A wind power prediction method, system and device based on multi-model fusion

Abstract

The present application discloses a wind power prediction method, system and device based on multi-model fusion, which relates to the field of wind power prediction technology. By combining the Pearson correlation coefficient, the frost optimization algorithm to optimize the variational mode decomposition model, the CNN-ASSA-Informer model and the quantile regression, the non-parametric probability prediction of short-term wind power prediction is realized. The complex relationship between different frequency components and wind power in the time series data is solved by processing complex time series data through the optimization algorithm. The corresponding prediction interval is constructed, which improves the accuracy and confidence of wind power prediction, and can provide more reliable wind power prediction results for regional wind power scheduling, renewable energy integration and power market.

Description

Technical Field

[0001] The present application relates to the technical field of wind power prediction, and in particular to a method, system and device for wind power prediction based on multi-model fusion. Background Art

[0002] As the main source of renewable energy, wind power generation has been widely used around the world. However, the output of wind power generation has great uncertainty, which is mainly affected by meteorological conditions. The characteristics of wind power generation are intermittent and volatile, which makes the prediction of wind power face many challenges. Especially when large-scale wind power is connected to the power grid, how to accurately predict the power generation of wind power and reduce the impact of uncertainty on the power grid has become a hot topic in current research.

[0003] Accurate wind power forecasting can effectively regulate grid load and alleviate the impact of wind power generation on grid stability. It is of great significance for grid management, reducing dependence on backup power sources and increasing wind energy penetration. Although studies have explored wind power forecasting methods in traditional technologies, current wind power forecasting methods mainly focus on deterministic wind power forecasting, usually using machine learning and deep learning technologies to build models to predict future wind power. However, due to the high uncertainty of wind speed and direction, traditional point forecasting methods have large errors and cannot fully consider the volatility and intermittency of wind power. Existing wind power forecasting methods are still insufficient in processing and feature selection of large-scale data sets, especially in terms of the time series characteristics and volatility of wind power data. Existing methods fail to fully mine effective information.

[0004] Therefore, how to use improved deep learning models combined with optimization algorithms to perform probabilistic prediction of wind power timing and improve the accuracy and reliability of overall wind power prediction has become a problem that needs to be solved urgently. Summary of the invention

[0005] In order to solve the above technical problems, this application proposes the following technical solutions:

[0006] In a first aspect, an embodiment of the present application provides a wind power prediction method based on multi-model fusion, comprising:

[0007] Collect the operating values ​​and weather forecast data of multiple wind turbines within a preset time period and process abnormal values ​​and missing values ​​in the data;

[0008] Use the Pearson correlation coefficient to perform feature selection, select feature data that is strongly correlated with the target wind power from the processed data, and establish a training set, a validation set, and a test set;

[0009] The Frost optimization algorithm is used to optimize the variational mode decomposition VMD model to obtain the best parameters. Based on the optimized VMD results, the data is decomposed into multiple frequency components.

[0010] Establish a CNN-ASSA-Informer model, use the training set to train the model, use the validation set to diagnose the model adaptability and optimize the model hyperparameters, and use the test set to make predictions and evaluate the model;

[0011] Quantile regression methods are used to construct prediction intervals at different confidence levels based on the results.

[0012] In a possible implementation, the collecting of operating values ​​and weather forecast data of multiple wind turbines within a preset time period and processing of abnormal values ​​and missing values ​​in the data includes:

[0013] Record any actual operating data related to wind power generation at preset intervals within a preset time period, including: power generation, wind direction, wind speed, temperature and humidity of each wind turbine;

[0014] The feature data with abnormal values ​​and missing values ​​are determined, and the missing values ​​or abnormal values ​​are replaced by the feature means of the corresponding feature data.

