial correlation matrix; The spatiotemporal feature vector set is input into a hybrid neural network model, and feature weights are dynamically allocated through an attention mechanism. Real-time multi-source data information is input into the hybrid neural network model to generate wind power prediction results, which include short-term prediction results, medium-term prediction results and long-term prediction results. The error between the predicted wind power result and the actual wind power result is obtained, and the model parameters of the hybrid neural network model are dynamically corrected according to the error to obtain the target hybrid neural network model. Wind power is predicted based on the target hybrid neural network model, and the wind power prediction results are pushed to the wind farm monitoring system and the power grid dispatch system in real time to achieve bidirectional data transmission.
[0006] Optionally, extract the time-series periodic features of the standardized dataset, including: The standardized dataset is arranged in chronological order, and it is determined whether the standardized data exhibits periodic characteristics. If present, the time-domain data is converted to frequency-domain data using Fourier transform, and the periodic components are obtained by analyzing the frequency-domain data. The periodic pattern is obtained based on the periodic components, and the statistical and temporal features of each period are extracted simultaneously.
[0007] Optionally, the spatial correlation matrix of the wind turbine is calculated based on its geographical coordinates, including: Extract the wind turbine geographic coordinates from the standardized dataset; The actual distance between each pair of wind turbines is calculated based on their geographical coordinates to obtain the measurement results; The measurement results will be filled into the distance matrix; The distance matrix is transformed into a spatial correlation matrix using an objective function.
[0008] Optionally, a spatiotemporal feature vector set is constructed using the time series periodicity features and the spatial correlation matrix, including: The periodic features of the time series and the spatial correlation matrix are fused by splicing and weighted summation to obtain high-dimensional features; The high-dimensional features are compressed to obtain a vector of uniform dimension; A spatiotemporal feature vector set is constructed based on the vector of the unified dimension.
[0009] Optionally, the spatiotemporal feature vector set is input into a hybrid neural network model, and feature weights are dynamically assigned through an attention mechanism, including: The spatiotemporal feature vector set is preprocessed to obtain a preprocessed vector; The preprocessed vector set is input into a hybrid neural network model, and the correlation between features and their impact on wind power prediction are analyzed through an attention mechanism to obtain evaluation results. The evaluation results are used to dynamically assign weights to the spatiotemporal feature vectors.
[0010] Optionally, the error between the predicted wind power result and the actual wind power result is obtained, and the model parameters of the hybrid neural network model are dynamically corrected based on the error to obtain the target hybrid neural network model, including: Obtain the quantitative error value between the wind power prediction results and the actual measured values; Based on the quantization error value, the analysis results of each parameter in the hybrid neural network model on the quantization error value are obtained; The model parameters of the hybrid neural network model are dynamically adjusted based on the analysis results to obtain the target hybrid neural network model.
[0011] Optionally, a standardized dataset is constructed based on the multi-source data information, including: The multi-source data information is preprocessed to obtain preprocessed information; The preprocessed information is integrated according to preset rules to construct a standardized dataset.
[0012] A second aspect of this application provides a wind power prediction device based on multi-parameter fusion, the device comprising: The acquisition unit is used to acquire multi-source wind power data information of the wind turbine, which includes wind turbine wind speed, wind turbine terrain, wind turbine geographical coordinates and wind turbine grid load information; The first construction unit is used to construct a standardized dataset based on the multi-source data information; An extraction unit is used to extract the time-series periodic features of the standardized dataset; The calculation unit is used to calculate the spatial correlation matrix of the wind turbine based on the wind turbine's geographical coordinates; The second construction unit is used to construct a spatiotemporal feature vector set through the periodic features of the time series and the spatial correlation matrix; The input unit is used to input the spatiotemporal feature vector set into the hybrid neural network model and dynamically allocate feature weights through an attention mechanism. The generation unit is used to input real-time multi-source data information into the hybrid neural network model to generate wind power prediction results, which include short-term prediction results, medium-term prediction results and long-term prediction results. The correction unit is used to obtain the error between the wind power prediction result and the actual wind power result, and dynamically correct the model parameters of the hybrid neural network model according to the error to obtain the target hybrid neural network model. The prediction unit is used to predict wind power based on the target hybrid neural network model and push the obtained wind power prediction results to the wind farm monitoring system and the power grid dispatch system in real time to achieve bidirectional data transmission.
[0013] A third aspect of this application provides a wind power prediction device based on multi-parameter fusion, the device comprising: Processor, memory, input / output units, and bus; The processor is connected to the memory, the input / output unit, and the bus; The storage device contains a program that the processor invokes to execute the first aspect and any optional method of the first aspect.
[0014] A fourth aspect of this application provides a computer-readable storage medium on which a program is stored, which, when executed on a computer, performs the first aspect and any optional method of the first aspect.
