Wind power prediction method and system based on data enhancement

Abstract

The invention discloses a wind power prediction method and system based on data enhancement, and the method comprises the steps: arranging a multi-measurement-point multivariable sensor in a wind power plant, and collecting the data in the wind power plant; carrying out normalization processing on the collected data, and carrying out feature selection by using adaptive LASSO to obtain a wind power plant prediction data set; constructing a space-time enhanced generative adversarial network SPATE-GAN model, and generating an enhanced data set; constructing a space-time prediction model KAN-BiTCN, outputting wind power prediction data of the wind power plant, improving an alpha evolutionary algorithm AE by adopting dynamic reverse learning and Fuch chaotic mapping, and optimizing model hyper-parameters; performing error correction on the prediction result through an RBF neural network, and outputting wind power prediction data of the wind power plant; important features can be selected comprehensively and accurately, and model parameters are optimized efficiently, so that the accuracy of wind power prediction of the wind power plant is ensured, and stable operation of a power system and reasonable arrangement of a power generation plan are realized.

Description

Technical Field

[0001] The present invention relates to a wind power prediction method and system, and particularly to a wind power prediction method and system based on data augmentation. Background Art

[0002] With the continuous increase in global energy demand and the enhancement of environmental protection awareness, wind energy, as an important clean and renewable energy source, has received extensive attention. Especially in China, the installed capacity of wind power has increased rapidly in recent years, and China has become one of the countries with the largest wind power generation in the world. The accuracy of wind power prediction is directly related to the stable operation of the power system and the reasonable arrangement of power generation plans.

[0003] Traditional wind power prediction methods mainly rely on meteorological data and simple statistical models. These methods have limitations in dealing with complex non-linear relationships and time series data, resulting in low prediction accuracy and reliability. For example, the changes in meteorological conditions and the complexity of factors such as the wind farm environment are difficult to accurately predict through traditional statistical models, and the prediction accuracy is difficult to meet the actual needs. To solve the problems of traditional methods, researchers have begun to explore more advanced wind power prediction technologies, and data-driven prediction models have gradually emerged. These models process a large amount of historical data through machine learning and deep learning technologies and can better capture the non-linear characteristics of wind power. For example, methods such as support vector machine (SVM) and neural network have been widely used in wind power prediction; these methods can effectively handle non-linear relationships and improve the prediction accuracy of wind power.

[0004] However, these models also have a series of problems in practical applications. First, the temporal and spatial characteristics of data are not fully considered, resulting in a decline in model performance when dealing with long sequence data. Second, feature selection is not precise enough to effectively capture the key features related to wind power prediction. In addition, the hyperparameter optimization process of the model is not efficient enough, which affects the prediction accuracy. Especially when facing complex and changeable meteorological conditions and wind farm environments, the prediction effect of the model is often not satisfactory. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to provide a wind power prediction method based on data augmentation to improve the accuracy and reliability of wind power prediction in wind farms. On the other hand, a wind power prediction system based on data augmentation is provided.

[0006] Technical Solution: A wind power prediction method based on data augmentation according to the present invention includes the following steps:

[0007] S1. Arrange multi-measurement-point and multi-variable sensors in the wind farm to collect data in the wind farm;

[0008] S2. Normalize the collected data, perform feature selection using Adaptive LASSO, and obtain the wind farm prediction dataset;

[0009] S3. Construct a spatio-temporal enhanced generative adversarial network SPATE-GAN model, generate an enhanced dataset, merge it with the wind farm prediction dataset, and then divide it into a training set, a validation set, and a test set;

[0010] S4. Construct a spatio-temporal prediction model KAN-BiTCN, output the wind power prediction data of the wind farm, and improve the alpha evolution algorithm AE using dynamic backpropagation learning and Fuch chaotic mapping to optimize the model hyperparameters;

[0011] S5. Perform error correction on the prediction results through an RBF neural network, and output the wind power prediction data of the wind farm.

[0012] Preferably, the data in the wind farm in S1 includes wind speed, wind direction, temperature, humidity, air pressure, actual power data, and the operating state data of wind turbines.

