A Wind Power Prediction Method and Device Based on Deep Learning and Error Correction

Abstract

The present invention provides a wind power prediction method and device based on deep learning and error correction, which relates to the technical fields of power systems and power prediction. The method combines the deep learning GRU model and the attention mechanism to establish an EDA-GRU model for power prediction and power error prediction. By using the volatility characteristics of wind power, VMD and SVMD are respectively used for modal division, and discrete wavelet transform (DWT) is used to capture feature information and integrate it into TimeGAN for data enhancement, which enhances the authenticity of the data and expands the scale of the data, solving the problems of strong volatility of wind power and difficulty in improving prediction accuracy in wind power prediction. Compared with traditional models such as CNN, GRU, LSTM and the combined model CNN-GRU, the prediction accuracy has been greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical fields of power systems and power prediction technologies, and in particular, to a wind power prediction method and device based on deep learning and error correction. Background Art

[0002] Wind power prediction is a key link in ensuring the safe and stable operation of the power grid and optimizing energy dispatching. Existing wind power prediction methods are usually combination methods that combine multiple prediction models. While combining the advantages of each model, they often face the characteristic of relatively high complexity. For error correction, most are based on statistical methods or directly predict the error sequence, and for data augmentation of the error sequence, there is often a lack of strong interpretability, making it difficult to clarify the reasons for the generation of model results or identify key influencing factors. This not only limits the practicability and popularization of the model, but also makes it difficult to effectively utilize or correct prediction errors in power grid operation.

[0003] Traditional power prediction methods mainly focus on using statistical methods, physical methods, and their hybrid methods, such as autoregressive moving average models and exponential smoothing methods. These methods are more suitable for situations with stable conditions and usually have poor effects when facing highly dynamic and non-linear wind energy output. Subsequently emerging machine learning methods, such as decision trees, random forests, and support vector machines, have better flexibility and accuracy for non-linear problems, but still cannot adapt to highly dynamic problems. Deep learning models, especially neural networks and their variants, have improved the adaptability to complex wind power prediction problems due to their superior time series data processing capabilities. Currently, short-term power prediction methods do not fully consider the strong volatility of the power curve and the complexity of error causes, and the models also lack interpretability, making it difficult to accurately predict the change trend of wind power.

[0004] Integrating existing solutions, the following problems still cannot be taken into account:

[0005] (1) The main influencing factor of wind power is wind speed. Due to the non-linearity and non-stationarity of wind speed, it is difficult to utilize the hidden information in its data.

[0006] (2) Deep learning models usually lack interpretability, pay more attention to the prediction results of the models, and it is difficult to reasonably explain the reasons behind the results. By introducing an attention mechanism, it is possible to effectively identify the key time periods or feature variables that play a major role in power prediction, thereby providing important reference value for prediction analysis.

[0007] (3) Commonly used error correction methods are often relatively simple, do not fully consider the change characteristics, formation reasons of errors, and the interpretability of error correction, reducing the correction accuracy.

[0008] The high proportion of renewable energy and high proportion of electronic devices connected to the power grid, and optimizing the operation of each energy source are of great significance for enhancing the resilience of the power grid, optimizing the energy structure, and environmental protection.

[0009] Therefore, how to solve the problems of strong volatility of wind power and difficulty in improving prediction accuracy in wind power prediction has become an urgent problem to be solved. Summary of the Invention

[0010] In order to overcome the deficiencies in the background technology, the present invention provides a wind power prediction method and device based on deep learning and error correction.

[0011] To achieve the above invention objectives, the present invention adopts the following technical solutions:

[0012] In the first aspect, the present invention provides a wind power prediction method based on deep learning and error correction, including the following steps:

[0013] S1. Collect wind power data within a preset time period, and use variational mode decomposition on the wind power data based on the particle swarm optimization algorithm to obtain multiple sub-modal wind power data;

[0014] S2. Divide the sub-modal wind power data into training set data, validation set data, and test set data;

[0015] S3. Input the training set data into a preset EDA-GRU model for training to obtain a power prediction model, and then input the validation set data and test set data into the power prediction model respectively to obtain the validation set power prediction result and the test set power prediction result;

[0016] S4. Calculate the residual as the prediction error according to the validation set power prediction result and the actual wind power value of the validation set data to obtain the validation set prediction error sequence;

[0017] S5. Divide the validation set prediction error sequence into multiple error subsequences through continuous variational mode decomposition, extract wavelet error features using discrete wavelet transform, and use TimeGAN to perform data augmentation on each error subsequence to obtain enhanced error sequences;

[0018] S6. Input the enhanced error sequences into the preset EDA-GRU model in step S3 for training to obtain an error prediction model, and predict the test set prediction error sequence with the same time series as the test set power prediction result through the error prediction model;

[0019] S7. Construct an error correction coefficient through the test set prediction error to correct the test set power prediction result to obtain the final prediction result.

[0020] Specifically, step S1 uses the particle swarm optimization algorithm to partition wind power data using variational mode decomposition as follows: optimizing the key parameters of variational mode decomposition using a combined objective function; the key parameters of the variational mode decomposition include the number of modes, the decomposition bandwidth, and the penalty factor; the combined objective function is shown in formula (1):

[0021] ,

[0022] where MSE is the minimum reconstruction error function; ST is the maximum signal smoothness function; POK is the mode number penalty function, which is used to limit the number of modes K within a reasonable range; is the weight factor for minimizing the reconstruction error, is the weight factor for maximizing the signal smoothness, is the weight factor for the mode number penalty, which is used to control the importance of the MSE function, the ST function, and the POK function in the combined objective function; α is the decomposition bandwidth, and β is the penalty factor.

[0023] Specifically, the preset EDA-GRU model is obtained by establishing an encoder-decoder structure and combining a GRU model with an attention mechanism. The encoder consists of three layers of GRU units, and then the hidden features of the input sequence are extracted, concatenated, and linearly reduced in dimension to obtain a vector as the input of the attention mechanism. Then, the attention mechanism calculates the attention scores and integrates them and passes them to the decoder; the decoder dynamically obtains the context information in combination with the attention mechanism and normalizes and further processes the features, and finally generates the final prediction result through a fully connected layer.

