Wind power plant generating capacity prediction method and system

Abstract

The invention relates to the technical field of machine learning, and discloses a wind power plant generating capacity prediction method and system. The method comprises the following steps: acquiring historical data and recent data of a wind power plant, and establishing an initial generating capacity prediction model; selecting historical meteorological factors and historical generating capacity related to the generating capacity from the historical data; training the initial generating capacity prediction model by using historical meteorological factors and historical generating capacity to obtain a generating capacity prediction model; judging whether recent data meet scoring requirements or not, and correspondingly adjusting the generating capacity prediction model; and predicting the generating capacity by using the generating capacity prediction model to obtain generating capacity data. The method has the following effects: historical data and recent data of the wind power plant are collected, and the data are put into the initial generating capacity prediction model for training, so that a complete prediction model is obtained. The prediction model can be adjusted in real time according to recent data, so that the generating capacity can be predicted more accurately and timely.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a method and system for predicting power generation of a wind farm. Background Art

[0002] After my country's wind power is connected to the power system, the randomness, intermittency and undispatchability of wind power in the power generation process have a great impact on the safe and economic dispatch of the power system. However, accurate wind power time series prediction will provide an important reference for the daily dispatch, state estimation and real-time dispatch of the power system, which is conducive to the optimal configuration of active reserve capacity on the power generation side and the formulation of demand-side dispatch plans, and improve the responsiveness of the power system dispatch to wind power access.

[0003] At present, most wind farm power generation forecasts are based on statistical analysis methods, that is, based on meteorological and wind farm historical data, power generation forecast models are calculated, such as using multiple linear regression, piecewise linear regression, time series model or grey theory forecasting method. However, such methods ignore the wake effect between units, the difference in wind rotor diameter and hub height, the failure rate and wind turbine control factors, and the forecast results are not accurate enough.

[0004] In order to improve the accuracy of the prediction results, there are also methods based on power prediction. The power prediction method usually considers the impact of wind speed on power generation and the wake effect between wind turbines. At present, the dynamic time warping algorithm (DTW) and Kalman filter algorithm are mainly used to build a power generation prediction model, and predict the future power output based on real-time wind speed, wind direction and atmospheric pressure; or based on short-term wind speed changes, according to the probability distribution of different wind speed segments and the regression curve of wind speed and power, calculate the future power output. However, the power prediction method does not consider the wind power utilization characteristics of wind turbines. At the same time, if the dynamic time warping algorithm (DTW) is used to build a power generation prediction model, since there is no objective evaluation standard for the similarity and smoothness of the time series, and the power curve of the wind farm is easily affected by the wind power control strategy, the algorithm is prone to error dispersion (Outlier); if the power output is predicted based on short-term wind speed changes, the prediction of short-term wind speed fluctuations is inaccurate and the error is large. Summary of the invention

[0005] The present invention provides a method and system for predicting the power generation of a wind farm, so as to solve the problem that in the process of predicting the power generation of a wind farm, the prediction model cannot be adjusted in real time according to the objective environment, resulting in inaccurate prediction.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides a method for predicting power generation of a wind farm, comprising: Obtain historical operation data and historical meteorological data of the wind farm, as well as recent data of the wind farm, and establish an initial power generation prediction model; the recent data includes recent meteorological data, recent operation data, and recent output log data within a preset time period from the current moment; Selecting historical meteorological factors related to wind farm power generation from the historical meteorological data; Selecting historical power generation related to the historical meteorological factors from the historical operation data; The initial power generation prediction model is trained using the historical meteorological factors and the historical power generation to obtain a first power generation prediction model; Performing multi-scale feature extraction on the recent data to obtain multi-scale feature data, and inputting the multi-scale feature data into a multi-scale normalized flow model to obtain multi-scale normalized flow data; Inputting the multi-scale normalized flow data into a hypersphere support vector machine model to perform multi-dimensional time series data classification to obtain classified output time series data; Calculate the contribution score according to the output time series data to obtain a real-time contribution score for each output time series data; When it is determined that the real-time contribution score is greater than a preset score threshold, the first power generation prediction model is trained using the recent data to obtain a second power generation prediction model; when it is determined that the real-time contribution score is less than or equal to the preset score threshold, the first power generation prediction model is directly output as the second power generation prediction model; The second power generation prediction model is used to predict the future power generation of the wind farm to obtain the power generation data of the wind farm.

[0007] In an optional implementation, the multi-scale normalized flow data is input into a hypersphere support vector machine model to perform multi-dimensional time series data classification to obtain classified output time series data, including: According to the multi-scale normalized flow data, a hypersphere is constructed in a feature space to obtain hypersphere data; According to the multi-scale normalized flow data and the hypersphere data, a data projection operation is performed and the distance between the data point and the center of the sphere is calculated to obtain the center-of-sphere distance data; According to the ball center distance data, multi-dimensional time series data classification is performed to obtain classified output time series data.

[0008] In an optional implementation, the contribution score calculation is performed according to the output time series data to obtain a real-time contribution score for each output time series data, including: According to the output time series data, the mean value of the core time series data is calculated to obtain the core center point; According to the output time series data and the core center point, the Euclidean distance between each time series data and the core center point is calculated to obtain a close distance; The contribution score of the time series data is calculated according to the proximity distance to obtain a real-time contribution score of each time series data.

[0009] In an optional implementation, the historical operation data includes the output and power of the wind farm units, and the corresponding power generation; the historical meteorological data includes the historical wind speed and the corresponding wind direction; The initial power generation prediction model includes a power prediction model based on physical factors and a statistical model based on statistical analysis methods; wherein the power prediction model includes a power function of the generated power with respect to time and wind speed; and the statistical model includes a statistical function of the output of the wind farm units with respect to wind speed.