[0015] As can be seen from the above technical solutions, this application has the following advantages: The following are the technical effects that this technical solution can achieve: 1. By integrating multi-source data such as wind turbine speed, terrain, geographical coordinates, and power grid load, the factors affecting wind power can be comprehensively captured, effectively solving the problems caused by traditional single-data prediction and significantly improving prediction accuracy.
[0016] 2. This application utilizes an attention mechanism to dynamically adjust feature weights, enabling the model to automatically adjust key parameters based on real-time data. This improves the model's adaptability and robustness to complex scenarios, while also outputting short-term, medium-term, and long-term prediction results to meet the scheduling needs of wind farms at different time scales and provide a reference for grid planning, equipment maintenance, and power trading.
[0017] 3. This application corrects model parameters in real time based on prediction error, thereby adaptively optimizing prediction accuracy to effectively cope with changes in meteorological conditions and ensure the prediction accuracy of the model.
[0018] 4. This application enables real-time data exchange with the wind farm monitoring system and the power grid dispatching system, realizing closed-loop management of prediction, feedback and adjustment, thereby providing a reliable basis for power grid dispatching. Moreover, the data update frequency is relatively fast, which can reflect the dynamic changes of meteorological conditions and power grid load in real time, so as to meet the real-time dispatching needs of wind farms. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 This is a schematic flowchart of an embodiment of the wind power prediction method based on multi-parameter fusion in this application; Figure 2 This is a schematic diagram of another embodiment of the wind power prediction method based on multi-parameter fusion of this application; Figure 3 This is a schematic diagram of another embodiment of the wind power prediction method based on multi-parameter fusion of this application; Figure 4 This is a schematic diagram of another embodiment of the wind power prediction method based on multi-parameter fusion of this application; Figure 5 This is a schematic diagram of another embodiment of the wind power prediction method based on multi-parameter fusion of this application; Figure 6 This is a schematic diagram of another embodiment of the wind power prediction method based on multi-parameter fusion of this application; Figure 7 This is a schematic diagram of an embodiment of the wind power prediction device based on multi-parameter fusion according to this application; Figure 8 This is a schematic diagram of another embodiment of the wind power prediction device based on multi-parameter fusion of this application. Detailed Implementation
[0021] This application provides a wind power prediction method, device, and storage medium based on multi-parameter fusion, which improves the safety of the wind farm, enhances the accuracy of wind power prediction, provides a reliable basis for power grid dispatch, and has a fast data update frequency, which can reflect the dynamic changes of meteorological conditions and power grid load in real time to meet the real-time dispatch needs of wind farms.
[0022] Please see Figure 1 The first aspect of this application provides an embodiment of a wind power prediction method based on multi-parameter fusion, the embodiment including: 101. Obtain multi-source wind power data information of the wind turbine, wherein the multi-source wind power data information includes wind turbine wind speed, wind turbine terrain, wind turbine geographical coordinates and wind turbine grid load information; 102. Construct a standardized dataset based on the multi-source data information; 103. Extract the time-series periodic features of the standardized dataset; 104. Calculate the spatial correlation matrix of the wind turbine based on its geographical coordinates; 105. Construct a spatiotemporal feature vector set using the time series periodicity features and the spatial correlation matrix; 106. Input the spatiotemporal feature vector set into a hybrid neural network model and dynamically allocate feature weights through an attention mechanism; 107. Input real-time multi-source data information into the hybrid neural network model to generate wind power prediction results, including short-term prediction results, medium-term prediction results and long-term prediction results; 108. Obtain the error between the predicted wind power result and the actual wind power result, and dynamically correct the model parameters of the hybrid neural network model according to the error to obtain the target hybrid neural network model; 109. Based on the target hybrid neural network model, wind power prediction is performed, and the obtained wind power prediction results are pushed to the wind farm monitoring system and the power grid dispatching system in real time to achieve bidirectional data transmission.
[0023] In this embodiment, firstly, multi-source wind power data of the wind turbine is acquired, including wind turbine wind speed, terrain where the wind turbine is located, geographical coordinates of the wind turbine, and grid load information of the wind turbine. Then, a standardized dataset is constructed based on the multi-source data, and the time-series periodic features of the standardized dataset are further extracted. The spatial correlation matrix of the wind turbine is calculated based on its geographical coordinates. A spatiotemporal feature vector set is constructed using the time-series periodic features and the spatial correlation matrix. This spatiotemporal feature vector set is input into a hybrid neural network model, and feature weights are dynamically allocated through an attention mechanism. Then, real-time multi-source data is input into the hybrid neural network model to generate wind power prediction results, including short-term, medium-term, and long-term predictions. The error between the wind power prediction results and the actual wind power results is obtained, and the model parameters of the hybrid neural network model are dynamically corrected based on the error to obtain a target hybrid neural network model. Finally, wind power prediction is performed based on the target hybrid neural network model, and the obtained wind power prediction results are pushed in real-time to the wind farm monitoring system and the grid dispatch system to achieve bidirectional data transmission.