[0015] In a possible implementation, the feature selection is performed using the Pearson correlation coefficient, a feature column strongly correlated with the target wind power is selected from the processed data, and a training set, a validation set, and a test set are established, including:

[0016] The Pearson correlation coefficient between each feature and the target wind power is calculated, and the feature columns with a strong correlation with the target wind power are selected for modeling to improve the prediction accuracy of the model. The range of the Pearson correlation coefficient is [-1, 1], where 0 represents no correlation, a positive value represents a positive correlation, and a negative value represents a negative correlation. The Pearson correlation coefficient calculation formula is:

[0017]

[0018] in, and is the feature sequence, and is a variable, For sequence expectations, For sequence expectations, For variables The variance of For variables The variance of for The mean of for The mean of

[0019] For sequence The length of the sequence Length.

[0020] In a possible implementation, the frost optimization algorithm is used to optimize the variational mode decomposition VMD model to obtain the best parameters, and based on the optimized VMD results, the data is decomposed into multiple frequency components, including:

[0021] Construct a variational mode decomposition (VMD) model for wind power prediction;

[0022] The wind power time series power data is decomposed by variational mode decomposition (VMD) to obtain several frequency components, each of which represents a different frequency component in the signal.

[0023] In the VMD decomposition process, the frost optimization algorithm is used to optimize the key parameters of VMD, including the number of modes, penalty coefficient and bandwidth, to improve the accuracy of signal decomposition; by simulating the natural cooling process, the parameters in VMD are gradually adjusted to find the optimal mode decomposition parameters;

[0024] The above process is repeated until all modal components are stable and meet the set convergence conditions, thereby obtaining different frequency characteristics of wind power data so that subsequent models can better capture the time series characteristics and change laws in the data.

[0025] In a possible implementation, the construction of a variational mode decomposition (VMD) model for wind power prediction includes:

[0026] Construct a variational problem to ensure that the decomposition sequence is a modal component with a finite bandwidth and a center frequency. At the same time, the sum of the estimated bandwidths of each mode is minimized. The constraint condition is that the sum of all modes is equal to the original signal. The VMD constrained variational model is as follows:

[0027]

[0028] in, is the initial signal, is the number of decomposed modes, are the components and center frequencies of the decomposed Kth mode, is the Dirac function, is the convolution operation;

[0029] Solving the variational problem, introducing a quadratic penalty factor and Lagrange multipliers , transforming the minimization problem into an unconstrained optimization problem, the augmented Lagrangian function is as follows:

[0030]

[0031] Iterative update using alternating direction multiplier method , find the saddle point of the augmented Lagrangian function in the iterative optimization sequence, and its calculation formula is:

[0032]

[0033]

[0034]

[0035] In the formula, represents the noise margin, , , , Corresponding to , , , Fourier transform of

[0036] Define the maximum number of iterations as ,make satisfy , for a given solution accuracy , when the iteration is completed, the following convergence condition is satisfied, and the final ;

[0037] .

[0038] In a possible implementation, the CNN-ASSA-Informer model is a neural network in which a convolutional neural network and an adaptive sparse attention mechanism are connected in series and then in parallel with an informer. The convolutional neural network includes an input layer, a convolution layer, an activation layer and an output layer. The sparse self-attention mechanism includes a window self-attention layer and a relative position bias layer. The informer includes an encoder layer and a decoder layer. The CNN model is used to extract local features in time series data and enhance the model's learning ability for key features through an adaptive sparse self-attention mechanism. The informer model is used to process long time series dependencies.

[0039] In a possible implementation, the method of training the model using the training set, diagnosing the model adaptability and optimizing the model hyperparameters using the validation set, and predicting and evaluating the model using the test set includes:

[0040] The data enters the convolution layer through the input layer, is convolved with the convolution kernel in the convolution layer, and the convolution layer outputs the convolution result to the activation layer;

[0041] The activation layer uses the ReLU activation function to perform nonlinear transformation on the convolution result; after being processed by the convolution layer and the activation layer, the convolutional neural network can extract local temporal features in the input data to form high-level information features;

[0042] The feature map output by the convolutional neural network is then processed by an adaptive sparse self-attention mechanism. The local temporal dependency is effectively modeled through self-attention calculation within the local window. At the same time, the feature relationship within the window is adjusted using relative position bias to enhance the expressiveness of the feature.

[0043] Then, the features processed by the sparse self-attention mechanism are added and fused with the global temporal information features output by the Informer encoder to generate the final multimodal features. The multimodal features enter the decoder as input. The decoder further models the global temporal dependencies through the self-attention mechanism, captures long-term dependencies, and obtains the prediction results.