[0010] In an optional implementation manner, the selecting of historical meteorological factors related to wind farm power generation from the historical meteorological data includes: Setting a correlation coefficient threshold between the historical meteorological data and the wind farm power generation; Calculating the correlation coefficient between the historical meteorological data and the power generation of the wind farm; The historical meteorological data corresponding to the correlation coefficient being greater than the correlation coefficient threshold is selected and output as the historical meteorological factor.

[0011] In an optional implementation, the using the historical meteorological factors and the historical power generation to train the initial power generation prediction model to obtain a first power generation prediction model includes: Using a polynomial fitting method, the power prediction model is trained according to the historical power generation to obtain a target power prediction model; Using the least square method, the statistical model is trained according to the historical meteorological factors to obtain a target statistical model; The target power prediction model and the target statistical model are subjected to a model fusion operation to obtain the first power generation prediction model.

[0012] In an optional implementation manner, the calculating the correlation coefficient between the historical meteorological data and the power generation of the wind farm includes: The correlation coefficient is calculated according to the following formula:

[0013] in, represents the correlation coefficient, Indicated in The wind speed at the moment, Indicates the average wind speed during the period of historical meteorological data. Indicated in The wind farm power generation at the time, represents the average value of wind farm power generation during the period of historical meteorological data, Represents the number of all time periods recorded in the historical meteorological factor.

[0014] In an optional implementation, the using a polynomial fitting method to train the power prediction model according to the historical power generation to obtain a target power prediction model includes: Calculate the power function according to the historical power generation; Substituting the power function into the power prediction model to obtain the target power prediction model; The power function is calculated according to the following formula:

[0015] in, Indicated in The power function at time, Indicated in The historical power generation at the time, Indicated in Wind speed at the moment.

[0016] In an optional implementation, the least square method is used to train the statistical model according to the historical meteorological factors to obtain the target statistical model, including: Calculate the statistical function according to the historical meteorological factors; Substituting the statistical function into the statistical model to obtain the target statistical model; The statistical function is calculated according to the following formula:

[0017] in, Indicated in The statistical function of time, Indicated in Wind speed at the moment Second power, Indicates the average wind speed during the period of the historical meteorological factors. Represents the number of all time periods recorded in the historical meteorological factor.

[0018] In a second aspect, the present invention provides a wind farm power generation prediction system, characterized in that it includes: The data acquisition module is used to obtain the historical operation data and historical meteorological data of the wind farm, as well as the recent data of the wind farm, and to establish an initial power generation prediction model; A historical meteorological factor selection module, used to select historical meteorological factors related to the power generation of the wind farm from the historical meteorological data; A historical power generation selection module, used to select historical power generation related to the historical meteorological factors from the historical operation data; A model training module, used to train the initial power generation prediction model using the historical meteorological factors and the historical power generation to obtain a first power generation prediction model; A model optimization module, used for performing multi-scale feature extraction on the recent data to obtain multi-scale feature data, and inputting the multi-scale feature data into a multi-scale normalized flow model to obtain multi-scale normalized flow data; inputting the multi-scale normalized flow data into a hypersphere support vector machine model to perform multi-dimensional time series data classification to obtain classified output time series data; performing contribution score calculation based on the output time series data to obtain a real-time contribution score for each output time series data; when it is determined that the real-time contribution score is greater than a preset score threshold, using the recent data to train the first power generation prediction model to obtain a second power generation prediction model; when it is determined that the real-time contribution score is less than or equal to the preset score threshold, directly outputting the first power generation prediction model as the second power generation prediction model; The model prediction module is used to predict the future power generation of the wind farm using the second power generation prediction model to obtain the power generation data of the wind farm.

[0019] In a third aspect, the present invention further provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the wind farm power generation prediction method described in any one of the above is implemented.

[0020] In a fourth aspect, the present invention further provides a computer-readable storage medium, the computer-readable storage medium comprising a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned wind farm power generation prediction methods.

[0021] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for predicting the power generation of a wind farm, comprising acquiring historical operation data and historical meteorological data of the wind farm, as well as recent data of the wind farm, and establishing an initial power generation prediction model; selecting historical meteorological factors related to the power generation of the wind farm from the historical meteorological data; selecting historical power generation related to the historical meteorological factors from the historical operation data; training the initial power generation prediction model using the historical meteorological factors and the historical power generation to obtain a first power generation prediction model; performing multi-scale feature extraction on the recent data to obtain multi-scale feature data, and inputting the multi-scale feature data into a multi-scale normalized flow model to obtain multi-scale normalized flow data; and extracting the multi-scale feature data from the recent data to obtain multi-scale feature data. The normalized flow data is input into the hypersphere support vector machine model to perform multidimensional time series data classification to obtain classified output time series data; based on the output time series data, a contribution score is calculated to obtain a real-time contribution score for each output time series data; when it is determined that the real-time contribution score is greater than a preset score threshold, the recent data is used to train the first power generation prediction model to obtain a second power generation prediction model; when it is determined that the real-time contribution score is less than or equal to the preset score threshold, the first power generation prediction model is directly output as the second power generation prediction model; the second power generation prediction model is used to predict the future power generation of the wind farm to obtain the power generation data of the wind farm.