[0024] In step 101, the first step is to acquire multi-source wind power data from the wind turbines. This is the foundation of the wind power prediction method. The multi-source wind power data includes various data related to wind power, specifically: wind speed information, which directly relates to the turbine's power generation capacity; the terrain where the turbine is located, as different terrains affect wind speed and direction; the turbine's geographical coordinates, used for subsequent spatial correlation analysis; and information on the grid load, to understand the grid's demand for wind power. This multi-source data comes from the wind turbine's own sensors, meteorological monitoring stations, geographic information systems, and grid operation data. By comprehensively collecting this multi-source data, a data foundation is provided for subsequent analysis and prediction. After obtaining the multi-source wind power data, step 102 is executed.
[0025] In step 102, a standardized dataset is constructed based on multi-source data. Specifically, since the collected multi-source data may come from different data sources and have different formats, dimensions, and ranges, this can affect subsequent analysis and model training. Therefore, it is necessary to standardize this data. The standardization process includes data cleaning to remove noisy data, missing values, and outliers; then, normalization or standardization operations are performed to transform data with different dimensions and ranges to a uniform scale. For example, the data is normalized to the [0,1] interval or converted to a standard normal distribution with a mean of 0 and a standard deviation of 1. The standardized dataset constructed in this way can ensure the consistency and comparability of the data, and improve the accuracy and stability of the subsequent model.
[0026] In step 103, the time-series periodic features of the standardized dataset are extracted. Specifically, since wind power data has obvious time-series characteristics and exhibits certain periodicity, such as daily, weekly, and seasonal cycles, extracting these periodic features helps to capture the variation patterns of wind power. Fourier transform can be used to extract these features. Fourier transform can convert the time-domain signal to the frequency domain, thereby identifying periodic features at different frequencies. By extracting these periodic features, the variation pattern of wind power over time can be better understood, providing information for subsequent prediction models.
[0027] In step 104, the spatial correlation matrix of the wind turbines is calculated based on their geographical coordinates. Specifically, the spatial relationship between wind turbines affects their wind power output. For example, adjacent wind turbines may be affected by similar meteorological conditions, while the correlation between wind turbines that are far apart may be weaker. Based on the geographical coordinates of the wind turbines, the spatial distance between them can be calculated, and the spatial correlation matrix can be further constructed. There are various methods for calculating spatial correlation, such as the inverse distance weighting method, which assumes that the closer the wind turbines are, the higher their correlation and the greater their weight; there is also a Gaussian function-based method, which can more flexibly describe the attenuation characteristics of spatial correlation. By constructing the spatial correlation matrix, the spatial dependence between wind turbines can be captured, providing a foundation for subsequent spatiotemporal feature fusion.
[0028] In step 105, a spatiotemporal feature vector set is constructed using the time series periodic features and the spatial correlation matrix. Specifically, after extracting the time series periodic features and calculating the spatial correlation matrix, these two features need to be merged to construct the spatiotemporal feature vector set. The time features reflect the variation of wind power over time, while the spatial features reflect the interrelationships between wind turbines. By concatenating the time series feature vector and the spatial correlation feature vector, a longer feature vector is formed. Constructing the spatiotemporal feature vector set allows for the simultaneous use of information from both time and space dimensions, improving the accuracy of wind power prediction.
[0029] In step 106, the spatiotemporal feature vector set is input into the hybrid neural network model, and feature weights are dynamically allocated through an attention mechanism. Specifically, after obtaining the spatiotemporal feature vector set, it is input into the hybrid neural network model. The hybrid neural network model combines the advantages of multiple neural networks, enabling it to better handle complex spatiotemporal data. The hybrid neural network model uses Long Short-Term Memory (LSTM) networks or Gated Recurrent Units (GRUs) to process time-series data, and uses Generative Neural Networks (GNNs) to process spatially correlated data.
[0030] It should be noted that the attention mechanism can dynamically assign feature weights based on the importance of the input data. In step 106, the spatiotemporal feature vector set is input into the hybrid neural network model. The model learns the importance of different features and assigns them corresponding weights through the attention mechanism. This highlights the features that have the most impact on wind power prediction, suppresses irrelevant or noisy features, and improves the model's predictive ability.
[0031] In step 107, real-time multi-source data information is input into the hybrid neural network model to generate wind power prediction results. Specifically, after the model training is completed, the real-time collected multi-source data information is processed according to the previous steps to obtain a standardized spatiotemporal feature vector set, which is then input into the trained hybrid neural network model. The model will generate wind power prediction results based on the input feature vectors. These prediction results include short-term prediction results (e.g., the next few hours to one day), medium-term prediction results (e.g., the next few days to one week), and long-term prediction results (e.g., the next few weeks or months). Prediction results at different time scales can meet the different needs of wind farms and grid dispatch systems. For example, short-term predictions can be used for real-time dispatch and power control, while medium- and long-term predictions can be used for power generation planning and grid planning. After obtaining the prediction results from the model, step 108 is executed.