[0044] In a possible implementation, the use of the quantile regression method to construct prediction intervals at different confidence levels according to the results includes:

[0045] Given a response variable and feature variables , define the conditional distribution function for:

[0046]

[0047] For each given ,definition Quantile for:

[0048]

[0049] This quantile represents the Next, the response variable of -Quantile;

[0050] Quantile regression estimates regression models at different quantiles by minimizing the quantile loss function. The definition is as follows:

[0051]

[0052] in is the actual observed value, is the predicted value of the model, is the chosen quantile;

[0053] Then use the quantile regression results to establish the prediction interval, given the input , prediction interval It can be expressed as:

[0054] .

[0055] In a second aspect, an embodiment of the present application provides a wind power prediction system based on multi-model fusion, including:

[0056] A data acquisition and processing module is used to collect operating values ​​and weather forecast data of multiple wind turbines within a preset time period and process abnormal values ​​and missing values ​​in the data;

[0057] A feature selection module is used to perform feature selection using the Pearson correlation coefficient, select feature data that is strongly correlated with the target wind power from the processed data, and establish a training set, a validation set, and a test set;

[0058] The data decomposition module is used to optimize the variational mode decomposition (VMD) model using the Frost optimization algorithm to obtain the best parameters, and decompose the data into multiple frequency components based on the optimized VMD results;

[0059] The model building module is used to build the CNN-ASSA-Informer model, train the model using the training set, diagnose the model adaptability and optimize the model hyperparameters using the validation set, and make predictions and evaluate the model using the test set;

[0060] The forecasting module is used to construct prediction intervals at different confidence levels based on the results using the quantile regression method.

[0061] In the third aspect, an embodiment of the present application provides a wind power prediction device based on multi-model fusion, comprising: a processor; a memory; and a computer program, wherein the computer program is stored in the memory, and the computer program includes instructions, and when the instructions are executed by the processor, the device executes the method described in the first aspect and any possible implementation method of the first aspect.

[0062] In the embodiments of the present application, by combining the Pearson correlation coefficient, the frost optimization algorithm to optimize the variational mode decomposition model, the CNN-ASSA-Informer model and the quantile regression, the non-parametric probabilistic prediction of short-term wind power forecasting is realized, the complex relationship between different frequency components in the time series data and the wind power is solved, the accuracy and confidence of the wind power forecast are improved, and more reliable wind power forecasting results can be provided for regional wind power scheduling, renewable energy integration and power markets. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A schematic diagram of a flow chart of a wind power prediction method based on multi-model fusion provided in an embodiment of the present application;

[0064] Figure 2 A heat map of correlation coefficients between all features of wind power data provided in the embodiment of the present application;

[0065] Figure 3 A structural diagram of the CNN-ASSA-Informer model provided in the embodiment of the present application;

[0066] Figure 4 A comparison chart of the model predicted power and the actual wind power provided in the embodiment of the present application;

[0067] Figure 5 A histogram of power error predicted by the model provided in the embodiment of the present application;

[0068] Figure 6 A schematic diagram of the prediction interval of the model provided in the embodiment of the present application at different confidence levels;

[0069] Figure 7 A schematic diagram of a wind power prediction system based on multi-model fusion provided in an embodiment of the present application. DETAILED DESCRIPTION

[0070] The present solution is described below in conjunction with the accompanying drawings and specific implementation methods.

[0071] See also Figure 1 The wind power prediction method based on multi-model fusion provided in this embodiment includes:

[0072] S101, collecting operating values ​​and weather forecast data of multiple wind turbines within a preset time period and processing abnormal values ​​and missing values ​​in the data.

[0073] Record any actual operation data related to wind power generation at preset intervals within a preset time period, including: power generation, wind direction, wind speed, temperature and humidity of each wind turbine. Determine the feature data with abnormal values ​​and missing values, and replace the missing values ​​or abnormal values ​​by using the feature mean of the corresponding feature data.