[0013] Preferably, the S2 includes:

[0014] S21. Convert each value of each feature into the interval [0, 1] to obtain a dataset containing different features;

[0015] S22. Estimate the initial coefficients by ordinary least squares OLS, and allocate weights according to the reciprocal of the absolute value of the initial coefficients;

[0016] S23. Multiply the feature matrix by the weight vector to construct a weighted feature matrix, and perform LASSO regression under the weighted condition to solve the objective function;

[0017] S24. According to the absolute value of the final regression coefficients, select the top 5 variables with the highest weights, which are the most important features for wind farm prediction, and form the wind farm prediction dataset.

[0018] Preferably, the S3 includes:

[0019] S31. Construct a SPATE-GAN model, define a generator, input a random noise vector, and output a wind farm prediction data sample;

[0020] S32. Define a discriminator, input a real data sample or a sample output by the generator, and output the probability that the sample is real data;

[0021] S33. Train the generator and the discriminator, input the random noise vector into the trained generator, output a large number of wind farm prediction data samples as enhanced data, and generate multiple enhanced samples by sampling different noise vectors multiple times;

[0022] S34. Combine the generated enhanced sample data with the original wind farm prediction data set, and divide it into a training set, a validation set, and a test set according to the ratio of 6:2:2.

[0023] Preferably, the construction of the spatio-temporal prediction model KAN-BiTCN in S4 includes:

[0024] Each convolutional layer of BiTCN uses multiple convolutional kernels for convolutional operations to extract local features of different scales. BiTCN captures past and future information of the time series through bidirectional convolution to construct the BiTCN part of the model;

[0025] Approximate complex functional relationships through learnable activation functions to construct the KAN part of the model;

[0026] Construct the KAN-BiTCN model and train it through the training set and the validation set. Input the test set into the trained KAN-BiTCN model, and perform feature extraction and transformation through the BiTCN layer and the KAN layer to output a prediction vector;

[0027] Correspond the dimension of the output prediction vector to the length of the time series, and output the wind power prediction data of the wind farm.

[0028] Preferably, the process of optimizing the model hyperparameters in S4 is as follows:

[0029] S41. Initialize the population using the Fuch chaotic mapping and the dynamic backtracking learning strategy. The formula is as follows:

[0030] P i = P min +(P max - P min )×Fuch(x i );

[0031] OP i = P max + P min - P i ;

[0032] Among them, P i represents the position of the i-th individual, P min and P max represent the minimum and maximum values of the hyperparameters respectively, x i represents a random number in the interval [0,1], Fuch(x i ) represents the Fuch chaotic mapping function, and OP i represents the inverse solution corresponding to each initial solution P i ;

[0033] S42. Calculate the fitness value of each individual. The fitness function is evaluated based on the performance of the spatio-temporal prediction model KAN-BiTCN under the current hyperparameters. The formula is as follows:

[0034] Fitness(P i ) = Error(KAN-BiTCN(P i ));

[0035] Among them, Error represents the prediction error, and KAN-BiTCN(P i ) represents the spatio-temporal prediction model under the hyperparameter P i ;

[0036] S43. Select the individual with the best fitness value as the alpha individual. According to the position of the alpha individual, use the Fuch chaotic mapping and dynamic reverse learning strategy to update the positions of other individuals. The formula is as follows:

[0037] P i(t+1) = P i(t) + β × (rand × P α (t) - P i(t) ) + γ × (rand × OP i(t) - P i(t) );

[0038] OP i(t+1) = P max + P min - P i(t+1 );

[0039] Among them, P i(t) represents the position of the i-th individual in the t-th generation, P i(t+1) represents the updated position, β and γ represent the learning factors, and rand represents a random number in the interval [0, 1];

[0040] S44. Recalculate the fitness value of the updated individual. The formula is the same as S43;

[0041] S45. Update the alpha individual. If the fitness value of an individual is better than the current alpha individual, update the alpha individual and its position;

[0042] S46. If the preset maximum number of iterations is reached or the fitness value converges, stop the iteration and output the optimal hyperparameters; otherwise, return to S43 and continue the iteration.