[0032] In step 108, the error between the predicted and actual wind power results is obtained, and the model parameters of the hybrid neural network model are dynamically adjusted based on the error. Specifically, to improve the prediction accuracy of the model, it is necessary to continuously evaluate and correct the model. By comparing the predicted and actual wind power results, the prediction error can be calculated. Commonly used error metrics include mean squared error (MSE) and mean absolute error (MAE). Based on these errors, the stochastic gradient descent method can be used to dynamically adjust the parameters of the hybrid neural network model. This process is similar to model retraining, but more efficient because it only requires fine-tuning the model parameters based on new error information. By continuously correcting the model parameters, the model can adapt to changes in the wind farm environment and data distribution drift, maintaining high prediction accuracy.
[0033] In step 109, wind power is predicted based on the target hybrid neural network model, and the predicted wind power results are pushed to the wind farm monitoring system and the power grid dispatching system in real time. Specifically, the target hybrid neural network model obtained after dynamic correction has high prediction accuracy. This model is used to predict wind power, and the prediction results are pushed to the wind farm monitoring system and the power grid dispatching system in real time.
[0034] Wind farm monitoring systems can rationally schedule wind turbine maintenance and optimize turbine operation based on forecast results, thereby improving the power generation efficiency and reliability of wind farms. Grid dispatching systems can better manage power production and distribution based on wind power forecasts, balancing the proportion of wind power with other energy sources and improving grid stability and economy. Simultaneously, wind farm and grid dispatching systems can feed actual operating data back to the target hybrid neural network model, forming a closed-loop, two-way data transmission system, further enhancing the accuracy and practicality of wind power forecasts.
[0035] Therefore, this application achieves real-time data exchange with the wind farm monitoring system and the power grid dispatching system, realizing closed-loop management of prediction, feedback and adjustment, thereby providing a reliable basis for power grid dispatching. Moreover, the data update frequency is relatively fast, which can reflect the dynamic changes of meteorological conditions and power grid load in real time, so as to meet the real-time dispatching needs of wind farms.
[0036] Please refer to Figure 2 According to some embodiments of the present invention, the extraction of time series periodic features from the standardized dataset in step 103 may specifically include, but is not limited to, the following: 201. Arrange the standardized dataset in chronological order and determine whether the standardized data exhibits periodic characteristics; 202. If it exists, the time-domain data is converted into frequency-domain data using Fourier transform, and the periodic components are obtained by analyzing the frequency-domain data; 203. Obtain the periodic pattern based on the periodic components, and simultaneously extract the statistical and temporal features of each period.
[0037] In this embodiment of the application, firstly, the standardized dataset needs to be arranged in chronological order to form an ordered time series, so as to better analyze the changes in the data over time.
[0038] After the data is arranged, it is necessary to determine whether the data exhibits periodicity. Periodicity refers to the recurring patterns in the data within a certain time interval, such as daily, weekly, or yearly repetitions. Several methods can be used to determine the presence of periodicity. For example, an intuitive method is to plot the data as a time series and observe whether there are obvious repeating patterns in the graph. Alternatively, the autocorrelation function (ACF) and partial autocorrelation function (PACF) can be used for analysis. The autocorrelation function measures the correlation between the same time series at different time points; if a significant correlation exists at certain fixed time intervals, it indicates that the data may exhibit periodicity.
[0039] If the data is found to have periodic characteristics, then Fourier transform is used to convert the time-domain data into frequency-domain data. Time-domain data refers to the original representation of the data arranged in chronological order, while frequency-domain data is the representation obtained after decomposing the data into components of different frequencies.
[0040] It should be noted that the Fourier transform is based on the principle that any complex time series can be regarded as a superposition of multiple sine and cosine waves of different frequencies. Through the Fourier transform, the data in the time domain can be converted into a spectrum in the frequency domain. The amplitude corresponding to each frequency in the spectrum represents the importance of that frequency component in the original data.
[0041] When analyzing frequency domain data, focus on frequency components with large amplitudes in the spectrogram. These components correspond to the main periodic patterns in the original data. For example, if a frequency with a high amplitude is found in the spectrogram, it indicates the presence of a strong periodic component in the original data with that frequency as its period. By identifying these important frequency components, it is possible to determine which periodic patterns exist in the data and their intensity.
[0042] Once the periodic components in the data are identified, the specific periodic patterns can be further obtained. A periodic pattern refers to the shape and characteristics of data changes within each period. For example, a daily periodic pattern might show a pattern where wind speed gradually increases each morning, reaches its peak at noon, and then gradually decreases in the afternoon.