[0022] The method obtains an initial power generation prediction model by fusing the power model and the statistical model, collects various historical data and recent data about the wind farm, and puts these data into the initial power generation prediction model for training to obtain a complete prediction model. Among them, the hypersphere support vector machine has good performance when processing high-dimensional data, can improve the accuracy of classification, and calculates the contribution score according to the classified output time series data to obtain the real-time contribution score of each output time series data, which can identify the factors that have the greatest impact on the power generation. If the real-time contribution score is greater than the preset score threshold, it means that the recent data has a significant impact on the prediction ability of the model, and it is necessary to use these data to update the first power generation prediction model to obtain the second power generation prediction model, so that it can adjust itself in real time according to the recent data, so as to more accurately and timely predict the future power generation of the wind farm. If the real-time contribution score is less than or equal to the preset score threshold, it means that the recent data has no significant impact on the prediction ability of the model, then there is no need to update the prediction model, and no additional memory and running memory of the system are occupied, which is conducive to the long-term stable and reliable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a schematic flow chart of a method for predicting wind farm power generation provided by a first embodiment of the present invention; Figure 2 It is a structural schematic diagram of a wind farm power generation prediction system provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] Reference Figure 1 The first embodiment of the present invention provides a method for predicting wind farm power generation, comprising the following steps: S1, obtain historical operation data and historical meteorological data of the wind farm, as well as recent data of the wind farm, and establish an initial power generation prediction model; S2, selecting historical meteorological factors related to the power generation of the wind farm from the historical meteorological data; S3, selecting historical power generation related to the historical meteorological factors from the historical operation data; S4, training the initial power generation prediction model using the historical meteorological factors and the historical power generation to obtain a first power generation prediction model; S5, performing multi-scale feature extraction on the recent data to obtain multi-scale feature data, and inputting the multi-scale feature data into a multi-scale normalized flow model to obtain multi-scale normalized flow data; S6, inputting the multi-scale normalized flow data into a hypersphere support vector machine model to perform multi-dimensional time series data classification to obtain classified output time series data; S7, performing contribution score calculation according to the output time series data to obtain a real-time contribution score for each output time series data; S8, when it is determined that the real-time contribution score is greater than a preset score threshold, the first power generation prediction model is trained using the recent data to obtain a second power generation prediction model; when it is determined that the real-time contribution score is less than or equal to the preset score threshold, the first power generation prediction model is directly output as the second power generation prediction model; S9: Use the second power generation prediction model to predict the future power generation of the wind farm to obtain wind farm power generation data.

[0026] It should be noted that the current wind farm power generation prediction methods mainly include statistical analysis methods and power prediction methods. Based on statistical analysis methods, that is, according to meteorological and wind farm historical data, the power generation prediction model is calculated, such as using multiple linear regression, piecewise linear regression, time series model or gray theory prediction method. However, this type of method ignores the wake effect between units, the difference in wind rotor diameter and hub height, the failure rate and wind turbine control factors, and the prediction results are not accurate enough. The power prediction method usually considers the impact of wind speed on power generation and the wake effect between wind turbines. At present, the dynamic time warping algorithm, Kalman filter algorithm, etc. are mainly used to construct the power generation prediction model, and the future power output is predicted by the real-time wind speed, wind direction and atmospheric pressure; or based on short-term wind speed changes, the future power output is calculated according to the probability distribution of different wind speed segments and the regression curve of wind speed and power. However, this method does not take into account the wind power utilization characteristics of wind turbines. At the same time, if the dynamic time warping algorithm is used to construct a power generation prediction model, since there is no objective evaluation standard for the similarity and smoothness of the time series, and the power curve of the wind farm is easily affected by the wind power control strategy, the algorithm is prone to error dispersion; if the power output is predicted based on short-term wind speed changes, the short-term wind speed fluctuations will be inaccurate and the error will be large.

[0027] In the present invention, the power model and the statistical model are integrated to obtain an initial power generation prediction model, and at the same time, various historical data and recent data about the wind farm are collected, and these data are put into the initial power generation prediction model for training to obtain a complete prediction model. The prediction model can adjust itself in real time according to recent data, so as to more accurately and timely predict the future power generation of the wind farm.

[0028] In step S1, the historical operation data and historical meteorological data of the current wind farm, as well as the recent data of the wind farm, are first obtained, and then a suitable machine learning algorithm is selected to establish an initial power generation prediction model.

[0029] Among them, the historical operation data and historical meteorological data of the wind farm, as well as the recent data of the wind farm, include the historical operation data including the output, power, and corresponding power generation of the wind farm units; the historical meteorological data include historical wind speed, and the corresponding wind direction; the recent data include recent meteorological data, recent operation data, and recent output log data within a preset time period from the current moment. Exemplarily, historical operation data refers to the operation records of the wind farm in the past period of time, including the output, power, and corresponding power generation of the wind farm units, such as data from the past few years or months, depending on the time span required by the model. Historical meteorological data refers to the records of meteorological conditions of the wind farm in the past period of time, including wind speed and wind direction. Similar to historical operation data, historical meteorological data refers to data from the past few years or months.

[0030] It is worth noting that recent data refers to data within a preset time period from the current moment, which is used to capture the latest operating and meteorological conditions to improve the accuracy of the prediction model. This preset time period can be a few days, weeks or a month, depending on the prediction requirements of the model and the availability of data. Among them, recent meteorological data includes recent meteorological parameters such as wind speed, wind direction, temperature, humidity, etc., recent operating data includes the output and power of wind farm units in the recent period, and recent output log data records the output of the units in the recent period, including output time, failures, maintenance and other events.

[0031] Specifically, it is necessary to first obtain the historical operation data and historical meteorological data of the current wind farm, as well as the recent data of the current wind farm through the records in the file data or database. Among them, the recent data of the current wind farm can also be obtained through real-time detection of various instruments installed in the wind farm. Then, according to the specific data type and characteristics of the acquired data, a suitable machine learning algorithm is selected to establish an initial power generation prediction model.

[0032] It is important to note that when choosing a suitable machine learning algorithm, you need to consider the specific type of data. For example, weather data will include satellite cloud image data, so you can choose an algorithm that is commonly used for image data. Commonly used machine learning algorithms are mainly concentrated in the field of computer vision, especially deep learning algorithms. Some common choices: Convolutional Neural Network, a neural network that can automatically learn features from images, is widely used in image classification, object detection, and image segmentation tasks; Scale Invariant Feature Transform, a feature detection algorithm used to detect and describe local features in an image, which is invariant to scaling, rotation, and lighting changes, and is suitable for tasks such as image matching and object recognition; Oriented Gradient Histogram Figure 1 A feature extraction technology used to detect objects in images. It represents the image by calculating the distribution of gradient directions in the image. It is often used in scenarios such as dynamic detection.