[0043] Preferably, the said S5 includes:

[0044] S51. Calculate the error between the initial prediction value and the actual value of the model. The formula is as follows:

[0045]

[0046] Among them, e represents the error, y represents the actual value, represents the initial predicted value of the model;

[0047] S52. Calculate the output of the hidden layer of the RBF neural network. The formula is as follows:

[0048]

[0049] Among them, H ij represents the output of the i-th input sample at the j-th hidden layer node, x i represents the i-th input sample, c j represents the center of the j-th hidden layer node, and σ j represents the width of the j-th hidden layer node;

[0050] S53. Calculate the error predicted value. The formula is as follows:

[0051]

[0052] Among them, represents the error predicted value, H represents the hidden layer output, and β represents the output weight of the RBF neural network;

[0053] S54. Add the predicted value of the initial model to the error predicted value of the RBF neural network to obtain the final predicted value. The formula is as follows:

[0054]

[0055] Among them, represents the final predicted value, represents the initial predicted value of the model, represents the error predicted value of the RBF neural network.

[0056] A wind power prediction system based on data augmentation according to the present invention includes:

[0057] A data acquisition module for arranging multi-measurement point and multi-variable sensors in a wind farm to collect data in the wind farm;

[0058] A data preprocessing and feature selection module for performing normalization processing on the collected data and using AdaptiveLASSO for feature selection to obtain a wind farm prediction data set;

[0059] A data augmentation and division module for constructing a spatio-temporal augmentation generative adversarial network SPATE-GAN model to generate an augmented data set, and dividing it into a training set, a validation set, and a test set after merging it with the wind farm prediction data set;

[0060] A model construction and optimization module, which is used to construct a spatio-temporal prediction model KAN-BiTCN, improve the alpha evolution algorithm AE by using dynamic backpropagation learning and Fuch chaotic mapping, and optimize the hyperparameters of the model;

[0061] An error correction module, which is used to correct the prediction results through an RBF neural network to obtain the wind power prediction data of the wind farm.

[0062] A computer device includes one or more processors, a memory, and one or more programs, where the one or more programs are stored in the memory and are configured to be executed by the one or more processors. When the program is executed by the processor, it implements the steps of a wind power prediction method based on data augmentation as described in any one of claims 1-7.

[0063] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps of a wind power prediction method based on data augmentation as described in any one of claims 1-7.

[0064] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages: 1. By constructing a data augmentation model based on the spatio-temporal enhanced generative adversarial network SPATE-GAN, giving full play to its ability to generate high-quality and diverse data, effectively solving problems such as insufficient and unbalanced data that may exist in the data collection process of the wind farm, providing richer and more representative training samples for the wind power prediction model, and thus improving the generalization ability and prediction accuracy of the model; 2. By using the spatio-temporal prediction model KAN-BiTCN to predict wind power, it can fully extract the spatio-temporal features in the wind farm data. Its bidirectional convolution operation can effectively capture the forward and backward relationships in time series data, and the spatio-temporal convolution operation can obtain the feature information at different time points and spatial positions. The multi-layer structure design can gradually extract the spatio-temporal features at the abstract level, further enhancing the model's ability to grasp the variation law of wind power; 3. By optimizing the hyperparameters of the KAN-BiTCN model through the alpha evolution algorithm AE improved by dynamic backpropagation learning and Fuch chaotic mapping, the improved AE algorithm constructs an evolutionary matrix to update the solution, effectively balancing the relationship between exploration and exploitation, accelerating the algorithm convergence and improving the efficiency, so as to find a better hyperparameter combination for the wind power prediction model and further improve the model performance; 4. By using the RBF neural network error correction method to correct the prediction results of the model, it can effectively reduce the deviation between the model prediction results and the actual wind power of the wind farm, reduce the prediction error, improve the reliability of the prediction results, and provide a more accurate basis for the stable operation of the power system and the reasonable arrangement of the power generation plan. Description of the Drawings

[0065] Figure 1Schematic diagram of the process of the present invention;

[0066] Figure 2 Schematic diagram of the process for optimizing the hyperparameters of the improved Alpha Evolution Algorithm AE optimization model of the present invention;

[0067] Figure 3 Schematic diagram of the process for error correction by the RBF neural network error correction method of the present invention. Detailed implementation manners

[0068] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0069] S1. Arrange multi-measurement-point and multi-variable sensors in the wind farm to collect the wind speed, wind direction, temperature, humidity, air pressure, actual power data in the wind farm, and the operating state data of the wind turbines in real time;