[0043] While acquiring the cyclical pattern, it is also necessary to simultaneously extract the statistical and temporal characteristics of each cycle. Statistical characteristics are quantitative indicators that describe the distribution and features of data. Common statistical characteristics include mean, median, standard deviation, maximum value, and minimum value. Through these statistical characteristics, we can understand the central tendency, dispersion, and extreme cases of the data within each cycle.
[0044] Temporal characteristics refer to attributes related to time, such as the start and end times of a cycle, and its duration. These characteristics are crucial for analyzing the patterns and trends in periodic data. For example, analyzing the start time and duration of cycles in different seasons can reveal the impact of seasonal variations on wind power. Therefore, by extracting the statistical and temporal characteristics of each cycle, we can more comprehensively describe and understand the properties of periodic data, providing richer and more accurate information for subsequent wind power forecasting.
[0045] Please refer to Figure 3 According to some embodiments of the present invention, the calculation of the spatial correlation matrix of the wind turbine based on the wind turbine's geographical coordinates in step 104 may specifically include, but is not limited to, the following: 301. Extract the wind turbine geographic coordinates from the standardized dataset; 302. Calculate the actual distance between every two wind turbines based on their geographical coordinates to obtain the measurement results; 303. The measurement results will be filled into the distance matrix; 304. Use the objective function to transform the distance matrix into a spatial correlation matrix.
[0046] In this embodiment of the application, after completing the data standardization process, it is necessary to extract the geographic coordinate information related to the wind turbine location from the dataset. The geographic coordinates are represented in the form of latitude and longitude, and the coordinate data is the basis for subsequent analysis of the spatial correlation of wind turbines.
[0047] After obtaining the geographic coordinates of all wind turbines, the actual physical distance between any two turbines is calculated. Since the Earth is approximately a sphere, directly using planar distance formulas will produce significant errors. Therefore, spatial distance calculation methods based on the Earth's curvature, such as the Haversine formula, are typically used. This calculation process generates a set of measurement results containing the distances between all turbine pairs. For example, a wind farm with 100 turbines requires calculating the distances between 4950 turbine pairs. However, in practical applications, terrain information is also incorporated to correct the distances, such as considering the impact of terrain obstacles like mountains and rivers on airflow propagation.
[0048] After calculating the spatial distance using the Earth's curvature, the distance measurements between all wind turbine pairs are organized into a matrix, known as a distance matrix. It's important to note that this distance matrix is symmetric, with the number of rows and columns equal to the total number of wind turbines. Each element in the matrix represents the actual distance between two corresponding wind turbines, and the elements on the diagonal are 0 (i.e., the distance between the turbines themselves). By constructing the distance matrix, the discrete distance measurements can be structured, facilitating subsequent processing and analysis. For example, if there are three wind turbines A, B, and C, their distance matrix might be represented as follows: ABC A0d1d2 Bd10d3 Cd2d30 Where d1, d2, and d3 are the measured distances between the corresponding fans.
[0049] The distance matrix only reflects the physical distance between wind turbines, but wind power prediction focuses on the spatial correlation between turbines, that is, the degree to which a power change in one turbine affects other turbines. Generally, the closer the turbines are, the higher their correlation, but the actual correlation is also affected by various factors such as terrain, wind direction, and turbine layout. Therefore, an objective function is needed to transform physical distance into spatial correlation.
[0050] The objective function is the inverse distance weighting method. The core of the inverse distance weighting method is that the closer the distance, the greater the weight. After processing by the inverse distance weighting method, the distance matrix will be transformed into a spatial correlation matrix. The spatial correlation matrix is also symmetric, and the element values are usually between 0 and 1. The closer the value is to 1, the stronger the correlation between the two wind turbines. The spatial correlation matrix will serve as an important input for subsequent spatiotemporal feature fusion, helping the model to capture the spatial dependence between wind turbines and improve the accuracy of wind power prediction.
[0051] Please refer to Figure 4 According to some embodiments of the present invention, the construction of a spatiotemporal feature vector set by the root through the time series periodicity features and the spatial correlation matrix in step 105 may specifically include, but is not limited to, the following: 401. The periodic features of the time series and the spatial correlation matrix are fused by splicing and weighted summation to obtain high-dimensional features; 402. Compress the high-dimensional features to obtain a vector of uniform dimension; 403. Construct a spatiotemporal feature vector set based on the vector of the unified dimension. Below, I will explain the spatiotemporal feature fusion steps in detail: In this embodiment of the application, it is necessary to fuse the previously extracted time series periodic features with the spatial correlation matrix to obtain high-dimensional features that simultaneously contain both temporal and spatial information.
[0052] The first step is the splicing operation, which combines the periodic features of the time series with the spatial correlation features. Specifically, you can imagine putting the time features describing "when" and the spatial features describing "where" side by side to form a longer feature vector. This allows each data point in the longer feature vector to contain information from both time and space dimensions.