[0033] In one embodiment, the initial power generation prediction model includes a power prediction model based on physical factors and a statistical model based on statistical analysis methods; wherein the power prediction model includes a power function of the generated power with respect to time and wind speed; and the statistical model includes a statistical function of the output of the wind farm units with respect to wind speed.

[0034] In step S2, it is necessary to select historical meteorological factors related to the current wind farm power generation from the historical meteorological data to serve as data for subsequent model training.

[0035] In one embodiment, the selecting of historical meteorological factors related to the power generation of a wind farm from the historical meteorological data includes: setting a correlation coefficient threshold between the historical meteorological data and the power generation of a wind farm; calculating the correlation coefficient between the historical meteorological data and the power generation of a wind farm; selecting the historical meteorological data corresponding to a correlation coefficient greater than the correlation coefficient threshold, and outputting the historical meteorological factors.

[0036] Specifically, firstly, according to the relationship between the historical meteorological data and the power generation of the wind farm, a suitable statistical indicator of the degree of correlation is selected, and a corresponding threshold value of the correlation coefficient is set, that is, if the historical meteorological data and the power generation of the wind farm are exceeded within a certain interval, it can be considered that the historical meteorological data and the power generation of the wind farm are correlated. Then, the correlation coefficient is calculated according to the formula. Finally, from all the historical meteorological data, the historical meteorological data corresponding to the correlation coefficient greater than the correlation coefficient threshold is selected and output as the historical meteorological factor, so as to be used as data for subsequent model training.

[0037] In one embodiment, the calculating the correlation coefficient between the historical meteorological data and the power generation of the wind farm includes: calculating the correlation coefficient according to the following formula:

[0038] in, represents the correlation coefficient, Indicated in The wind speed at the moment, Indicates the average wind speed during the period of historical meteorological data. Indicated in The wind farm power generation at the time, represents the average value of wind farm power generation during the period of historical meteorological data, Indicates the number of all time periods recorded in the historical meteorological factors. The correlation coefficient is the Pearson correlation coefficient, which is mainly used to measure the degree of linear correlation between two variables, and its value is between -1 and 1. When the value of the correlation coefficient is positive, it means that the two variables are monotonically positively correlated; when the value of the correlation coefficient is negative, it means that the two variables are monotonically negatively correlated; the closer the value of the correlation coefficient is to 0, the less monotonous relationship there is between the two variables.

[0039] In another embodiment, the Spearman correlation coefficient is selected to determine the correlation between the historical meteorological data and the power generation of the wind farm. The Spearman correlation coefficient is calculated according to the following formula:

[0040] in, represents the Spearman correlation coefficient, represents the difference in rank between each pair of observations, that is, the difference in the positions of the corresponding observations of the two variables in the sorted list. Represents the number of observations. The Spearman correlation coefficient ranges from -1 to 1. The larger its absolute value, the stronger the rank correlation between the two variables. It should be noted that the Spearman correlation coefficient is calculated based on the rank, so its value is not affected by the distribution of the original data and has better robustness to outliers. This makes it more applicable than the Pearson correlation coefficient when dealing with data sets that do not meet the normal distribution assumption or have outliers.

[0041] In step S3, it is necessary to select historical power generation related to the historical meteorological factors from the historical operation data to serve as data for subsequent model training.

[0042] Specifically, the historical operation data includes historical power generation, and the wind farm will generate different power generation due to the historical meteorological factors. Therefore, selecting the historical power generation related to the historical meteorological factors from the historical operation data is to select representative wind farm power generation under different historical meteorological factors, such as different wind speeds and different weather conditions. For multiple power generation under similar historical meteorological factors, any power generation can be selected as the power generation under this historical meteorological factor. Finally, the power generation of wind farms under all different historical meteorological factors is combined to form the historical power generation.

[0043] It should be noted that the historical meteorological factors have many factors. To determine whether two different historical meteorological factors are similar, it is necessary to comprehensively consider indicators such as wind speed, time, location, humidity, temperature, etc., and set a certain interval for each indicator. Indicators within the same interval can be considered to be similar. Furthermore, only two different historical meteorological factors with all indicators within the same interval can be considered to be similar.

[0044] In step S4, the initial power generation prediction model is trained using the historical meteorological factors and the historical power generation to obtain a first power generation prediction model.

[0045] In one embodiment, the training of the initial power generation prediction model using the historical meteorological factors and the historical power generation to obtain the first power generation prediction model includes the following steps: S41, using a polynomial fitting method, training the power prediction model according to the historical power generation to obtain a target power prediction model.

[0046] S42, using the least square method to train the statistical model according to the historical meteorological factors to obtain a target statistical model.

[0047] S43, performing a model fusion operation on the target power prediction model and the target statistical model to obtain the first power generation prediction model.

[0048] In step S41, the historical power generation is substituted into the power prediction model for training using a polynomial fitting method, so as to obtain a target power prediction model.

[0049] It should be noted that the core of the power prediction model is the power function. By determining and calculating the power function, a specific calculation mode can be set for the power prediction model to obtain the target power prediction model, so that the model can perform accurate analysis and prediction based on the input data.

[0050] In one embodiment, the use of a polynomial fitting method to train the power prediction model according to the historical power generation to obtain a target power prediction model includes: calculating the power function according to the historical power generation; substituting the power function into the power prediction model to obtain the target power prediction model; calculating the power function according to the following formula,

[0051] in, Indicated in The power function at time, Indicated in The historical power generation at the time, Indicated in The wind speed at the moment. The power function is calculated by the polynomial fitting method, which can well simulate the actual wind farm power generation. At the same time, as the power of the polynomial fitting increases, the fitting accuracy can be improved in theory, but the amount of calculation will also increase significantly; therefore, only a cubic polynomial is fitted during the fitting, which avoids a significant increase in the amount of calculation while maintaining acceptable accuracy.

[0052] In step S42, the historical meteorological factors are substituted into the statistical model for training using the least squares method, so as to achieve the target statistical model.