[0024] Specifically, the vector obtained by concatenating and linearly reducing the hidden features of the wind power data is used as the input of the attention mechanism, and then the attention mechanism calculates the attention scores as follows:

[0025] First, the hidden states of multiple layers of GRU are concatenated, and the hidden states of all layers are concatenated into a high-dimensional vector , as shown in formula (9):

[0026] ,

[0027] where, represents the hidden feature of the L-th layer GRU of the encoder at time step t; Concat() is the concatenation function;

[0028] Secondly, a low-dimensional vector is obtained by dimensionality reduction through a linear layer , as shown in formula (10):

[0029] ,

[0030] where, A high-dimensional vector obtained by concatenating the hidden states of all layers at time step t; W is the weight matrix of the linear layer, and b is the bias term; is A low-dimensional vector after linear dimensionality reduction;

[0031] Then, calculate the attention scores , as shown in Equation (11):

[0032] ,

[0033] where is the decay weight based on the time step distance; exp() is the exponential function; i is the time step index of the input sequence; t is the current decoder time step; is the adaptive decay coefficient, which controls the influence of long distances, ; is the hidden feature at decoder time step t and the dot product of the hidden state at the i-th time step of the encoder .

[0034] Specifically, the formula for calculating the adaptive decay coefficient is as shown in Equation (12):

[0035] ,

[0036] where is the weight matrix; is the bias vector, which improves the dynamics; the softplus function outputs non-negative values to ensure , meeting the requirements of the decay coefficient, as shown in Equation (13):

[0037] .

[0038] Specifically, the decoder combines the attention mechanism to dynamically obtain context information and normalizes and further processes the features, and finally generates the final prediction result through the fully connected layer, specifically:

[0039] First, normalize the attention scores of all time steps to obtain the attention weights , as shown in Equation (14):

[0040] ,

[0041] where T is the length of the input sequence;

[0042] Based on the attention weights , calculate the context vector , as shown in Equation (15):

[0043] ,

[0044] Among them, is the hidden state of the encoder at time step t;

[0045] The context vector will be used as the input of the decoder to generate the final prediction result.

[0046] Specifically, step S5 specifically includes the following steps:

[0047] S51. Divide the verification set prediction error sequence into multiple error subsequences through continuous variational mode decomposition;

[0048] S52. Normalize each error subsequence to obtain a normalized error subsequence, then use discrete wavelet transform to decompose the normalized error subsequence into low-frequency components and high-frequency components, and obtain the wavelet error features of each error subsequence according to the low-frequency components and high-frequency components;

[0049] S53. Generate similar error features of the wavelet error features through TimeGAN;

[0050] S54. Restore the similar error features through inverse wavelet transform to obtain a normalized enhanced error sequence;

[0051] S55. Denormalize the normalized enhanced error sequence to obtain an enhanced error sequence, and finally obtain the enhanced error sequence of each error subsequence.

[0052] Specifically, step S7 is specifically: correct the test set power prediction result through formula (23):

[0053] ,

[0054] Among them, is the corrected predicted value corresponding to time point i, with the unit of kilowatt; is the preliminary predicted wind power value corresponding to time point i, with the unit of kilowatt; is the error correction coefficient corresponding to time point i, The value range of is (0, 1);

[0055] The error correction coefficient is composed of a correction coefficient and a bias coefficient, as shown in formula (24):

[0056] ,

[0057] Among them, is the error correction coefficient corresponding to time point i; is the correction coefficient corresponding to time point i; is the bias coefficient corresponding to time point i;

[0058] Correction coefficient Specifically, it is shown in formula (25) as follows:

[0059] ,

[0060] wherein, represents the actual wind power value at time point i; represents the preliminarily predicted wind power value at time point i; represents the prediction error of the test set at time point i;

[0061] Bias coefficient Specifically, it is shown in formula (26) as follows:

[0062] ,

[0063] wherein, the gust wind speed at time point i, unit: m / s, is the wind speed at time point i, unit: m / s.

[0064] Specifically, the wind power data includes the hourly wind power generation, unit: kW; the wind speed, unit: m / s; the gust wind speed, unit: m / s; the preset time period includes a first preset time period and a second preset time period; the first 80% of the sub-modal wind power data in the first preset time period is used as the training set data in chronological order, and the last 20% is used as the validation set data, and the sub-modal wind power data in the second preset time period is used as the test set data.

[0065] In a second aspect, the present invention provides a wind power prediction device based on deep learning and error correction, including the following units:

[0066] The decomposition and optimization unit is used to collect the wind power data within a preset time period, and divide the wind power data into multiple sub-modal wind power data by using variational mode decomposition based on the particle swarm optimization algorithm;

[0067] The data division unit is used to divide the sub-modal wind power data into training set data, validation set data and test set data;

[0068] The first prediction unit is used to input the training set data into a preset EDA-GRU model for training to obtain a power prediction model, and then input the validation set data and the test set data into the power prediction model respectively to obtain the validation set power prediction result and the test set power prediction result;

[0069] An error calculation unit, configured to calculate a residual as a prediction error based on the verification set power prediction result and the actual wind power value of the verification set data, and obtain a verification set prediction error sequence;

[0070] An error enhancement unit, configured to divide the verification set prediction error sequence into multiple error subsequences through continuous variational mode decomposition, extract wavelet error features using discrete wavelet transform, and perform data augmentation on each error subsequence using TimeGAN to obtain enhanced error sequences;

[0071] A second prediction unit, configured to input the enhanced error sequences into a preset EDA-GRU model for training to obtain an error prediction model, and predict a test set prediction error sequence with the same time series as the test set power prediction result through the error prediction model;

[0072] A prediction correction unit, configured to construct an error correction coefficient through the test set prediction error to correct the test set power prediction result to obtain a final prediction result.

[0073] In a third aspect, the present invention further provides an electronic device, including a processor, a memory, a communication interface, and one or more programs, where the one or more programs are stored in the memory and are configured to be executed by the processor, and the programs include instructions for performing the steps in the method described in the first aspect.