[0070] S2. Perform normalization processing on the collected data, and use AdaptiveLASSO for feature selection to obtain the wind farm prediction data set;

[0071] (1) Convert each value of each feature to the interval [0, 1] to obtain a data set containing different features:

[0072]

[0073] Among them, X represents the original data point, and X min represents the minimum value of this feature in the data set, which is used to ensure that the minimum value after normalization is 0, and X max represents the maximum value of this feature in the data set, which is used to ensure that the maximum value after normalization is 1, and X norm represents the data point after normalization;

[0074] (2) Use ordinary least squares OLS to estimate the initial coefficient β, and the calculation formula is as follows:

[0075] β = (X T X) -1 X T y;

[0076] Among them, X represents the feature matrix, and y represents the response variable. This formula is used to calculate the initial coefficient and provides a basis for subsequent calculation of weights;

[0077] (3) Allocate weights according to the reciprocal of the absolute value of the initial coefficient, and the calculation formula is as follows:

[0078]

[0079] If β j = 0, then let w j = 0 to avoid division by zero;

[0080] (4) Construct a weighted feature matrix by multiplying the feature matrix X with the weight vector w to obtain the weighted feature matrix X weighted , and the calculation formula is as follows:

[0081] X weighted = X · w;

[0082] This step is to introduce weights in the subsequent LASSO regression so that different features are punished to different degrees;

[0083] (5) Perform LASSO regression under weighted conditions to solve the objective function.

[0084]

[0085] Among them, λ represents the regularization parameter used to control the intensity of the penalty term; p represents the number of features. By solving this objective function, the final regression coefficient β can be obtained, and then feature selection can be achieved.

[0086] (6) Select the top 5 variables with the largest absolute values of the final regression coefficients. These variables are the most important features for wind farm prediction. Combine them to form a new data set as the wind farm prediction data set.

[0087] S3. Construct a spatio-temporal enhanced generative adversarial network SPATE-GAN model, generate an enhanced data set, and after merging it with the wind farm prediction data set, divide it into a training set, a validation set, and a test set;

[0088] (1) Define the generator G: The generator G is a deep neural network. Its input is a random noise vector z, and its output is the generated wind farm prediction data sample G(z). Its goal is to generate samples as close as possible to the true data distribution. The formula is as follows:

[0089] G: z → G(z);

[0090] Among them, z represents the noise vector sampled from the prior distribution p z (z), and G(z) represents the wind farm prediction data sample generated by the generator according to the noise vector;

[0091] Define the discriminator D: The discriminator D is also a deep neural network. Its input is the real data sample x or the sample G(z) generated by the generator, and its output is the probability D(x) or D(G(z)) that the sample is real data;

[0092] D: x → D(x), G(z) → D(G(z));

[0093] Among them, D(x) represents the probability that the discriminator judges the sample x as real data, and D(G(z)) represents the probability that the discriminator judges the generated sample G(z) as real data;

[0094] (2) Train the SPATE-GAN model and train the discriminator D. During the training process, the goal of the discriminator D is to maximize the discrimination accuracy of real data and generated data, that is, to maximize the following loss function:

[0095]

[0096] Among them, D(G(z)) is as close to 1 as possible, that is, the generated samples can make the discriminator misjudge them as real data;

[0097] (3) Generate enhanced data: Use the trained generator G to input the random noise vector z to generate a large number of wind farm prediction data samples as enhanced data:

[0098] G(z) enhanced ;

[0099] Among them, G(z) enhanced represents the generated enhanced data samples. By sampling different noise vectors z multiple times, multiple enhanced samples can be generated.

[0100] (4) Merge the enhanced data, merge the generated enhanced data with the original wind farm prediction data set, and then divide it into a training set, a validation set, and a test set according to the ratio of 6:2:2.