[0053] Next is the weighted summation operation. To make the model focus more on important features, different features are assigned different weights. Important features are given larger weights, while less important features are given smaller weights. The concatenated features are integrated through this weighting method to finally obtain high-dimensional features. These high-dimensional features not only retain the periodic variation of the time series, but also include the spatial correlation information between wind turbines.
[0054] The high-dimensional features obtained after the fusion operation have a very high dimension. The high dimension of the features increases the computational complexity and training difficulty of the model. Therefore, it is necessary to compress the high-dimensional features and transform them into vectors of uniform dimension.
[0055] Compression is a process of refining and summarizing information. Through dimensionality reduction methods, redundant information in the features is removed, retaining only the most crucial and representative details. This not only reduces the dimensionality of the data and lowers computational costs, but also avoids overfitting caused by the model processing too much irrelevant information. The resulting unified-dimensional vector, while having reduced dimensionality, still retains the main information from the original high-dimensional features.
[0056] After obtaining vectors of a unified dimension, these vectors can be combined to construct a spatiotemporal feature vector set. Each vector of a unified dimension represents feature information at a specific point in time and spatial location. Arranging all these vectors in a certain order forms a complete spatiotemporal feature vector set. This spatiotemporal feature vector set comprehensively describes the state of each wind turbine in the wind farm at different points in time and the interrelationships between them.
[0057] Spatiotemporal feature vector sets are input information for hybrid neural network models, which can help the models better capture the variation patterns and spatial distribution characteristics of wind power, thereby improving the accuracy and reliability of wind power prediction.
[0058] Please refer to Figure 5 According to some embodiments of the present invention, step 106, in which the spatiotemporal feature vector set is input into the hybrid neural network model and feature weights are dynamically allocated through an attention mechanism, may specifically include, but is not limited to, the following: 501. Preprocess the spatiotemporal feature vector set to obtain a preprocessed vector; 502. Input the preprocessed vector set into the hybrid neural network model, and analyze the correlation between features and their impact on wind power prediction through the attention mechanism to obtain the evaluation results; 503. Based on the evaluation results, dynamically assign weights to the spatiotemporal feature vectors. In this embodiment, before inputting the spatiotemporal feature vector set into the model, preprocessing is required to optimize the data structure and quality. The preprocessed vector set has a more regular structure. The preprocessed vector set is then input into the hybrid neural network model, and the correlation between features and its impact on wind power prediction are analyzed through an attention mechanism. Specifically, the hybrid neural network model combines the advantages of multiple network structures and can simultaneously handle time series and spatial relationships. The recurrent neural network (such as LSTM or GRU) in the model is responsible for processing time series features and capturing the variation of wind power over time; the graph neural network (such as GCN or GAT) processes spatial correlation features and mines the spatial dependencies between wind turbines. After the two networks extract information in the time and spatial dimensions respectively, they are fused through a fully connected layer.
[0059] Attention mechanisms can automatically analyze the correlation between different features and assess the importance of each feature to wind power prediction. For example, they can detect strong correlations between the historical power data of a wind turbine and the current prediction time, or identify significant power synergy among geographically adjacent wind turbines.
[0060] By analyzing the correlations between features, the attention mechanism generates an importance score for each feature. These scores constitute the evaluation results, reflecting the degree to which each feature contributes to wind power prediction.
[0061] Based on the evaluation results obtained in the previous step, the model dynamically assigns weights to each spatiotemporal feature vector. Specifically, features with high importance scores are given larger weights, meaning that these features play a more crucial role in wind power prediction, while features with low importance scores have smaller weights, and their influence is appropriately weakened. For example, if the attention mechanism finds that the periodic changes in wind speed over a certain time period have a significant impact on the prediction results, then the weight of the temporal feature vector for that time period will be increased accordingly.
[0062] It's important to note that the weights are not fixed but dynamically adjusted based on the input data. For example, under different weather conditions, such as strong winds, weak winds, or changes in wind direction, the model automatically identifies which features become more important and reassigns their weights accordingly. This dynamic adjustment capability allows the model to adapt to various complex scenarios, improving the flexibility and accuracy of predictions.
[0063] Please refer to Figure 6 According to some embodiments of the present invention, in step 108, the error between the predicted wind power result and the actual wind power result is obtained, and the model parameters of the hybrid neural network model are dynamically corrected according to the error to obtain the target hybrid neural network model. Specifically, this may include, but is not limited to, the following: 601. Obtain the quantitative error value between the predicted wind power and the actual measured value; 602. Based on the quantization error value, obtain the analysis results of each parameter in the hybrid neural network model on the quantization error value; 603. Dynamically adjust the model parameters of the hybrid neural network model based on the analysis results to obtain the target hybrid neural network model.