[0074] The present invention provides a wind power prediction method and device based on deep learning and error correction. The method combines the deep learning GRU model and the attention mechanism to establish an EDA-GRU model for power prediction and power error prediction. The VMD and SVMD are respectively used for modal division based on the volatility characteristics of wind power, and the discrete wavelet transform (DWT) is used to capture feature information and integrate it into TimeGAN for data augmentation, enhancing the authenticity of the data and expanding the scale of the data, solving the problems of strong volatility of wind power and difficulty in improving prediction accuracy in wind power prediction. Compared with traditional models such as CNN, GRU, LSTM, and the combined model CNN-GRU, the prediction accuracy has been greatly improved. Description of the Drawings

[0075] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0076] Figure 1Schematic diagram of a wind power prediction method based on deep learning and error correction according to an embodiment of the present invention;

[0077] Figure 2 Schematic diagram of the EDA-GRU model structure diagram according to an embodiment of the present invention;

[0078] Figure 3 Schematic diagram of the GRU model framework used according to an embodiment of the present invention;

[0079] Figure 4 Schematic diagram of the TimeGAN data augmentation framework combined with wavelet transform according to an embodiment of the present invention;

[0080] Figure 5 Schematic diagram of a wind power prediction device based on deep learning and error correction according to an embodiment of the present invention;

[0081] Figure 6 Schematic diagram of a wind power prediction device based on deep learning and error correction according to an embodiment of the present invention. Detailed implementation manners

[0082] The present invention can be explained in detail through the following embodiments. The purpose of providing the present invention is to protect all technical improvements within the scope of the present invention. In the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "front", "rear", "left", "right", etc. indicating the orientation or positional relationship, they are only corresponding to the drawings of the present application for the convenience of describing the present invention, rather than indicating or implying that the device or element referred to must have a specific orientation.

[0083] Embodiment 1

[0084] Refer to Figure 1 , this embodiment provides a wind power prediction method based on deep learning and error correction, including the following steps:

[0085] S1. Collect wind power data within a preset time period, and use variational mode decomposition to divide the wind power data based on the particle swarm optimization algorithm to obtain multiple sub-modal wind power data;

[0086] The wind power data includes the wind power generation per hour, unit: kilowatt; wind speed, unit: meter per second; gust wind speed, unit: meter per second; the preset time period includes January 1, 2013 to December 31, 2016;

[0087] The preset time period includes a first preset time period and a second preset time period; the first preset time period is from January 1, 2013 to December 31, 2015; the second preset time period is from January 1, 2016 to December 31, 2016;

[0088] The collected data includes: hourly wind power generation (unit: kilowatt), wind speed (unit: m / s), gust wind speed (unit: m / s), which refers to the wind speed of the wind that suddenly increases in a short period of time.

[0089] Equipment: data acquisition system (including wind speed sensor, generator power monitoring equipment); data storage tool (saved as a CSV file); data cleaning tool (Pandas library).

[0090] Specifically: the key parameters of variational mode decomposition are optimized by constructing a combined objective function through the particle swarm optimization algorithm, and then the optimized variational mode decomposition is used to divide the wind power data to obtain multiple sub-modes; the key parameters of the variational mode decomposition include the number of modes, decomposition bandwidth and penalty factor;

[0091] Use particle swarm optimization (PSO) to find the optimal parameters of variational mode decomposition, and then use the variational mode decomposition (VMD) optimized by PSO to divide the wind power data to obtain multiple sub-mode wind power data;

[0092] Variational mode decomposition (VMD) is an adaptive signal decomposition method that decomposes the original signal into several intrinsic mode functions (IMFs), that is, sub-modes. Each sub-mode represents different frequency components of the signal. The advantage of VMD is that it can retain more local features in signal decomposition and can effectively process non-stationary and non-linear signals.

[0093] Particle swarm optimization (PSO) is a swarm intelligence optimization algorithm. Based on the behavior model of bird flocks foraging in nature, it searches for the optimal solution of the problem by simulating the movement of "particles" in the search space.

[0094] In this embodiment, the PSO algorithm is used to optimize the key parameters in VMD: the number of modes, decomposition bandwidth, penalty factor.

[0095] In this embodiment, the PSO algorithm used defines a combined objective function to optimize the key parameters in VMD, and the combined objective function is shown in formula (1):

[0096] ,

[0097] Among them, MSE is the minimum reconstruction error function; ST is the maximum signal smoothness function; POK is the mode number penalty function, which is used to limit the mode number K within a reasonable range; is the minimum reconstruction error weight factor, is the maximum signal smoothness weight factor, is the modal quantity penalty weight factor, used to control the importance of the MSE function, ST function, and POK function in the combined objective function; α is the decomposition bandwidth, and β is the penalty factor.

[0098] Preferably, , , The values of depend on the importance of the three factors of modal quantity, decomposition bandwidth, and penalty factor. However, all three of them should be between 0 and 1. In this embodiment, , , take the values of 0.4, 0.4, and 0.2 respectively;

[0099] Modal quantity (K): The modal quantity selected during VMD decomposition affects the fine-grainedness of the result. If the modal quantity is too large, overfitting may occur; if the modal quantity is too small, the characteristics of the signal may not be effectively captured. PSO can find the optimal decomposition structure by adjusting the modal quantity; its value is an integer selected in the range [2, 10]. Preferably, in this embodiment, K takes 6;

[0100] Decomposition bandwidth (α): This parameter controls the bandwidth size of each mode, thus affecting the frequency distribution of the modes. PSO can optimize the decomposition effect by adjusting the bandwidth to ensure that the modes can reasonably represent the frequency components of the signal; its value is a real number selected in the range [0, 10]. Preferably, in this embodiment, α takes 5.

[0101] Penalty factor (β): This parameter affects the decomposition accuracy of the modes. PSO can optimize this parameter to make the decomposed modes smoother and without too much high-frequency noise. Its value is a real number selected in the range [0, 10]. Preferably, in this embodiment, β takes 5.

[0102] Through PSO to optimize VMD, finally the initial data is divided into multiple sub-modes.