[0053] It should be noted that the core of the statistical model is the statistical function. By determining and calculating the statistical function, a specific statistical mode can be set for the statistical model, thereby obtaining the target statistical model, so that the model can perform accurate statistical analysis on the input data.

[0054] In one embodiment, the least square method is used to train the statistical model according to the historical meteorological factors to obtain the target statistical model, including: calculating the statistical function according to the historical meteorological factors; substituting the statistical function into the statistical model to obtain the target statistical model; calculating the statistical function according to the following formula,

[0055] in, Indicated in The statistical function of time, Indicated in Wind speed at the moment Second power, Indicates the average wind speed during the period of the historical meteorological factors. It represents the number of all time periods recorded in the historical meteorological factors. The statistical model obtained by the least squares method can well reflect the overall trend of the data, and can also weaken the impact of anomalies when there are noise or outliers between data points. This statistical model can significantly improve the accuracy of data analysis and prediction.

[0056] In step S43, the target power prediction model and the target statistical model obtained previously are firstly obtained, and then the two models are subjected to a model fusion operation, integrated into a new model, and output as a first power generation prediction model.

[0057] Specifically, the two functions of the two models are weighted averaged to obtain a new function after weighted average, and then a suitable method is selected to stack the structural parts of the two models to obtain a new model structure, and finally the new function is put into the new model structure to obtain the first power generation prediction model. By fusing the two types of models, their advantages can be combined to improve the overall performance and accuracy, and outliers and missing values ​​can be better handled.

[0058] It should be noted that when performing weighted averaging, appropriate weights need to be selected based on actual conditions to ensure that the actual situation can be accurately simulated; similarly, when stacking the structural parts of two models, appropriate methods should be selected, such as single-layer stacking and cross-validation stacking, to ensure the integrity of the model structure.

[0059] In step S5, multi-scale feature extraction is performed on the recent data to obtain multi-scale feature data, and the multi-scale feature data is input into a multi-scale normalized flow model to obtain multi-scale normalized flow data.

[0060] In one implementation, the features of the spatial dimension are extracted from the recent data through a convolutional neural network to obtain wind power spatial features; based on the recent data, the features of the time dimension are extracted through a gated recurrent unit to obtain output time features; based on the wind power spatial features and the output time features, a multi-scale convolution operation is performed to obtain multi-scale feature data.

[0061] It should be noted that Convolutional Neural Network (CNN) is a deep learning model that can be used to extract features of spatial dimensions. These features help capture the spatial distribution characteristics of meteorological conditions such as wind speed and wind direction at different locations in a wind farm. CNN processes the input recent data (meteorological condition data, spatial distribution map of wind farm data) through its convolution layer to extract the spatial features of the wind farm. These features include local changes in wind speed, consistency of wind direction, etc., which are important factors affecting the power generation of wind farms. By using convolution kernels of different sizes and shapes, CNN can detect spatial features of different scales, which is crucial for understanding the complex meteorological patterns inside wind farms. In multi-scale convolution operations, CNN can use multiple convolution kernels of different sizes at the same time to capture spatial features of different scales, which is conducive to identifying small-scale wind speed changes in the local area and large-scale wind speed patterns of the entire wind farm. In the final stage of CNN, the extracted multi-scale spatial features are fused through fully connected layers to form a comprehensive understanding of the spatial characteristics of the wind farm, which can then be input into subsequent models together with other time dimension features for further processing.

[0062] Furthermore, a scale unification operation is performed on the multi-scale feature data to obtain unified-scale data; a reversible mapping operation is performed on the unified-scale data to obtain low-dimensional feature data; a multi-scale data enhancement operation is performed on the unified-scale data to obtain enhanced distribution data; a multi-distribution fusion operation is performed on the low-dimensional feature data and the enhanced distribution data to obtain multi-scale normalized flow data.

[0063] Among them, the scale unification operation refers to converting features with different scales or units to the same scale when processing input data, so that subsequent calculations and processing processes can be performed on a unified basis. Specifically, different data features may have different ranges, units or distributions in their original form. Through the scale unification operation, these features can be mapped to the same scale range through standardization, normalization or other transformation methods, so that the model can use these features more efficiently during training or inference, avoiding numerical deviations or instabilities caused by different scales. In the contribution scoring of recent data, the scale unification operation reduces the sensitivity of the model to outliers and noise, because the impact of outliers is weakened after standardization, thereby enhancing the robustness of the model, thereby improving the accuracy and performance of the overall evaluation.

[0064] Specifically, the calculation formula of the reversible mapping operation is: ; in, Represents the input unified scale data, such as meteorological condition data, wind farm data and output log data, obtained by the scale unification operation; Represents the latent space data transformed by the multi-scale normalized flow model; Reversible mapping function representing the normalized flow model; Represents the inverse operation of the mapping, that is, the representation of the latent space Mapping back to the input space ; The flow model Layer mapping operations, usually reversible affine transformations; express The corresponding inverse transformation.

[0065] It is worth noting that the reversible mapping operation refers to mapping high-dimensional data to low-dimensional space through a series of transformations while ensuring the reversibility of the mapping process, that is, the original high-dimensional data can be accurately restored from the low-dimensional data. The core of this operation is to minimize information loss and ensure that the important features of the original data are retained in the process of reducing the data dimension. Exemplarily, in an embodiment of the present invention, the reversible mapping operation adopts the flow model technology and is implemented through continuous bidirectional transformation (forward transformation is used for dimensionality reduction, and inverse transformation is used for recovery). In the contribution scoring of recent data, the use of reversible mapping operations can effectively compress complex multi-dimensional time series data features into low-dimensional data to reduce the computational complexity, and can restore them when necessary for more accurate contribution evaluation and feature analysis. This operation ensures the efficiency and accuracy of data processing.