[0064] In this embodiment, it is necessary to compare the prediction results of the hybrid neural network model with the actual measured wind power value and calculate the quantization error between the two. The specific process is as follows: This involves mapping the predicted results to actual measurements at the same point in time. For example, if a model predicts the hourly power output of a wind turbine over the next 24 hours, these predicted values need to be matched with the actual power measurements of the turbine at the corresponding time points. Statistical metrics are used to quantify the difference between the predicted and actual values. Common error metrics include Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE). These metrics measure the accuracy of the prediction from different perspectives; for example, MAE reflects the average deviation between the predicted and actual values, while RMSE is more sensitive to larger errors.
[0065] By summing the error values for all time points and all wind turbines, a comprehensive quantified error value is obtained. This value reflects the overall prediction accuracy of the model. After obtaining the quantified error, it is necessary to further analyze the influence of each parameter in the hybrid neural network model on the error. This step is similar to "diagnosing" the model, identifying which parameters caused inaccurate predictions. The specific method is as follows: By changing a parameter in the model, observe how the error value changes. If a small change in a parameter leads to a significant increase or decrease in error, it indicates that the parameter is highly sensitive to the prediction results and needs to be adjusted. Then, use the backpropagation algorithm to calculate the gradient of the error with respect to each parameter. The gradient represents the rate and direction of error change with the parameter; parameters with larger gradients have a more significant impact on the error.
[0066] By combining the output of the attention mechanism, the contribution of different spatiotemporal features to the error can be evaluated. For example, if the feature weights for a certain time period or a certain geographical location are large, but the corresponding prediction error is also large, it indicates that the model may have insufficient processing of these features.
[0067] Considering the interactions between parameters, some parameters may have little impact on the error when viewed alone, but they can have a significant impact when combined with other parameters. By analyzing these relationships, we can gain a more comprehensive understanding of the model's behavior.
[0068] Based on the analysis results of the previous step, the parameters of the hybrid neural network model are dynamically adjusted to improve the model's prediction accuracy. After multiple iterations and parameter adjustments, the model's prediction capability will be continuously improved, and the target hybrid neural network model will be obtained. This model can more accurately predict wind power and provide reliable support for wind farm operation and grid dispatch.
[0069] Please see Figure 7 The second aspect of this application provides a wind power prediction device based on multi-parameter fusion, comprising: The acquisition unit 701 is used to acquire multi-source wind power data information of the wind turbine, which includes wind turbine wind speed, wind turbine terrain, wind turbine geographical coordinates and wind turbine grid load information. The first construction unit 702 is used to construct a standardized dataset based on the multi-source data information; Extraction unit 703 is used to extract the time series periodic features of the standardized dataset; Calculation unit 704 is used to calculate the spatial correlation matrix of the wind turbine based on the wind turbine's geographical coordinates; The second construction unit 705 is used to construct a spatiotemporal feature vector set through the time series periodic features and the spatial correlation matrix; Input unit 706 is used to input the spatiotemporal feature vector set into the hybrid neural network model and dynamically allocate feature weights through an attention mechanism; The generation unit 707 is used to input real-time multi-source data information into the hybrid neural network model to generate wind power prediction results, including short-term prediction results, medium-term prediction results and long-term prediction results. The correction unit 708 is used to obtain the error between the wind power prediction result and the actual wind power result, and dynamically correct the model parameters of the hybrid neural network model according to the error to obtain the target hybrid neural network model. The prediction unit 709 is used to predict wind power based on the target hybrid neural network model and push the obtained wind power prediction results to the wind farm monitoring system and the power grid dispatch system in real time to achieve bidirectional data transmission.
[0070] Please see Figure 8 This application also provides a wind power prediction device based on multi-parameter fusion, comprising: Processor 801, memory 802, input / output unit 803, bus 804; The processor 801 is connected to the memory 802, the input / output unit 803, and the bus 804; The memory 802 stores a program, and the processor 801 calls the program to execute any of the methods described above.
[0071] This application also relates to a computer-readable storage medium on which a program is stored, which, when run on a computer, causes the computer to perform any of the methods described above.
[0072] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0073] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.
[0074] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0075] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.
[0076] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
Claims
1. A wind power prediction method based on multi-parameter fusion, characterized in that, The method includes: The wind power multi-source data information of the wind turbine is obtained, including the wind speed of the wind turbine, the terrain where the wind turbine is located, the geographical coordinates of the wind turbine, and the grid load information of the wind turbine. A standardized dataset is constructed based on the multi-source data information; Extract the time-series periodic features from the standardized dataset; Calculate the spatial correlation matrix of the wind turbine based on its geographical coordinates; A spatiotemporal feature vector set is constructed using the periodic features of the time series and the spatial correlation matrix; The spatiotemporal feature vector set is input into a hybrid neural network model, and feature weights are dynamically allocated through an attention mechanism. Real-time multi-source data information is input into the hybrid neural network model to generate wind power prediction results, which include short-term prediction results, medium-term prediction results and long-term prediction results. The error between the predicted wind power result and the actual wind power result is obtained, and the model parameters of the hybrid neural network model are dynamically corrected according to the error to obtain the target hybrid neural network model. Wind power is predicted based on the target hybrid neural network model, and the wind power prediction results are pushed to the wind farm monitoring system and the power grid dispatch system in real time to achieve bidirectional data transmission.