[0103] The combined objective function includes the following parts:

[0104] A. Minimizing the reconstruction error (MSE): The mean square error between the reconstructed signal and the original signal, reflecting the quality of the decomposition,

[0105] B. Maximizing the smoothness of the signal (ST): By adjusting the values of α and β, optimizing the frequency difference between the modes to make the signal smoother and with less noise.

[0106] C. Balancing the modal quantity (K): An appropriate value of K helps to balance the fine-grainedness of the decomposition. Too many modes may lead to overfitting, and too few may not be able to capture the details of the signal.

[0107] In this embodiment, the three partial functions of the objective function are reconstructed as follows:

[0108] a) Minimize the reconstruction error (MSE): The mean square error is used to represent the difference between the reconstructed signal and the original signal, as shown in Equation (2):

[0109] ,

[0110] where is the value of the original signal at time , is the reconstructed value of the signal obtained by variational mode decomposition (VMD) at time , N is the number of samples, α is the decomposition bandwidth, and β is the penalty factor.

[0111] The MSE function itself is a simple error metric that directly calculates the error between the original signal and the reconstructed signal. It does not need to know the specific parameters of VMD. The adjustment of the VMD parameters affects factors such as the accuracy and smoothness of signal decomposition, indirectly determines the quality of the reconstructed signal, and thus affects the value of MSE. The expressions in the brackets , , indicate that MSE is related to these three parameters and are not the inputs for calculating MSE.

[0112] b) Maximize the smoothness of the signal (ST): It is achieved by calculating the difference between modal spectra, as shown in Equation (3):

[0113] ,

[0114] where is the central frequency of the nth mode, is the central frequency of the (n + 1)th mode, K is the number of modes, and a smaller frequency difference means higher smoothness.

[0115] The SK function is similar to the MSE function. The expressions in the brackets , indicate that SK is related to these three parameters and are not the inputs for calculating SK.

[0116] The smoothness calculates the difference between adjacent modal frequencies. If the modal frequency difference is large, the smoothness will be low, indicating that the signal decomposition is not smooth enough. By optimizing K, α, and β, it is desired to maximize the smoothness, that is, to make the frequency variation of the modes as small as possible.

[0117] c) Balance the number of modes (K): This is a penalty term used to control the number of modes K and avoid selecting too many or too few modes. Too many modes may lead to overfitting, while too few modes may lead to insufficient decomposition. Therefore, a penalty mechanism is needed to ensure the rationality of the number of modes.

[0118] ,

[0119] Where: and are the reasonable ranges of the number of modes, which can usually be set according to the characteristics of the signal. For example = 2 and = 10.

[0120] If the number of modes K exceeds the predetermined range , the penalty term will become infinite, indicating that this is an unacceptable solution. Therefore, through the optimization process, PSO will avoid selecting an unreasonable number of modes.

[0121] The PSO algorithm is to calculate the final within the given through different combinations, and select the when the maximum as the parameters of the model.

[0122] By optimizing VMD with PSO, the initial data (wind power data) will finally be divided into multiple sub - modes.

[0123] S2. Divide the sub - mode wind power data into training set data, validation set data, and test set data;

[0124] Take the first 80% of the sub - mode wind power data in chronological order within the first preset time period as the training set data, and the last 20% as the validation set data, and take the sub - mode wind power data within the second preset time period as the test set data;

[0125] Among them, the hourly wind power generation power is used for single - feature prediction to construct a two - dimensional input matrix composed of the time series, and the wind speed and gust wind speed are used for the bias coefficient modeling in step S7.

[0126] The training set is used for model training; the validation set is used to calculate the prediction error sequence and predict the prediction error of the test set; the test set is used to finally evaluate the generalization ability of the model and measure the performance of the model in the real scenario.

[0127] S3. Input the training set data into the preset EDA - GRU model for training to obtain a power prediction model, and then input the validation set data and test set data into the power prediction model respectively to obtain the validation set power prediction result and the test set power prediction result;

[0128] The preset EDA-GRU model is obtained by establishing an encoder-decoder structure and combining the GRU model with an attention mechanism;

[0129] Specifically, the preset EDA-GRU model adopts a deep learning network structure as shown in Figure 2 . The input encoder consists of three GRU units, which enhances the feature extraction ability and reduces the influence of noise. Then, the hidden features of the input sequence are spliced and linearly reduced to obtain a vector as the input of the attention mechanism. Then, the attention mechanism calculates the attention scores and integrates them and passes them to the decoder; the decoder dynamically obtains context information in combination with the attention mechanism and normalizes and further processes the features, and finally generates a prediction result through a fully connected layer.

[0130] In this embodiment, the training set data obtained in step S2 is input into the preset EDA-GRU model for training to obtain a power prediction model;

[0131] The validation set data is input into the power prediction model to obtain the validation set power prediction result, and the test set data is input into the power prediction model to obtain the test set power prediction result;

[0132] It can be understood that each sub-modal obtained by VMD decomposition of the original data represents different frequency components of the original data. The number of data of this sub-modal is the same as that of the original data, and adding the sub-modal obtained by VMD decomposition can accurately reconstruct the original signal. This is because one of the goals of VMD is to ensure the orthogonality and completeness of signal decomposition.

[0133] It can be seen from steps S1 and S2 that the validation set data contains multiple sub-modal wind power data, and the time series of each sub-modal wind power data is the same. Therefore, after inputting the validation set data into the power prediction model for prediction, the power prediction result data of the multiple sub-modal wind power data are added to finally obtain the validation set power prediction result;

[0134] Similarly, the test set data also contains multiple sub-modal wind power data, and the time series of each sub-modal wind power data is the same. Therefore, after inputting the test set data into the power prediction model for prediction, the power prediction result data of the multiple sub-modal wind power data are added to finally obtain the test set power prediction result;

[0135] Both the validation set power prediction result and the test set power prediction result contain a two-dimensional matrix sequence of time points and the corresponding predicted power results at the time points.