[0074] In this embodiment, any actual operation data related to wind power generation is recorded every 10 minutes for 134 wind turbines in the wind farm for 245 days, including but not limited to the power generation, wind direction, wind speed, temperature and humidity of each wind turbine. In this embodiment, the time scale for obtaining the actual operation data is 10 minutes. Compared with hourly or daily data, 10-minute data can more accurately reflect the dynamic changes and fluctuations of wind power generation, and can ensure that the system can obtain the latest wind power generation and consumption in a timely manner, so as to make timely responses and adjustments.

[0075] Furthermore, outliers and missing values ​​are processed by the mean filling method. By using the mean of the feature in the data to replace missing values ​​or outliers, the integrity of the data can be retained without introducing too much deviation, and the model training error caused by missing values ​​or outliers can be reduced, thereby improving the quality and reliability of subsequent modeling. It also includes normalizing the actual operation data and predicted data, that is, converting data of different ranges to the same dimension (between 0 and 1) to facilitate the learning and training of subsequent models.

[0076] S102, using the Pearson correlation coefficient to perform feature selection, selecting feature data that is strongly correlated with the target wind power from the processed data, and establishing a training set, a validation set, and a test set.

[0077] The Pearson correlation coefficient is used to calculate the correlation between each feature and the target wind power. Feature columns with a strong correlation with the target power are selected for modeling to improve the prediction accuracy of the model. The range of the Pearson correlation coefficient is [-1,1], where 0 represents no correlation, a positive value represents a positive correlation, and a negative value represents a negative correlation. The Pearson correlation coefficient is optimized relative to the Euclidean distance by decentralizing the target vector and then using the decentralized value to calculate the cosine distance. Let the sequence X = ( , ,… ), Y=( , ,… ) variables are x and y, then the formula for calculating the correlation coefficient is:

[0078]

[0079] in, and is the feature sequence, and is a variable, For sequence expectations, For sequence expectations, For variables The variance of For variables The variance of for The mean of for The mean of

[0080] For sequence The length of the sequence The length of the data features is Figure 2 shown.

[0081] Figure 2 The data feature labels corresponding to the vertical direction are wind speed, wind direction, ambient temperature, cabin temperature, cabin angle, first blade pitch angle, second blade pitch angle, third blade pitch angle, reactive power, and active power from top to bottom. The data feature labels corresponding to the horizontal direction are wind speed, wind direction, ambient temperature, cabin temperature, cabin angle, first blade pitch angle, second blade pitch angle, third blade pitch angle, reactive power, and active power from left to right. For example, Figure 2 The correlation coefficient between the first row and the first column, i.e. the wind speed and itself, is 1; the correlation coefficient between the first row and the second column, i.e. the wind speed and the wind direction, is -0.027; the correlation coefficient between the first row and the third column, i.e. the wind speed and the ambient temperature, is 0.16; the remaining data correspond to the above-mentioned identification information in sequence, and will not be described in detail.

[0082] S103, using the Frost optimization algorithm to optimize the variational mode decomposition VMD model to obtain optimal parameters, and based on the optimized VMD results, decomposing the data into multiple frequency components.

[0083] VMD is a non-recursive signal processing method that is very suitable for processing non-stationary data such as wind power. It aims to decompose complex nonlinear non-stationary signals into a series of modal functions with different center frequencies.

[0084] The core idea of ​​VMD is to construct and solve variational problems. The steps to construct a variational mode decomposition VMD model for wind power prediction are as follows:

[0085] Step 1: Construct a variational problem. Ensure that the decomposed sequence is a modal component with a finite bandwidth and a center frequency, and the sum of the estimated bandwidths of each mode is minimized. The constraint condition is that the sum of all modes is equal to the original signal. The VMD constrained variational model is as follows:

[0086]

[0087] in, is the initial signal, is the number of decomposed modes, are the components and center frequencies of the decomposed Kth mode, is the Dirac function, is the convolution operation.