[0101] S4. Build a spatio-temporal prediction model KAN-BiTCN to output the wind power prediction data of the wind farm, and use dynamic reverse learning and Fuch chaotic mapping to improve the alpha evolutionary algorithm AE to optimize the model hyperparameters;

[0102] (1) Build a prediction model KAN-BiTCN, build the BiTCN part. BiTCN consists of multiple convolutional layers. Each convolutional layer uses multiple convolutional kernels for convolution operations to extract local features of different scales;

[0103] Assume that the input sequence is x ∈ R T×C , where T represents the time series length, C represents the feature dimension, and the output of the convolutional layer can be expressed as:

[0104] h = ReLU(W * x + b);

[0105] Among them, W represents the convolutional kernel weight, b represents the bias term, * represents the convolution operation, ReLU represents the activation function, and BiTCN captures the past and future information of the time series through bidirectional convolution.

[0106] (2) Construct the KAN part. Based on the Kolmogorov-Arnold representation theorem, KAN approximates complex functional relationships through a learnable activation function;

[0107] Assume that the feature output by BiTCN is h ∈ R T×D , where D represents the feature dimension output by BiTCN, and the output of KAN can be expressed as:

[0108]

[0109] where n represents the number of nodes in KAN, λ i represents the node weight, and φ i represents a learnable activation function used to map the input features to the output space;

[0110] (3) Train the model. Use the training set and the validation set to train the constructed KAN-BiTCN model;

[0111] (4) Predict data. Input the test set into the trained KAN-BiTCN model. The model will sequentially perform feature extraction and transformation through the BiTCN layer and the KAN layer, and finally output the prediction result. The formula is as follows:

[0112] y = KAN-BiTCN(x test );

[0113] where x test represents the test set input, and y represents the predicted vector output by the model, which contains the predicted values of the wind power operation data for a future period of time;

[0114] (5) Output the predicted vector. The predicted vector y output by the model is the predicted result of the wind power operation data for a future period of time. Its dimension corresponds to the length n of the time series and can be expressed as:

[0115]

[0116] where y i represents the predicted value of the wind power at the i-th time point;

[0117] (6) Initialize the population. Use the Fuch chaotic mapping and the dynamic reverse learning strategy to initialize the population to improve the diversity and ergodicity of the population. The formula is as follows:

[0118] P i = P min + (P max - P min ) × Fuch(x i );

[0119] OP i= P max + P min - P i ;

[0120] where P i represents the position of the i-th individual, P min and P max represent the minimum and maximum values of the hyperparameters respectively, x i represents a random number in the interval [0, 1], Fuch(x i ) represents the Fuch chaotic mapping function, OP i represents the inverse solution corresponding to each initial solution P i In this way, an initial population is generated, which can expand the search space and improve the quality of the initial solution;

[0121] (7) Fitness evaluation, calculate the fitness value of each individual. The fitness function is evaluated according to the performance of the spatio-temporal prediction model KAN - BiTCN under the current hyperparameters, such as the prediction error, etc. The formula is:

[0122] Fitness(P i ) = Error(KAN - BiTCN(P i ));

[0123] where Error represents the prediction error, and KAN - BiTCN(P i ) represents the spatio-temporal prediction model under the hyperparameter P i ;

[0124] (8) Alpha individual selection, select the individual with the best fitness value as the alpha individual, denoted as α, and its position is P α ;

[0125] Position update, according to the position of the alpha individual, use the Fuch chaotic mapping and dynamic reverse learning strategy to update the positions of other individuals. The formula is as follows:

[0126] P i(t+1) = P i(t) + β × (rand × P α (t) - P i(t) ) + γ × (rand × OP i(t) - P i(t) );

[0127] OP i(t+1) = P max + P min - P i(t+1) ;

[0128] where P i(t) represents the position of the i-th individual in the t-th generation, Pi(t+1) It represents the updated position, β and γ represent the learning factors, and rand represents a random number within the interval [0, 1]. In this way, the individual can not only learn from the alpha individual but also explore the new solution space by using the reverse learning strategy;

[0129] (9) Fitness re-evaluation, recalculate the fitness value of the updated individual, and the formula is the same as that in the step of fitness evaluation; update the alpha individual. If the fitness value of an individual is better than the current alpha individual, then update the alpha individual and its position; judge the termination condition. If the preset maximum number of iterations or the fitness value converges, stop the iteration and output the optimal hyperparameters; otherwise, return to the step of position update to continue the iteration;