[0136] GRU (Gated Recurrent Unit) is an improved recurrent neural network model designed specifically for processing time series data. By introducing update gates and reset gates, it effectively solves the vanishing gradient problem of traditional RNNs and can capture long-term and short-term dependencies in time series. Its specific structure is as shown in Figure 3 shown, and its specific internal calculation process is as shown in formulas (5)-(8):

[0137] ,

[0138] where, , , is the weight matrix corresponding to the network activation function, is the weight matrix of the update gate; is the weight matrix of the reset gate, is the weight matrix of the candidate hidden state; is the candidate matrix, which contains all vectors that may be the output of the GRU layer, and are the outputs of the internal reset gate and update gate of the GRU respectively, σ and ɸ are the sigmoid and tanh activation functions respectively, is the hidden state at the previous time step t, is the hidden state at the current time step t, is the input at the current time step t, is the candidate hidden state at the current time step t.

[0139] The attention layer judges the importance of the information contained in the predicted input from the GRU and assigns different weights. The weight calculation process is as follows:

[0140] First, concatenate the hidden states of multiple layers of GRUs, and concatenate the hidden states of all layers into a high-dimensional vector , as shown in formula (9):

[0141] ,

[0142] where, represents the hidden feature of the GRU in the L-th layer of the encoder at time step t; Concat() is the concatenation function;

[0143] Secondly, reduce the dimension through a linear layer to obtain a low-dimensional vector , as shown in formula (10):

[0144] ,

[0145] where, A high-dimensional vector obtained by concatenating the hidden states of all layers at time step t; W is the weight matrix of the linear layer, and b is the bias term; is a low-dimensional vector after linear dimensionality reduction;

[0146] Then, calculate the attention scores , as shown in Equation (11):

[0147] ,

[0148] where calculating the attention scores consists of two parts:

[0149] The first part: is the decay weight based on the time-step distance:

[0150] where exp() is the exponential function; i is the time-step index of the input sequence; t is the current decoder time step;

[0151] is the adaptive decay coefficient, controlling the influence of long distances:

[0152] A larger λ: the weights of distant time steps decay faster, and the model pays more attention to neighboring time steps;

[0153] A smaller λ: distant time steps still have a greater influence, and the model has a wider attention range;

[0154] This part introduces the proximity assumption of time steps, that is, short-term dependencies in the time series are more important;

[0155] The adaptive decay coefficient is calculated as shown in Equation (12):

[0156] ,

[0157] where is the hidden state of the decoder at the current time step t, containing the context information of time step t; is the weight matrix; is the bias vector, enhancing the dynamics; the softplus function outputs non-negative values to ensure , meeting the requirements of the decay coefficient, as shown in Equation (13):

[0158] ;

[0159] The second part: is the hidden feature of the decoder at time step t dot product (measuring similarity) with the hidden state of the encoder at the i-th time step ;

[0160] Use Softmax to normalize the attention scores at all time steps to obtain attention weights as shown in Equation (14):

[0161] ,

[0162] where T is the length of the input sequence;

[0163] Based on the attention weights , calculate the context vector as shown in Equation (15):

[0164] ,

[0165] The context vector will be used as the input of the decoder to generate the final prediction result;

[0166] The input at each time step is only the output of the previous time step and the hidden state of the decoder ;

[0167] After introducing the context vector, the input of GRU will be extended as shown in Equation (16):

[0168] ,

[0169] where is the output generated by the decoder at the previous time step; is the context vector at the current time step (from the attention mechanism);

[0170] In this embodiment, variational mode decomposition (VMD) based on particle swarm optimization (PSO) is used to decompose the data into multiple modes, and an EDA-GRU model is constructed based on TensorFlow.

[0171] S4. Calculate the residual as the prediction error according to the power prediction result of the validation set and the actual wind power value of the validation set data to obtain the validation set prediction error sequence; the calculation formula of the residual is as shown in Equation (17):

[0172] ,

[0173] where: is the prediction error at time i; is the actual wind power value at time i; is the preliminary predicted wind power value at time i;

[0174] It is understandable that the actual wind power values at different times are subtracted from the predicted wind power values at the corresponding times to obtain the verification set prediction error sequence;

[0175] S5. The verification set prediction error sequence is divided into multiple error subsequences through Sequential Variational Mode Decomposition (SVMD). Discrete wavelet transform is used to extract wavelet error features, and TimeGAN is used to perform data augmentation on each error subsequence to obtain enhanced error sequences;

[0176] To enhance the authenticity of the data, this method uses Discrete Wavelet Transform (DWT) to extract the features of the real error data, that is, the features of the verification set prediction error sequence, so that the error data generated by TimeGAN is more in line with the real distribution. The data augmentation process is as Figure 4 shown.

[0177] S51. The verification set prediction error sequence is divided into multiple error subsequences through Sequential Variational Mode Decomposition (SVMD);

[0178] Sequential Variational Mode Decomposition (SVMD) is a signal processing and data analysis method. It is mainly used to decompose complex signals into a series of mode functions, and each mode function represents a specific frequency component in the signal.

[0179] The basic principle of SVMD is to decompose the signal into multiple mode functions through variational mode decomposition. In each iteration step, SVMD updates the mode functions by minimizing the difference between the signal and the mode functions, and this process will be repeated continuously until convergence. The key feature of SVMD is sequential decomposition, that is, in each iteration step, a main frequency component is extracted from the signal and removed from the signal until all frequency components are extracted.

[0180] In this embodiment, the prediction error sequence will obtain multiple error subsequences (mode functions) after Sequential Variational Mode Decomposition (SVMD);

[0181] S52. Each error subsequence is normalized to obtain a normalized error subsequence, and then the discrete wavelet transform is used to decompose the normalized error subsequence into low-frequency components and high-frequency components, and the wavelet error features of each error subsequence are obtained according to the low-frequency components and high-frequency components;

[0182] Among them, the normalization of each error subsequence is as shown in formula (18):

[0183] ,

[0184] Among them, X is the error data in the error subsequence (which may contain values in different ranges), is the minimum value in the error subsequence, is the maximum value in the error subsequence, is the normalized error data;

[0185] Among them, each normalized error subsequence is decomposed using the discrete wavelet transform (DWT) as shown in formula (19):

[0186] ,

[0187] where, is the value of the m-th sample of the validation set error sequence at time step t; is the low-frequency component, reflecting the overall trend of the error; is the high-frequency component of the j-th layer, reflecting the detailed changes of the error; j is the number of layers;

[0188] Calculate the wavelet error features as shown in formula (20):

[0189] ,

[0190] where, is the low-frequency component, is the high-frequency component of the j-th layer, reflecting the detailed changes of the error; j is the number of layers;

[0191] S53. Generate similar error features of the wavelet error features through TimeGAN;

[0192] TimeGAN is a deep learning model for generating time series data. It combines the ideas of generative adversarial networks (GAN) and self-supervised learning, and can generate high-quality data with real time series characteristics. In the generated time series data, TimeGAN preserves the time dependence and the complex relationships between features.