[0066] In an optional implementation, the multi-scale data enhancement operation includes: Based on the unified scale data, a probability density estimation of the data distribution is performed to obtain probability distribution data; According to the probability distribution data, a layer-by-layer normalization operation is performed to obtain enhanced distribution data; The calculation formula of the probability density estimate is as follows: , in, Represents the input unified scale data, such as meteorological condition data, wind farm data and output log data, which are obtained by the scale unification operation; Represents the latent space data transformed by the multi-scale normalized flow model; Reversible mapping function representing the normalized flow model; express The probability density in the latent space is generally assumed to be Follow a simple known distribution, such as the standard normal distribution; Represents the determinant of the Jacobian matrix, which is used to describe the change in the feature space volume during the mapping process.

[0067] It should be noted that the multi-scale data enhancement operation includes two steps: probability density estimation and layer-by-layer normalization. The probability density estimation step is used to estimate the probability distribution of data of different scales. The distribution of the data is modeled by statistical methods to determine the probability of occurrence of each eigenvalue at different scales. This helps to identify abnormal data points because anomalies usually have lower density values ​​in the probability distribution. Through accurate probability density estimation, the model can effectively distinguish between normal and abnormal data patterns and improve the accuracy of the evaluation. The layer-by-layer normalization step ensures that different features are of the same numerical magnitude by normalizing each layer of the multi-scale data. Layer-by-layer normalization can avoid the adverse effects of the order of magnitude difference of the eigenvalues ​​on the model learning process, while enhancing the stability and robustness of the model when processing data of different scales. This step adjusts the data to a uniform range through a standardized or normalized formula to ensure that the model can better utilize these enhanced feature data in subsequent multi-distribution fusion operations.

[0068] Specifically, the calculation formula of the multi-distribution fusion operation is: ; in, Represents the fused multi-scale normalized flow data; Indicates The probability distribution of feature dimensions; Represents each distribution The weight parameter is used to adjust the contribution of each distribution in the fusion process; is the number of distributions that need to be fused.

[0069] It should be noted that the multi-distribution fusion operation is to fuse the data distributions of multiple different feature dimensions in a weighted manner to form a comprehensive probability distribution, which represents the final multi-scale normalized stream data. In this operation, the probability distribution of each feature dimension According to its weight parameter To adjust its contribution to the overall distribution. By weighting each distribution by power and then multiplying them, we finally get the fused distribution , which ensures that the influence of different feature dimensions can be dynamically adjusted during the fusion process. The core function of the multi-distribution fusion operation is to integrate the information of multiple feature dimensions to generate a unified probability model. This model can more accurately represent the overall data distribution, help capture cross-dimensional correlation features during contribution evaluation, and improve the accuracy of contribution evaluation. In addition, the weight parameter This makes the operation flexible and can adaptively adjust the contribution of each distribution according to the importance of different features, further improving the evaluation accuracy.

[0070] In step S6, the multi-scale normalized flow data is input into a hypersphere support vector machine model to perform multi-dimensional time series data classification to obtain classified output time series data, including: According to the multi-scale normalized flow data, a hypersphere is constructed in a feature space to obtain hypersphere data; According to the multi-scale normalized flow data and the hypersphere data, a data projection operation is performed and the distance between the data point and the center of the sphere is calculated to obtain the center-of-sphere distance data; According to the ball center distance data, multi-dimensional time series data classification is performed to obtain classified output time series data.

[0071] Among them, in machine learning, feature space refers to the high-dimensional space to which the original data is mapped through some transformation (such as kernel function). In this space, data points can be more easily distinguished by linear or nonlinear models. Through the kernel function, the original data can be mapped to a high-dimensional feature space, so that the nonlinear relationship of the data in the original space becomes linearly separable in the feature space. In high-dimensional space, a hypersphere is a geometric shape, similar to a circle in two-dimensional space or a ball in three-dimensional space. The center of the hypersphere is a certain center point of all data points in the feature space (such as the mean point), and the radius is determined by the distance between the data point and the center point. In the feature space, a hypersphere is constructed by calculating the center and radius of the data point so that the core data point is contained inside the hypersphere as much as possible. The sphere center distance refers to the distance between the data point and the center point of the hypersphere in the feature space. This distance can be used to measure the closeness of the data point to the core data distribution. Hypersphere support vector machine is a support vector machine model based on hypersphere geometry, which is used for classification tasks, especially in multidimensional time series data. It can effectively process multi-dimensional time series data, identify the relative distribution of data points and core data, and has good robustness and classification performance.

[0072] In step S7, a contribution score calculation is performed based on the output time series data to obtain a real-time contribution score for each output time series data, including: According to the output time series data, the mean value of the core time series data is calculated to obtain the core center point; According to the output time series data and the core center point, the Euclidean distance between each time series data and the core center point is calculated to obtain a close distance; The contribution score of the time series data is calculated according to the proximity distance to obtain a real-time contribution score of each time series data.

[0073] It should be noted that the core center point is obtained by calculating the mean value of the core time series data. The core center point represents the central position of the core data and is used to subsequently calculate the proximity of each time series data point to the core data center. The proximity distance is obtained by calculating the Euclidean distance between each time series data point and the core center point. The proximity distance is used to measure the proximity of each time series data point to the core data center. The smaller the proximity distance, the closer the data point is to the core data distribution. The real-time contribution score is calculated by normalization based on the proximity distance. The smaller the proximity distance, the higher the real-time contribution score. The real-time contribution score is used to quantify the contribution of each time series data point. The higher the score, the closer the data point is to the core point, the higher its contribution, and the greater the impact on the final prediction. These steps work together to effectively identify the contributing data points close to the core data distribution and provide a real-time contribution score for each time series data point to help evaluate whether the contribution of recent data is significant.

[0074] In step S8, when it is determined that the real-time contribution score is greater than the preset score threshold, the first power generation prediction model is trained using the recent data to obtain a second power generation prediction model; when it is determined that the real-time contribution score is less than or equal to the preset score threshold, the first power generation prediction model is directly output as the second power generation prediction model.