[0130] S5. Perform error correction on the prediction results through the RBF neural network and output the wind power prediction data of the wind farm;

[0131] (1) Calculate the error between the initial predicted value and the actual value of the model. The formula is:

[0132]

[0133] where e represents the error, y represents the actual value, represents the initial predicted value of the model;

[0134] (2) Calculate the output of the hidden layer. The output H of the hidden layer of the RBF neural network can be calculated by the following formula:

[0135]

[0136] where H ij represents the output of the i-th input sample at the j-th hidden layer node, x i represents the i-th input sample, c j represents the center of the j-th hidden layer node, and σ j represents the width of the j-th hidden layer node;

[0137] (3) Calculate the error prediction value. The error prediction value can be calculated by the following formula:

[0138]

[0139] where represents the error prediction value, H represents the output of the hidden layer, and β represents the output weight of the RBF neural network;

[0140] (4) Calculate the final predicted value. Add the predicted value of the initial model and the error prediction value of the RBF neural network to obtain the final predicted value. The formula is:

[0141]

[0142] Among them, represents the final predicted value, represents the initial predicted value of the model, represents the error predicted value of the RBF neural network.

[0143] Finally, the wind power prediction result is obtained, which is used to ensure the stable operation of the power system and ensure the reasonable arrangement of the power generation plan.

Claims

1. A wind power prediction method based on data augmentation, characterized in that It includes the following steps: S1. Arrange multi-point multi-variable sensors in the wind farm to collect data in the wind farm; S2. Perform normalization processing on the collected data, use Adaptive LASSO for feature selection, and obtain the wind farm prediction data set; S3. Construct a spatio-temporal enhanced generative adversarial network SPATE-GAN model, generate an enhanced data set, merge it with the wind farm prediction data set, and divide it into a training set, a validation set, and a test set; S4. Construct a spatio-temporal prediction model KAN-BiTCN, output the wind power prediction data of the wind farm, and adopt dynamic reverse learning and Fuch chaotic mapping to improve the alpha evolution algorithm AE to optimize the model hyperparameters; S5. Perform error correction on the prediction results through an RBF neural network and output the wind power prediction data of the wind farm.

2. The wind power prediction method according to claim 1, characterized in that, The data in the wind farm described in S1 includes wind speed, wind direction, temperature, humidity, air pressure, actual power data, and the operating state data of wind turbines.

3. The wind power prediction method according to claim 1, wherein The S2 includes: S21. Convert each value of each feature into the interval [0,1] to obtain a data set containing different features; S22. Estimate the initial coefficients by ordinary least squares OLS and assign weights according to the reciprocal of the absolute value of the initial coefficients; S23. Multiply the feature matrix by the weight vector to construct a weighted feature matrix, and perform LASSO regression under weighted conditions to solve the objective function; S24. Select the 5 variables with the top 5 weights according to the absolute value of the final regression coefficients, which are the most important features for wind farm prediction, and form the wind farm prediction data set.

4. The wind power prediction method according to claim 1, wherein The S3 includes: S31. Construct a SPATE-GAN model, define a generator, input a random noise vector, and output a wind farm prediction data sample; S32. Define a discriminator, input a real data sample or a sample output by the generator, and output the probability that the sample is real data; S33. Train the generator and the discriminator, input the random noise vector into the trained generator, output a large number of wind farm prediction data samples as enhanced data, and generate multiple enhanced samples by sampling different noise vectors multiple times; S34. Merge the generated enhanced sample data with the original wind farm prediction data set and divide it into a training set, a validation set, and a test set according to the ratio of 6:2:

2.

5. The wind power prediction method according to claim 1, characterized in that, The construction of the spatio-temporal prediction model KAN-BiTCN described in S4 includes: Each convolutional layer of BiTCN uses multiple convolutional kernels for convolutional operations to extract local features of different scales. BiTCN captures the past and future information of the time series through bidirectional convolution to construct the BiTCN part of the model; Approximate complex functional relationships through a learnable activation function to construct the KAN part of the model; Construct a KAN-BiTCN model and train it through the training set and the validation set. Input the test set into the trained KAN-BiTCN model, perform feature extraction and transformation through the BiTCN layer and the KAN layer, and output a prediction vector; Correspond the dimension of the output prediction vector to the length of the time series and output the wind power prediction data of the wind farm.