[0193] Due to pattern generation, the training objective of TimeGAN is to generate similar error features with a distribution similar to the input features .

[0194] S54. Restore the similar error features through the inverse wavelet transform to obtain the normalized enhanced error sequence;

[0195] Therefore, the inverse wavelet transform (IDWT) is used to restore the generated similar error features to the normalized enhanced error sequence. The inverse wavelet transform is as shown in formula (21):

[0196] ,

[0197] wherein, is the value of the m-th sample of the original error sequence after the inverse wavelet transform at time step t; The value of the m-th sample of the original error sequence after the inverse wavelet transform at time step t; is the low-frequency component after the inverse wavelet transform, reflecting the overall trend of the error; is the high-frequency component of the j-th layer after the inverse wavelet transform, reflecting the detailed changes of the error;

[0198] S55. Denormalize the normalized enhanced error sequence to obtain the enhanced error sequence, and finally obtain the enhanced error sequence of each error subsequence;

[0199] Denormalize the normalized enhanced error sequence to obtain the generated error data, that is, the enhanced error sequence, and then obtain the enhanced error sequence of each error subsequence.

[0200] The denormalization formula is as shown in formula (22):

[0201] ,

[0202] wherein, is the error data obtained after normalization processing (i.e., the data generated by the model), is the minimum value in the error subsequence, is the maximum value in the error subsequence, is the error data after denormalization, that is, the data restored to the original data range.

[0203] In this embodiment, the residual sequence is decomposed by using continuous variational mode decomposition (SVMD), wavelet feature extraction is performed by using the PyWavelets library, and then the data is augmented based on TimeGAN to twice the original data, and the tool used is PyCharm.

[0204] S6. Input the enhanced error sequence into the pre-set EDA-GRU model in step S3 for training to obtain an error prediction model, and predict the test set prediction error sequence of the same time series as the test set power prediction result through the error prediction model;

[0205] Specifically, the error values at the same time point in multiple enhanced error sequences are averaged to reconstruct a new enhanced error sequence, and the new enhanced error sequence and its corresponding time point are constructed into a matrix as the input of the EDA-GRU model to train the error prediction model, and the error prediction model can predict and output the power error values at different time points.

[0206] The enhanced error sequence is composed of the original error sequence and the error sequence expanded by TimeGAN. The time series of the enhanced error sequence of each error subsequence is the same. After being expanded by TimeGAN, multiple error values correspond to one time point. The average value of the multiple error values is taken as the error at that time point, and then a matrix constructed by this error and its corresponding time point is used as the input of the preset EDA-GRU model to train an error prediction model; the predicted error at the corresponding time point can be obtained by inputting the time point into the error prediction model.

[0207] In step S3, the power prediction result of the test set is obtained as a two-dimensional matrix sequence containing the time points and the predicted power results corresponding to the time points. The time series of the power prediction result of the test set is extracted and input into the power prediction model to obtain the test set prediction error sequence.

[0208] In step S6, the structure adopted by the preset EDA-GRU model is the same as that of the EDA-GRU model adopted in step S3;

[0209] S7. The error correction coefficient is constructed through the test set prediction error to correct the power prediction result of the test set to obtain the final prediction result.

[0210] Specifically, the power prediction result of the test set is corrected by formula (23):

[0211] ,

[0212] where, is the corrected predicted value (kilowatt) corresponding to time point i; is the preliminarily predicted wind power value (kilowatt) corresponding to time point i; is the error correction coefficient corresponding to time point i, The value range of is (0, 1);

[0213] In this embodiment, the error correction coefficient is used to correct the prediction result. The error correction coefficient is composed of a correction coefficient and a bias coefficient, as shown in formula (24):

[0214] ,

[0215] where, is the error correction coefficient corresponding to time point i; is the correction coefficient corresponding to time point i; is the bias coefficient corresponding to time point i;

[0216] The correction coefficient Specifically, it is shown in formula (25):

[0217] ,

[0218] wherein, represents the actual wind power value at time point i; represents the preliminarily predicted wind power value at time point i; represents the prediction error of the test set at time point i;

[0219] bias coefficient Specifically, it is shown in formula (26):

[0220] ,

[0221] wherein, the gust wind speed at time point i, with the unit of m / s, is the wind speed at time point i, with the unit of m / s;

[0222] Traditional power prediction methods are shown in Table 1:

[0223]

[0224] This embodiment provides a wind power prediction method based on deep learning and error correction. Compared with traditional CNN, GRU, LSTM and their combined model CNN-GRU, this method can significantly improve the accuracy of wind power prediction. The key to this improvement lies in several aspects: First, the model combines deep learning and an error correction mechanism. By correcting the prediction results for errors, it can effectively correct the prediction biases generated in traditional methods, thereby improving accuracy. Second, the combination of the GRU model and the attention mechanism enables the model to adaptively focus on important temporal features, thus enhancing the sensitivity to key fluctuations in wind power data. In addition, VMD and SVMD are used for modal separation, effectively extracting useful information in each frequency band from complex wind power signals, reducing noise interference, and further improving the prediction accuracy. Furthermore, discrete wavelet transform (DWT) is used for feature extraction, enhancing the model's ability to capture different frequency components in the data. Finally, by using TimeGAN for data augmentation, the model can utilize more high-quality data for training, which not only improves the model's generalization ability but also makes it more stable in the face of the uncertainty and volatility of wind power data.