[0088] Step 2: Solve the variational problem. Introduce the quadratic penalty factor and Lagrange multipliers , transforming the minimization problem into an unconstrained optimization problem, the augmented Lagrangian function is as follows:

[0089]

[0090] Step 3: Iterative update using alternating direction multiplier method , find the saddle point of the augmented Lagrangian function in the iterative optimization sequence, and its calculation formula is:

[0091]

[0092]

[0093]

[0094] In the formula, represents the noise margin, , , , Corresponding to , , , The Fourier transform of .

[0095] Define the maximum number of iterations as ,make satisfy , for a given solution accuracy , when the iteration is completed, the following convergence condition is satisfied, and the final .

[0096] .

[0097] Furthermore, the Rime Optimization Algorithm (RIME) is used to optimize the modal decomposition parameters of VMD and obtain the optimal parameters to ensure the optimal decomposition of the data. The wind power time series power data is decomposed by variational mode decomposition (VMD) to obtain several frequency components, each modal component representing a different frequency component in the signal; in the VMD decomposition process, the Rime Optimization Algorithm (RIME) is used to optimize the key parameters of VMD, including the number of modes, penalty coefficient and bandwidth, to improve the accuracy of signal decomposition; specifically, the RIME algorithm gradually adjusts the parameters in VMD by simulating the natural cooling process to find the optimal modal decomposition parameters; the RIME optimization process reduces the influence of local optimal solutions and enhances the global adaptability of the decomposition results by optimizing the VMD parameters; the above process is repeated until all modal components are stable and meet the set convergence conditions, thereby obtaining different frequency characteristics of wind power data, so that subsequent models can better capture the time series characteristics and change laws in the data.

[0098] S104, establish a CNN-ASSA-Informer model, use the training set to train the model, the validation set to diagnose the model adaptability and optimize the model hyperparameters, and the test set to make predictions and evaluate the model.

[0099] Specifically, the CNN-ASSA-Informer model is a convolutional neural network and an adaptive sparse attention mechanism in series and then in parallel with an informer. Its structure is as follows: Figure 3 As shown in the figure. The convolutional neural network includes an input layer, a convolution layer, an activation layer and an output layer. The sparse self-attention mechanism includes a window self-attention layer and a relative position bias layer. The Informer includes an encoder layer and a decoder layer. The CNN model is used to extract local features in time series data and enhance the model's learning ability for key features through an adaptive sparse self-attention mechanism. The Informer model is used to process long time series dependencies.

[0100] Furthermore, the prediction process of the CNN-ASSA-Informer model is as follows: first, the data enters the convolution layer through the input layer, and is convolved with the convolution kernel in the convolution layer. The convolution layer outputs the convolution result to the activation layer, and the activation layer uses the ReLU activation function to perform nonlinear transformation on the convolution result; after being processed by the convolution layer and the activation layer, the convolution neural network can extract the local temporal features in the input data and form high-level information features; then, the feature map output by the convolution neural network is processed by the adaptive sparse self-attention mechanism, and the local temporal dependency is effectively modeled through the self-attention calculation in the local window, and the feature relationship in the window is adjusted by the relative position bias to enhance the feature expression ability; then, the features processed by the sparse self-attention mechanism are combined with the global temporal information features output by the Informer encoder through addition to generate the final multimodal features. The multimodal features are input to the decoder, and the decoder further models the global temporal dependency through the self-attention mechanism, captures the long-term dependency, and obtains the prediction results.

[0101] To verify the prediction performance of the proposed model, the prediction results of the model in this application are compared with the long short-term memory network LSTM and Informer on the same data set. The comparison between the model prediction power and the actual wind power is shown in the figure below. Figure 4 As shown, Figure 5 The error comparison of model prediction evaluation indicators is shown, and the prediction accuracy of the model is evaluated by mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE).

[0102] S105, use the quantile regression method to construct prediction intervals at different confidence levels based on the results.

[0103] Specifically, the quantile regression method is a non-parametric model that does not rely on the distribution assumption of the data and can effectively deal with outliers and heteroscedasticity problems. The goal of quantile regression is to estimate the quantile of the response variable by fitting the conditional distribution function of different quantiles. Specifically, given the response variable and feature variables , we can define the conditional distribution function for:

[0104]

[0105] For each given , we can define -Quantile for:

[0106]

[0107] This quantile represents the Next, the response variable of -Quantile.