[0075] Among them, if the real-time contribution score is greater than the preset score threshold, it means that the recent data has a significant impact on the prediction ability of the model, and it is necessary to use these data to update the first power generation prediction model to obtain the second power generation prediction model. If the score is less than or equal to the preset threshold, it means that the recent data has little effect on the prediction ability of the model, and the first power generation prediction model can be directly used as the second model. Exemplarily, when the operation of the wind farm has general requirements for prediction accuracy, a higher threshold value, such as 0.8 or 0.9, can be set to ensure that the model is updated only when recent data has a significant impact on the prediction; when the operation of the wind farm has very high requirements for the real-time nature of the prediction, a lower threshold value, such as 0.5 or 0.6, can be set to update the model more frequently to capture the latest data changes.

[0076] In step S9, the future power generation of the current wind farm is predicted using the trained second power generation prediction model to obtain wind farm power generation data, and then the data is output and saved.

[0077] To facilitate understanding of the present invention, the working process of the present invention is described below using a relatively common scenario as an example.

[0078] In this embodiment, a system is planned to be implemented that can predict the future power generation of a wind farm. The system aims to accurately and in real time predict the power generation of a wind farm, and to make corresponding adjustments to the system itself according to changes in objective factors such as the wind farm and the weather.

[0079] Step 1: Obtain historical operation data, historical meteorological data and recent data of the wind farm from relevant databases, and select a suitable initial power generation prediction model.

[0080] Step 2: Select historical meteorological factors related to wind farm power generation from historical meteorological data and record them.

[0081] Step 3: Select and record the historical power generation related to historical meteorological factors from the historical operation data.

[0082] Step 4: Use historical meteorological factors and historical power generation to train the initial power generation prediction model to obtain the first power generation prediction model.

[0083] Step 5: Determine whether the recent data meets the scoring requirements, and adjust the first power generation prediction model according to the corresponding results to obtain the second power generation prediction model.

[0084] Step 6: Use the second power generation prediction model to predict the future power generation of the wind farm to obtain the power generation data of the wind farm.

[0085] Through the above steps, the wind farm power generation prediction system can accurately and timely predict the future power generation of the wind farm and make corresponding adjustments based on real-time data.

[0086] In summary, this method obtains an initial power generation prediction model by fusing the power model and the statistical model, and collects various historical data and recent data about the wind farm at the same time, and puts these data into the initial power generation prediction model for training to obtain a complete prediction model. Among them, the hypersphere support vector machine has good performance in processing high-dimensional data, which can improve the accuracy of classification. According to the classified output time series data, the contribution score is calculated to obtain the real-time contribution score of each output time series data, which can identify the factors that have the greatest impact on power generation. If the real-time contribution score is greater than the preset score threshold, it means that the recent data has a significant impact on the prediction ability of the model, and it is necessary to use these data to update the first power generation prediction model to obtain the second power generation prediction model, so that it can adjust itself in real time according to the recent data, so as to more accurately and timely predict the future power generation of the wind farm.

[0087] Reference Figure 2 The second embodiment of the present invention provides a wind farm power generation prediction system, comprising: The data acquisition module is used to obtain the historical operation data and historical meteorological data of the wind farm, as well as the recent data of the wind farm, and to establish an initial power generation prediction model; A historical meteorological factor selection module, used to select historical meteorological factors related to the power generation of the wind farm from the historical meteorological data; A historical power generation selection module, used to select historical power generation related to the historical meteorological factors from the historical operation data; A model training module, used to train the initial power generation prediction model using the historical meteorological factors and the historical power generation to obtain a first power generation prediction model; A model optimization module, used for performing multi-scale feature extraction on the recent data to obtain multi-scale feature data, and inputting the multi-scale feature data into a multi-scale normalized flow model to obtain multi-scale normalized flow data; inputting the multi-scale normalized flow data into a hypersphere support vector machine model to perform multi-dimensional time series data classification to obtain classified output time series data; performing contribution score calculation based on the output time series data to obtain a real-time contribution score for each output time series data; when it is determined that the real-time contribution score is greater than a preset score threshold, using the recent data to train the first power generation prediction model to obtain a second power generation prediction model; when it is determined that the real-time contribution score is less than or equal to the preset score threshold, directly outputting the first power generation prediction model as the second power generation prediction model; The model prediction module is used to predict the future power generation of the wind farm using the second power generation prediction model to obtain the power generation data of the wind farm.

[0088] It should be noted that a wind farm power generation prediction system provided in an embodiment of the present invention is used to execute all process steps of a wind farm power generation prediction method in the above embodiment, and the working principles and beneficial effects of the two correspond one to one, so they will not be described in detail.

[0089] The embodiment of the present invention further provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a wind farm power generation prediction program. When the processor executes the computer program, the steps in the above-mentioned wind farm power generation prediction method embodiments are implemented, such as Figure 1 Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned system embodiments are realized, such as the historical meteorological factor selection module.

[0090] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the electronic device.

[0091] The electronic device may be a computing device such as a desktop computer, a notebook, a PDA, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. The electronic device may include more or fewer components than the above components, or may combine certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0092] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the electronic device, and uses various interfaces and lines to connect various parts of the entire electronic device.

[0093] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the electronic device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0094] Wherein, if the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or system capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0095] It should be noted that the system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the system embodiment provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0096] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting wind farm power generation, characterized in that: include: Obtain historical operation data and historical meteorological data of the wind farm, as well as recent data of the wind farm, and establish an initial power generation prediction model; the recent data includes recent meteorological data, recent operation data, and recent output log data within a preset time period from the current moment; Selecting historical meteorological factors related to wind farm power generation from the historical meteorological data; Selecting historical power generation related to the historical meteorological factors from the historical operation data; The initial power generation prediction model is trained using the historical meteorological factors and the historical power generation to obtain a first power generation prediction model; Performing multi-scale feature extraction on the recent data to obtain multi-scale feature data, and inputting the multi-scale feature data into a multi-scale normalized flow model to obtain multi-scale normalized flow data; Inputting the multi-scale normalized flow data into a hypersphere support vector machine model to perform multi-dimensional time series data classification to obtain classified output time series data; Calculate the contribution score according to the output time series data to obtain a real-time contribution score for each output time series data; When it is determined that the real-time contribution score is greater than a preset score threshold, the first power generation prediction model is trained using the recent data to obtain a second power generation prediction model; when it is determined that the real-time contribution score is less than or equal to the preset score threshold, the first power generation prediction model is directly output as the second power generation prediction model; The second power generation prediction model is used to predict the future power generation of the wind farm to obtain the power generation data of the wind farm.