[0225] This embodiment provides a wind power prediction method and device based on deep learning and error correction. The method integrates the deep learning GRU model and the attention mechanism to establish an EDA-GRU model for power prediction and power error prediction. The variational mode decomposition (VMD) and the split variational mode decomposition (SVMD) are used for mode division respectively according to the volatility characteristics of wind power. The discrete wavelet transform (DWT) is used to capture feature information and integrate it into TimeGAN for data augmentation, which enhances the authenticity of the data and expands the scale of the data, and solves the problems of strong volatility of wind power and difficulty in improving prediction accuracy in wind power prediction. Compared with the traditional models CNN, GRU, LSTM and the combined model CNN-GRU, the prediction accuracy has been greatly improved.

[0226] Embodiment 2

[0227] Reference Figure 5 , this embodiment provides a wind power prediction device based on deep learning and error correction, which includes the following units:

[0228] The decomposition and optimization unit is used to collect wind power data within a preset time period, and use variational mode decomposition for the wind power data based on the particle swarm optimization algorithm to obtain multiple sub-mode wind power data;

[0229] The data division unit is used to divide the sub-mode wind power data into training set data, validation set data and test set data;

[0230] The first prediction unit is used to input the training set data into a preset EDA-GRU model for training to obtain a power prediction model, and then input the validation set data and the test set data into the power prediction model respectively to obtain the validation set power prediction result and the test set power prediction result;

[0231] The error calculation unit is used to calculate the residual as the prediction error according to the validation set power prediction result and the actual wind power value of the validation set data, and obtain the validation set prediction error sequence;

[0232] The error enhancement unit is used to divide the validation set prediction error sequence into multiple error subsequences through continuous variational mode decomposition, extract wavelet error features using discrete wavelet transform, and use TimeGAN to perform data augmentation on each error subsequence to obtain enhanced error sequences;

[0233] The second prediction unit is used to input the enhanced error sequences into a preset EDA-GRU model for training to obtain an error prediction model, and predict the test set prediction error sequence with the same time series as the test set power prediction result through the error prediction model;

[0234] A prediction correction unit is configured to construct an error correction coefficient through the prediction error of the test set to correct the test set power prediction result and obtain the final prediction result

[0235] Embodiment III

[0236] Reference Figure 6 , Figure 6 FIG. 10 is a schematic structural diagram of a wind power prediction device based on deep learning and error correction according to this embodiment. The wind power prediction device 20 based on deep learning and error correction in this embodiment includes a processor 21, a memory 22, and a computer program stored in the memory 22 and executable on the processor 21. When the processor 21 executes the computer program, the steps in the above method embodiments are implemented. Alternatively, when the processor 21 executes the computer program, the functions of each module / unit in the above device embodiments are implemented

[0237] Exemplarily, the computer program may be divided into one or more modules / units. The one or more modules / units are stored in the memory 22 and executed by the processor 21 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the wind power prediction device 20 based on deep learning and error correction. For example, the computer program may be divided into the respective modules in Embodiment II. For the specific functions of each module, please refer to the working process of the device described in the above embodiments, and details are not described herein again

[0238] The wind power prediction device 20 based on deep learning and error correction may include, but is not limited to, a processor 21 and a memory 22. Those skilled in the art can understand that the schematic diagram is only an example of the wind power prediction device 20 based on deep learning and error correction, and does not constitute a limitation on the wind power prediction device 20 based on deep learning and error correction. It may include more or fewer components than shown in the figure, or combine some components, or different components. For example, the wind power prediction device 20 based on deep learning and error correction may further include an input / output device, a network access device, a bus, etc

[0239] The processor 21 can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor 21 is the control center of the wind power prediction device 20 based on deep learning and error correction, and connects all parts of the wind power prediction device 20 based on deep learning and error correction through various interfaces and lines.

[0240] The memory 22 can be used to store the computer programs and / or modules. The processor 21 realizes various functions of the wind power prediction device 20 based on deep learning and error correction by running or executing the computer programs and / or modules stored in the memory 22, and by calling the data stored in the memory 22. The memory 22 mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory 22 can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, at least one magnetic disk storage device, flash device, or other volatile solid-state storage devices.

[0241] Among them, if the modules / units integrated in the wind power prediction device 20 based on deep learning and error correction are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor 21, the steps of the above various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0242] It should be noted that the device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0243] The parts not detailed in the present invention are prior art. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention; therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, aiming to include all changes falling within the meaning and scope of the equivalent elements within the present invention.

Claims

1. A wind power prediction method based on deep learning and error correction, characterized in that Specifically, it includes the following steps: S1. Collect wind power data within a preset time period, and use variational mode decomposition on the wind power data based on the particle swarm optimization algorithm to obtain multiple sub-modal wind power data; S2. Divide the sub-modal wind power data into training set data, validation set data, and test set data; S3. Input the training set data into a preset EDA-GRU model for training to obtain a power prediction model, and then input the validation set data and test set data into the power prediction model respectively to obtain the validation set power prediction result and the test set power prediction result; S4. Calculate the residual as the prediction error according to the validation set power prediction result and the actual wind power value of the validation set data to obtain the validation set prediction error sequence; S5. Divide the validation set prediction error sequence into multiple error subsequences through continuous variational mode decomposition, extract wavelet error features using discrete wavelet transform, and use TimeGAN to perform data augmentation on each error subsequence to obtain enhanced error sequences; S6. Input the enhanced error sequences into the preset EDA-GRU model in step S3 for training to obtain an error prediction model, and predict the test set prediction error sequence with the same time series as the test set power prediction result through the error prediction model; S7. Construct an error correction coefficient through the test set prediction error to correct the test set power prediction result to obtain the final prediction result.

2. The wind power prediction method according to claim 1, characterized in that The specific process of using variational mode decomposition to divide the wind power data based on the particle swarm optimization algorithm in step S1 is as follows: use a combined objective function to optimize the key parameters of variational mode decomposition; the key parameters of variational mode decomposition include the number of modes, decomposition bandwidth, and penalty factor; the combined objective function is shown in formula (1): ， Among them, MSE is the minimum reconstruction error function; ST is the maximum signal smoothness function; POK is the modal number penalty function, which is used to limit the modal number K within a reasonable range; is the weight factor for minimizing the reconstruction error, is the weight factor for maximizing the signal smoothness, is the weight factor of the modal number penalty, which is used to control the importance of the MSE function, the ST function and the POK function in the combined objective function; α is the decomposition bandwidth, and β is the penalty factor.