[0108] Quantile regression estimates regression models at different quantiles by minimizing the quantile loss function. The definition is as follows:

[0109]

[0110] in is the actual observed value, is the model's predicted value, is the chosen quantile.

[0111] Then, the prediction interval is established using the quantile regression results. For example, to obtain the 90% prediction interval, the model will output the 5% and 95% quantile prediction values. Specifically, given the input , prediction interval It can be expressed as:

[0112]

[0113] here, is a given feature The 5% quantile below represents the lower limit forecast of wind power; The given feature The 95% quantile under the above value indicates the upper limit forecast of wind power. This method provides an upper and lower limit forecast interval, rather than just a point estimate, thereby realizing the uncertainty estimation of regional wind power generation and providing risk assessment and decision support for wind farm operation and management. Figure 6 The prediction intervals of the CNN-ASSA-Informer-QR model at 95%, 75%, 50%, and 25% confidence levels are shown.

[0114] This embodiment combines a variety of advanced technologies, such as RIME optimization algorithm, VMD, CNN, ASSA and Informer models, and proposes a wind power forecasting method based on multi-model fusion. Experimental results show that this method can improve the accuracy and reliability of wind power forecasting by effectively processing noise, missing values ​​and outliers in the data and optimizing the model structure. It is also highly adaptable and can perform effective forecasting in complex and changing environments.

[0115] Corresponding to the wind power prediction method based on multi-model fusion provided in the above embodiment, the present application also provides an embodiment of a wind power prediction system based on multi-model fusion.

[0116] See also Figure 7The wind power prediction system 20 based on multi-model fusion in this embodiment includes:

[0117] The data collection and processing module 201 is used to collect the operating values ​​and weather forecast data of multiple wind turbines within a preset time period and process abnormal values ​​and missing values ​​in the data.

[0118] The feature selection module 202 is used to perform feature selection using the Pearson correlation coefficient, select feature data that is strongly correlated with the target wind power from the processed data, and establish a training set, a validation set, and a test set.

[0119] The data decomposition module 203 is used to use the frost optimization algorithm to optimize the variational mode decomposition VMD model to obtain the best parameters, and decompose the data into multiple frequency components based on the optimized VMD results.

[0120] The model building module 204 is used to build a CNN-ASSA-Informer model, train the model using a training set, diagnose the model adaptability and optimize the model hyperparameters using a validation set, and make predictions and evaluate the model using a test set.

[0121] The prediction module 205 is used to construct prediction intervals at different confidence levels based on the results using the quantile regression method.

[0122] This embodiment also provides a wind power prediction device based on multi-model fusion, including: a processor; a memory; and a computer program, wherein the computer program is stored in the memory, and the computer program includes instructions, and when the instructions are executed by the processor, the device executes the method described in the embodiment of the wind power prediction method based on multi-model fusion. For details, please refer to the embodiment of the method of this application, which will not be repeated here.

[0123] Specifically, in this embodiment, the computer program can be divided into one or more modules, and the one or more modules are stored in the memory and executed by the processor to complete the present application. The one or more modules can be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in the device; the processor can be a central processing unit (CPU), or 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. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.; 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 (SMC), a 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.

[0124] The present application also provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps in the wind power prediction method based on multi-model fusion described in the method embodiment of the present application are implemented. For specific contents, please refer to the method embodiment of the present application, which will not be repeated here. The computer-readable storage medium may include but is not limited to a read-only memory (ROM), a random access memory (RAM), a solid-state drive (SSD), an optical disc, a magnetic tape, a flash memory, an optical data storage device, etc. When the computer program in the storage medium is executed by a processor, the prediction steps provided in the present application can be completed.