6. The wind power prediction method according to claim 5, characterized in that, The process of optimizing the model hyperparameters described in S4 is: S41. Initialize the population using the Fuch chaotic mapping and dynamic reverse learning strategy, with the formula as follows: P i = P min +(P max - P min ) × Fuch(x i ); OP i = P max + P min - P i ; Among them, P i represents the position of the i-th individual, P min and P max represent the minimum and maximum values of the hyperparameters respectively, x i represents a random number in the interval [0, 1], Fuch(x i ) represents the Fuch chaotic mapping function, OP i represents the inverse solution corresponding to each initial solution P i ; S42. Calculate the fitness value of each individual. The fitness function is evaluated based on the performance of the spatio-temporal prediction model KAN-BiTCN under the current hyperparameters, with the formula as follows: Fitness(P i ) = Error(KAN - BiTCN(P i )); where Error represents the prediction error, and KAN-BiTCN(P i ) represents the spatio-temporal prediction model under the hyperparameter P i ; S43. Select the individual with the best fitness value as the alpha individual. According to the position of the alpha individual, update the positions of other individuals using the Fuch chaotic mapping and dynamic reverse learning strategy, with the formula as follows: P i(t+1) = P i(t) + β × (rand × P α (t) - P i(t) ) + γ × (rand × OP i(t) - P i(t) ); OP i(t+1) = P max + P min - P i(t+1) ; Among them, $P$ i(t) represents the position of the $i$-th individual in the $t$-th generation, and $P$ i(t+1) represents the updated position. $\beta$ and $\gamma$ represent learning factors, and $rand$ represents a random number in the interval $[0, 1]$; S44. Recalculate the fitness values of the updated individuals, with the formula the same as S43; S45. Update the alpha individual. If the fitness value of an individual is better than the current alpha individual, update the alpha individual and its position; S46. If the preset maximum number of iterations is reached or the fitness value converges, stop the iteration and output the optimal hyperparameters; otherwise, return to S43 and continue the iteration.

7. The wind power prediction method according to claim 1, wherein The above S5 includes: S51. Calculate the error between the initial prediction value and the actual value of the model, with the formula as follows: where e represents the error, y represents the actual value, represents the initial predicted value of the model; S52. Calculate the output of the hidden layer of the RBF neural network, with the formula as follows: Among them, H ij represents the output of the i-th input sample at the j-th hidden layer node, x i represents the i-th input sample, c j represents the center of the j-th hidden layer node, σ j represents the width of the j-th hidden layer node; S53. Calculate the error prediction value, with the formula as follows: Among them, represents the error prediction value, H represents the output of the hidden layer, and β represents the output weight of the RBF neural network; S54. Add the prediction value of the initial model and the error prediction value of the RBF neural network to obtain the final prediction value, with the formula as follows: Among them, represents the final predicted value, represents the initial predicted value of the model, represents the error predicted value of the RBF neural network.

8. A wind power prediction system based on data augmentation, characterized in that, It includes: A data acquisition module, which is used to arrange multi-measurement-point and multi-variable sensors in the wind farm to collect data in the wind farm; A data preprocessing and feature selection module, which is used to perform normalization processing on the collected data and use AdaptiveLASSO for feature selection to obtain the wind farm prediction data set; A data augmentation and division module, which is used to construct a spatio-temporal augmentation generative adversarial network SPATE-GAN model to generate an augmented data set, and after merging it with the wind farm prediction data set, divide it into a training set, a validation set and a test set; A model construction and optimization module, which is used to construct a spatio-temporal prediction model KAN-BiTCN, and improve the alpha evolutionary algorithm AE using dynamic reverse learning and Fuch chaotic mapping to optimize the model parameters; An error correction module, which is used to correct the prediction result through an RBF neural network to obtain the wind power prediction data of the wind farm.

9. A computer device, characterized in that, It includes one or more processors, a memory, and one or more programs, where the one or more programs are stored in the memory and are configured to be executed by the one or more processors. When the program is executed by the processor, it implements the steps of a wind power prediction method based on data augmentation as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a wind power prediction method based on data augmentation as described in any one of claims 1-7.

Images