3. The wind power prediction method according to claim 1, wherein The preset EDA-GRU model is obtained by establishing an encoder-decoder structure and combining a GRU model with an attention mechanism. The encoder consists of three layers of GRU units, and then the vector obtained by splicing and linearly reducing the hidden features of the input sequence is used as the input of the attention mechanism. Then the attention mechanism calculates the attention score and integrates it and passes it to the decoder; The decoder dynamically obtains context information in combination with the attention mechanism, normalizes and further processes the features, and finally generates the final prediction result through a fully connected layer.

4. The wind power prediction method according to claim 3, wherein The specific process of using the vector obtained by splicing and linearly reducing the hidden features of the wind power data as the input of the attention mechanism, and then the attention mechanism calculates the attention score is as follows: First, concatenate the hidden states of the multi-layer GRU, and concatenate the hidden states of all layers into a high-dimensional vector , as shown in formula (9): ， Among them, represents the hidden feature of the GRU in the L-th layer of the encoder at time step t; Concat() is the concatenation function; Secondly, a low-dimensional vector is obtained by dimensionality reduction through a linear layer , as shown in formula (10): ， Among them, is the high-dimensional vector spliced by the hidden states of all layers at time step t; W is the weight matrix of the linear layer, and b is the bias term. is the low-dimensional vector after linear dimensionality reduction; Then, calculate the attention scores , as shown in Equation (11): ， Among them, is the decay weight based on the time-step distance; exp() is the exponential function; i is the time-step index of the input sequence; t is the current decoder time step; is the adaptive decay coefficient, controlling the influence of long distances ; is the hidden feature at decoder time step t and the dot product of the hidden state at the i-th time step of the encoder .

5. The wind power prediction method according to claim 4, wherein The adaptive attenuation coefficient is calculated by the formula shown in Formula (12) as follows: ， Among them, is the weight matrix; is the bias vector, which improves the dynamic performance; the softplus function outputs non-negative values to ensure that , meeting the requirements of the attenuation coefficient, as shown in Equation (13): 。 6. The wind power prediction method according to claim 4, wherein The specific process of the decoder dynamically obtaining context information in combination with the attention mechanism, normalizing and further processing the features, and finally generating the final prediction result through a fully connected layer is as follows: First, normalize the attention scores for all time steps to obtain the attention weights , as shown in Equation (14): ， Where, T is the length of the input sequence; Based on the attention weights , calculate the context vector , as shown in Equation (15): ， where, is the hidden state of the encoder at time step t; This context vector will be used as the input to the decoder to generate the final prediction result.

7. The wind power prediction method according to claim 1, wherein Step S5 specifically includes the following steps: S51. Divide the validation set prediction error sequence into multiple error subsequences through continuous variational mode decomposition; S52. Perform data normalization on each error subsequence to obtain a normalized error subsequence, then use discrete wavelet transform to decompose the normalized error subsequence into low-frequency components and high-frequency components, and obtain the wavelet error features of each error subsequence according to the low-frequency components and high-frequency components; S53. Generate similar error features of wavelet error features through TimeGAN; S54. Restore the similar error features through inverse wavelet transform to obtain a normalized enhanced error sequence; S55. Denormalize the normalized enhanced error sequence to obtain an enhanced error sequence, and finally obtain the enhanced error sequences of each error subsequence.

8. The wind power prediction method according to claim 1, wherein Step S7 is specifically: correct the power prediction results of the test set through formula (23): ， Among them, is the corrected predicted value corresponding to time point i, in kilowatts; is the preliminary predicted wind power value corresponding to time point i, in kilowatts; is the error correction coefficient corresponding to time point i, whose value range is (0, 1); The error correction coefficient consists of a correction coefficient and a bias coefficient, as shown in formula (24): ， Among them, is the error correction coefficient corresponding to time point i; is the calibration coefficient corresponding to time point i; is the bias coefficient corresponding to time point i; Correction coefficient Specifically, as shown in formula (25): ， Among them, represents the actual wind power value at time point i; represents the initially predicted wind power value at time point i; represents the prediction error of the test set at time point i; Bias coefficient Specifically, as shown in formula (26): ， wherein, the gust wind speed at time point i, in meters per second, is the wind speed at time point i, in meters per second.

9. The wind power prediction method according to claim 1, wherein The wind power data includes the hourly wind power generation, unit: kilowatt; Wind speed, unit: m / s; Gust wind speed, unit: m / s; The preset time period includes a first preset time period and a second preset time period; The preset time period includes a first preset time period and a second preset time period; 80% of the sub-modal wind power data within the first preset time period is used as training set data in chronological order, and 20% is used as validation set data, where the sub-modal wind power data within the second preset time period is used as test set data.

10. A wind power prediction device based on deep learning and error correction, characterized in that It includes the following units: A decomposition and optimization unit, configured to collect wind power data within a preset time period, and use variational mode decomposition on the wind power data based on the particle swarm optimization algorithm to obtain multiple sub-modal wind power data; A data partitioning unit, configured to partition the sub-modal wind power data into training set data, validation set data, and test set data; A first prediction unit, configured to input the training set data into a preset EDA-GRU model for training to obtain a power prediction model, and then input the validation set data and the test set data into the power prediction model respectively to obtain the validation set power prediction result and the test set power prediction result; An error calculation unit, configured to calculate the residual as the prediction error according to the validation set power prediction result and the actual wind power value of the validation set data, and obtain the validation set prediction error sequence; An error enhancement unit, configured to divide the validation set prediction error sequence into multiple error subsequences through continuous variational mode decomposition, extract wavelet error features using discrete wavelet transform, and use TimeGAN to perform data augmentation on each error subsequence to obtain an enhanced error sequence; A second prediction unit, configured to input the enhanced error sequence into a preset EDA-GRU model for training to obtain an error prediction model, and predict the test set prediction error sequence with the same time series as the test set power prediction result through the error prediction model; A prediction correction unit, configured to construct an error correction coefficient through the test set prediction error to correct the test set power prediction result to obtain the final prediction result.

Images