[0125] In the embodiments of the present application, "at least one" refers to one or more, and "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0126] The above is only a specific implementation of the present application. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A wind power prediction method based on multi-model fusion, characterized in that: include: Collect the operating values ​​and weather forecast data of multiple wind turbines within a preset time period and process abnormal values ​​and missing values ​​in the data; Use the Pearson correlation coefficient to perform feature selection, select feature data that is strongly correlated with the target wind power from the processed data, and establish a training set, a validation set, and a test set; The Frost optimization algorithm is used to optimize the variational mode decomposition VMD model to obtain the best parameters. Based on the optimized VMD results, the data is decomposed into multiple frequency components. A CNN-ASSA-Informer model is established, and the model is trained using a training set, the model adaptability is diagnosed using a validation set, and the model hyperparameters are optimized using a test set. The model is predicted and evaluated using a test set. The CNN-ASSA-Informer model is a convolutional neural network and an adaptive sparse attention mechanism that are connected in series and in parallel with an informer. The sparse self-attention mechanism includes a window self-attention layer and a relative position bias layer. The CNN model is used to extract local features from time series data, and the model's learning ability for key features is enhanced through the adaptive sparse self-attention mechanism. The feature map output by the convolutional neural network is processed by an adaptive sparse self-attention mechanism. The local temporal dependency is effectively modeled through self-attention calculation within the local window. At the same time, the feature relationship within the window is adjusted by using relative position bias to enhance the expressiveness of the feature. Then, the features processed by the sparse self-attention mechanism are added and fused with the global temporal information features output by the informer encoder to generate the final multimodal features. The multimodal features are input into the decoder. The decoder further models the global temporal dependency through the self-attention mechanism, captures the long-term dependency, and obtains the prediction result. Quantile regression methods are used to construct prediction intervals at different confidence levels based on the results.

2. The wind power prediction method based on multi-model fusion according to claim 1 is characterized in that: The collecting of operating values ​​and weather forecast data of multiple wind turbines within a preset time period and processing of abnormal values ​​and missing values ​​in the data includes: Record any actual operating data related to wind power generation at preset intervals within a preset time period, including: power generation, wind direction, wind speed, temperature and humidity of each wind turbine; The feature data with abnormal values ​​and missing values ​​are determined, and the missing values ​​or abnormal values ​​are replaced by the feature means of the corresponding feature data.

3. The wind power prediction method based on multi-model fusion according to claim 1 is characterized in that: The feature selection is performed using the Pearson correlation coefficient to select feature columns that are strongly correlated with the target wind power from the processed data, and to establish a training set, a validation set, and a test set, including: The Pearson correlation coefficient between each feature and the target wind power is calculated, and the feature columns with a strong correlation with the target wind power are selected for modeling to improve the prediction accuracy of the model. The range of the Pearson correlation coefficient is [-1, 1], where 0 represents no correlation, a positive value represents a positive correlation, and a negative value represents a negative correlation. The Pearson correlation coefficient calculation formula is: in, and is the feature sequence, and is a variable, For sequence expectations, For sequence expectations, For variables The variance of For variables The variance of for The mean of for The mean of For sequence The length of the sequence Length.

4. The wind power prediction method based on multi-model fusion according to claim 1 is characterized in that: The frost optimization algorithm is used to optimize the variational mode decomposition VMD model to obtain the best parameters. Based on the optimized VMD results, the data is decomposed into multiple frequency components, including: Construct a variational mode decomposition (VMD) model for wind power prediction; The wind power time series power data is decomposed by variational mode decomposition (VMD) to obtain several frequency components, each of which represents a different frequency component in the signal. In the VMD decomposition process, the frost optimization algorithm is used to optimize the key parameters of VMD, including the number of modes, penalty coefficient and bandwidth, to improve the accuracy of signal decomposition; by simulating the natural cooling process, the parameters in VMD are gradually adjusted to find the optimal mode decomposition parameters; The above process is repeated until all modal components are stable and meet the set convergence conditions, thereby obtaining different frequency characteristics of wind power data so that subsequent models can better capture the time series characteristics and change laws in the data.