2. The method for predicting wind farm power generation according to claim 1, characterized in that: The multi-scale normalized flow data is input into the hypersphere support vector machine model to perform multi-dimensional time series data classification to obtain the classified output time series data, including: According to the multi-scale normalized flow data, a hypersphere is constructed in a feature space to obtain hypersphere data; According to the multi-scale normalized flow data and the hypersphere data, a data projection operation is performed and the distance between the data point and the center of the sphere is calculated to obtain the center-of-sphere distance data; According to the ball center distance data, multi-dimensional time series data classification is performed to obtain classified output time series data.

3. The method for predicting wind farm power generation according to claim 2, characterized in that: The step of calculating the contribution score according to the output time series data to obtain the real-time contribution score of each output time series data includes: According to the output time series data, the mean value of the core time series data is calculated to obtain the core center point; According to the output time series data and the core center point, the Euclidean distance between each time series data and the core center point is calculated to obtain a close distance; The contribution score of the time series data is calculated according to the proximity distance to obtain a real-time contribution score of each time series data.

4. The method for predicting wind farm power generation according to claim 1, characterized in that: The historical operation data includes the output and power of the wind farm units, and the corresponding power generation; the historical meteorological data includes the historical wind speed and the corresponding wind direction; The initial power generation prediction model includes a power prediction model based on physical factors and a statistical model based on statistical analysis methods; wherein the power prediction model includes a power function of the generated power with respect to time and wind speed; and the statistical model includes a statistical function of the output of the wind farm units with respect to wind speed.

5. The method for predicting wind farm power generation according to claim 1, characterized in that: The selecting of historical meteorological factors related to the power generation of the wind farm from the historical meteorological data includes: Setting a correlation coefficient threshold between the historical meteorological data and the wind farm power generation; Calculating the correlation coefficient between the historical meteorological data and the power generation of the wind farm; The historical meteorological data corresponding to the correlation coefficient being greater than the correlation coefficient threshold is selected and output as the historical meteorological factor.

6. The method for predicting wind farm power generation according to claim 4, characterized in that: The using the historical meteorological factors and the historical power generation to train the initial power generation prediction model to obtain a first power generation prediction model includes: Using a polynomial fitting method, the power prediction model is trained according to the historical power generation to obtain a target power prediction model; Using the least square method, the statistical model is trained according to the historical meteorological factors to obtain a target statistical model; The target power prediction model and the target statistical model are subjected to a model fusion operation to obtain the first power generation prediction model.

7. The method for predicting wind farm power generation according to claim 5, characterized in that: The calculating of the correlation coefficient between the historical meteorological data and the power generation of the wind farm comprises: The correlation coefficient is calculated according to the following formula: in, represents the correlation coefficient, Indicated in The wind speed at the moment, Indicates the average wind speed during the period of historical meteorological data. Indicated in The wind farm power generation at the time, represents the average value of wind farm power generation during the period of historical meteorological data, Represents the number of all time periods recorded in the historical meteorological factor.

8. The method for predicting wind farm power generation according to claim 6, characterized in that: The method of using a polynomial fitting method to train the power prediction model according to the historical power generation to obtain a target power prediction model includes: Calculate the power function according to the historical power generation; Substituting the power function into the power prediction model to obtain the target power prediction model; The power function is calculated according to the following formula: in, Indicated in The power function at time, Indicated in The historical power generation at the time, Indicated in Wind speed at the moment.

9. The method for predicting wind farm power generation according to claim 6, characterized in that: The method of using the least square method to train the statistical model according to the historical meteorological factors to obtain the target statistical model includes: Calculate the statistical function according to the historical meteorological factors; Substituting the statistical function into the statistical model to obtain the target statistical model; The statistical function is calculated according to the following formula: in, Indicated in The statistical function of time, Indicated in Wind speed at the moment Second power, Indicates the average wind speed during the period of the historical meteorological factors. Represents the number of all time periods recorded in the historical meteorological factor.

10. A wind farm power generation prediction system, characterized in that: A method for predicting wind farm power generation according to any one of claims 1 to 9, comprising: The data acquisition module is used to obtain the historical operation data and historical meteorological data of the wind farm, as well as the recent data of the wind farm, and to establish an initial power generation prediction model; A historical meteorological factor selection module, used to select historical meteorological factors related to the power generation of the wind farm from the historical meteorological data; A historical power generation selection module, used to select historical power generation related to the historical meteorological factors from the historical operation data; A model training module, used to train the initial power generation prediction model using the historical meteorological factors and the historical power generation to obtain a first power generation prediction model; A model optimization module, used for performing multi-scale feature extraction on the recent data to obtain multi-scale feature data, and inputting the multi-scale feature data into a multi-scale normalized flow model to obtain multi-scale normalized flow data; inputting the multi-scale normalized flow data into a hypersphere support vector machine model to perform multi-dimensional time series data classification to obtain classified output time series data; performing contribution score calculation based on the output time series data to obtain a real-time contribution score for each output time series data; when it is determined that the real-time contribution score is greater than a preset score threshold, using the recent data to train the first power generation prediction model to obtain a second power generation prediction model; when it is determined that the real-time contribution score is less than or equal to the preset score threshold, directly outputting the first power generation prediction model as the second power generation prediction model; The model prediction module is used to predict the future power generation of the wind farm using the second power generation prediction model to obtain the power generation data of the wind farm.

Images