2. The wind power prediction method based on multi-parameter fusion according to claim 1, characterized in that, Extracting the time-series periodic features of the standardized dataset includes: The standardized dataset is arranged in chronological order, and it is determined whether the standardized data exhibits periodic characteristics. If present, the time-domain data is converted to frequency-domain data using Fourier transform, and the periodic components are obtained by analyzing the frequency-domain data. The periodic pattern is obtained based on the periodic components, and the statistical and temporal features of each period are extracted simultaneously.
3. The wind power prediction method based on multi-parameter fusion according to claim 1, characterized in that, Calculate the spatial correlation matrix of the wind turbine based on its geographical coordinates, including: Extract the wind turbine geographic coordinates from the standardized dataset; The actual distance between each pair of wind turbines is calculated based on their geographical coordinates to obtain the measurement results; The measurement results will be filled into the distance matrix; The distance matrix is transformed into a spatial correlation matrix using an objective function.
4. The wind power prediction method based on multi-parameter fusion according to claim 1, characterized in that, A spatiotemporal feature vector set is constructed using the time series periodic features and the spatial correlation matrix, including: The periodic features of the time series and the spatial correlation matrix are fused by splicing and weighted summation to obtain high-dimensional features; The high-dimensional features are compressed to obtain a vector of uniform dimension; A spatiotemporal feature vector set is constructed based on the vector of the unified dimension.
5. The wind power prediction method based on multi-parameter fusion according to claim 1, characterized in that, The spatiotemporal feature vector set is input into a hybrid neural network model, and feature weights are dynamically allocated through an attention mechanism, including: The spatiotemporal feature vector set is preprocessed to obtain a preprocessed vector; The preprocessed vector set is input into a hybrid neural network model, and the correlation between features and their impact on wind power prediction are analyzed through an attention mechanism to obtain evaluation results. The evaluation results are used to dynamically assign weights to the spatiotemporal feature vectors.
6. The wind power prediction method based on multi-parameter fusion according to claim 1, characterized in that, The process involves obtaining the error between the predicted wind power result and the actual wind power result, and dynamically correcting the model parameters of the hybrid neural network model based on the error to obtain the target hybrid neural network model, including: Obtain the quantitative error value between the wind power prediction results and the actual measured values; Based on the quantization error value, the analysis results of each parameter in the hybrid neural network model on the quantization error value are obtained; The model parameters of the hybrid neural network model are dynamically adjusted based on the analysis results to obtain the target hybrid neural network model.
7. The wind power prediction method based on multi-parameter fusion according to claim 1, characterized in that, A standardized dataset is constructed based on the multi-source data information, including: The multi-source data information is preprocessed to obtain preprocessed information; The preprocessed information is integrated according to preset rules to construct a standardized dataset.
8. A wind power prediction device based on multi-parameter fusion, characterized in that, The device includes: The acquisition unit is used to acquire multi-source wind power data information of the wind turbine, which includes wind turbine wind speed, wind turbine terrain, wind turbine geographical coordinates and wind turbine grid load information; The first construction unit is used to construct a standardized dataset based on the multi-source data information; An extraction unit is used to extract the time-series periodic features of the standardized dataset; The calculation unit is used to calculate the spatial correlation matrix of the wind turbine based on the wind turbine's geographical coordinates; The second construction unit is used to construct a spatiotemporal feature vector set through the periodic features of the time series and the spatial correlation matrix; The input unit is used to input the spatiotemporal feature vector set into the hybrid neural network model and dynamically allocate feature weights through an attention mechanism. The generation unit is used to input real-time multi-source data information into the hybrid neural network model to generate wind power prediction results, which include short-term prediction results, medium-term prediction results and long-term prediction results. The correction unit is used to obtain the error between the wind power prediction result and the actual wind power result, and dynamically correct the model parameters of the hybrid neural network model according to the error to obtain the target hybrid neural network model. The prediction unit is used to predict wind power based on the target hybrid neural network model and push the obtained wind power prediction results to the wind farm monitoring system and the power grid dispatch system in real time to achieve bidirectional data transmission.
9. A wind power prediction device based on multi-parameter fusion, characterized in that, The device includes: Processor, memory, input / output units, and bus; The processor is connected to the memory, the input / output unit, and the bus; The storage device contains a program that the processor invokes to perform the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains a program that, when executed on a computer, performs the method as described in any one of claims 1 to 7.