5. The wind power prediction method based on multi-model fusion according to claim 4 is characterized in that: The construction of a variational mode decomposition (VMD) model for wind power prediction includes: Construct a variational problem to ensure that the decomposition sequence is a modal component with a finite bandwidth and a center frequency. At the same time, the sum of the estimated bandwidths of each mode is minimized. The constraint condition is that the sum of all modes is equal to the original signal. The VMD constrained variational model is as follows: in, is the initial signal, is the number of decomposed modes, are the components and center frequencies of the decomposed Kth mode, is the Dirac function, is the convolution operation; Solving the variational problem, introducing a quadratic penalty factor and Lagrange multipliers , transforming the minimization problem into an unconstrained optimization problem, the augmented Lagrangian function is as follows: Iterative update using alternating direction multiplier method , find the saddle point of the augmented Lagrangian function in the iterative optimization sequence, and its calculation formula is: In the formula, represents the noise margin, , , , Corresponding to , , , Fourier transform of Define the maximum number of iterations as ,make satisfy , for a given solution accuracy , when the iteration is completed, the following convergence condition is satisfied, and the final ; 。 6. The wind power prediction method based on multi-model fusion according to claim 1 is characterized in that: The convolutional neural network consists of an input layer, a convolutional layer, an activation layer, and an output layer, and the Informer consists of an encoder layer and a decoder layer.

7. The wind power prediction method based on multi-model fusion according to claim 6 is characterized in that: The training set is used to train the model, the validation set is used to diagnose the model adaptability and optimize the model hyperparameters, and the test set is used to make predictions and evaluate the model, including: The data enters the convolution layer through the input layer, is convolved with the convolution kernel in the convolution layer, and the convolution layer outputs the convolution result to the activation layer; The activation layer uses the ReLU activation function to perform nonlinear transformation on the convolution result; after being processed by the convolution layer and the activation layer, the convolutional neural network can extract local temporal features in the input data to form high-level information features.

8. The wind power prediction method based on multi-model fusion according to claim 1 is characterized in that: The quantile regression method is used to construct prediction intervals at different confidence levels based on the results, including: Given a response variable and feature variables , define the conditional distribution function for: For each given ,definition Quantile for: This quantile represents the Next, the response variable of -Quantile; Quantile regression estimates regression models at different quantiles by minimizing the quantile loss function. The definition is as follows: in is the actual observed value, is the predicted value of the model, is the chosen quantile; Then use the quantile regression results to establish the prediction interval, given the input , prediction interval It can be expressed as: 。 9. A wind power prediction system based on multi-model fusion, characterized in that: include: A data acquisition and processing module is used to collect operating values ​​and weather forecast data of multiple wind turbines within a preset time period and process abnormal values ​​and missing values ​​in the data; A feature selection module is used to perform feature selection using the Pearson correlation coefficient, select feature data that is strongly correlated with the target wind power from the processed data, and establish a training set, a validation set, and a test set; The data decomposition module is used to optimize the variational mode decomposition (VMD) model using the Frost optimization algorithm to obtain the best parameters, and decompose the data into multiple frequency components based on the optimized VMD results; A model building module is used to establish a CNN-ASSA-Informer model, train the model using a training set, diagnose the model adaptability and optimize the model hyperparameters using a validation set, and predict and evaluate the model using a test set. The CNN-ASSA-Informer model is a convolutional neural network and an adaptive sparse attention mechanism in series and in parallel with an informer. The sparse self-attention mechanism includes a window self-attention layer and a relative position bias layer. The CNN model is used to extract local features from time series data, and the adaptive sparse self-attention mechanism is used to enhance the model's learning ability for key features. The feature map output by the convolutional neural network is processed by an adaptive sparse self-attention mechanism. The local temporal dependency is effectively modeled through self-attention calculation within the local window. At the same time, the feature relationship within the window is adjusted by using relative position bias to enhance the expressiveness of the feature. Then, the features processed by the sparse self-attention mechanism are added and fused with the global temporal information features output by the informer encoder to generate the final multimodal features. The multimodal features are input into the decoder. The decoder further models the global temporal dependency through the self-attention mechanism, captures the long-term dependency, and obtains the prediction result. The forecasting module is used to construct prediction intervals at different confidence levels based on the results using the quantile regression method.

10. A wind power prediction device based on multi-model fusion, characterized in that: include: processor; Memory; and a computer program, wherein the computer program is stored in the memory, and the computer program comprises instructions, which, when executed by the processor, enable the device to perform the method according to any one of claims 1 to 8.

Images