LSTM-Transform medium-and-long-term wind power prediction combination model method and LSTM-Transform medium-and-long-term wind power prediction combination model system

Abstract

The invention discloses an LSTM-Transform medium and long term wind power prediction combination model method and system, and the method comprises the steps: obtaining meteorological data of a wind power plant, carrying out the preprocessing of the meteorological data, capturing the internal characteristics of an original data set through employing a generative adversarial network, and generating and synthesizing a historical weather data sample, and carrying out the expansion of an existing data set; performing empirical mode decomposition on the enhanced data set to obtain intrinsic mode components of different time scales, and performing variational mode decomposition on the enhanced data set to obtain a fan power trend sequence and a periodic sequence; constructing a SwinLSTM network model, training the SwinLSTM network model by using data obtained by signal decomposition, and respectively predicting fan position comprehensive environment data and fan position wind power; and constructing an adaptive weight adjustment method based on deep reinforcement learning flexible action evaluation, and performing adaptive weight adjustment on the theoretical wind power and the predicted wind power to obtain the final predicted wind power.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and in particular to an LSTM-Transformer medium- and long-term wind power prediction combined model method and system. Background Art

[0002] As the global demand for clean energy continues to grow, wind power generation, as an important part of renewable energy, is gradually becoming a dominant force in energy production. However, the characteristics of wind power generation, including its uncertainty and volatility, pose a challenge to the stable operation of the power system. In this context, accurate prediction of wind power generation has become one of the key factors in optimizing grid dispatch and achieving efficient energy management.

[0003] Although traditional prediction methods have achieved certain results, they still show their limitations when faced with the complex influence of wind speed, wind direction and other environmental factors. Therefore, using advanced data analysis technology and machine learning algorithms is becoming a new direction to improve the accuracy of wind power prediction. A major challenge for using deep learning methods for prediction is the lack of sufficient data sets. Since data-driven methods are prone to overfitting the training data set, this will directly reduce the prediction accuracy of the neural network. Especially for medium- and long-term predictions, due to the longer prediction time limit, the demand for data sets is greater, and insufficient data will significantly affect the prediction accuracy of wind farm power generation. Generative Adversarial Networks (GAN) provides a solution for data-driven algorithms to enhance training data sets, effectively avoiding the overfitting problem. In addition, GAN can also repair and enhance data. By learning the distribution characteristics of data, GAN can recover information from damaged or incomplete data and improve the integrity and quality of data. Using GAN networks to process data helps us explore the structure and characteristics of wind power data sets more deeply, thereby enhancing our understanding and analysis of the relationship between wind power and environmental data. Signal decomposition technology has an important impact on wind power prediction using deep learning. It can effectively process complex wind power time series data. At the same time, the signal decomposition method helps to deeply analyze the dynamic changes and periodic characteristics in wind power data. On this basis, the most important thing is the selection of the prediction model. A reasonable and effective prediction strategy can often greatly improve the prediction accuracy.

[0004] The existing research on combined prediction models is divided into two stages. First, some sub-models are selected, and then the optimal weights of the sub-models are generated based on the given training data set. Most of the existing combined model prediction methods use a set of fixed weights to merge the sub-model results. In recent years, reinforcement learning, as one of the popular technologies in the field of artificial intelligence, has demonstrated powerful capabilities in many fields. Through trial and error and feedback mechanisms, reinforcement learning agents can continuously optimize decision-making strategies and improve system performance and efficiency. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, in view of the problem that the existing medium- and long-term wind power prediction technology takes less historical meteorological factors into consideration and has low prediction accuracy, the present invention provides a LSTM-Transformer medium- and long-term wind power prediction combined model method and system, which solves the problem of insufficient historical wind power meteorological data and improves prediction accuracy.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In the first aspect, the present invention provides a LSTM-Transformer medium- and long-term wind power forecasting combined model method, including: obtaining and preprocessing the meteorological data of the wind farm, using a generative adversarial network to capture the intrinsic characteristics of the original data set, and generating synthetic historical weather data samples to expand the existing data set; performing empirical mode decomposition on the enhanced data set to obtain intrinsic mode components of different time scales, and performing variational mode decomposition on the enhanced data set to obtain wind turbine power trend series and periodic series; constructing a SwinLSTM network model, training the SwinLSTM network model using data obtained by signal decomposition, and predicting the comprehensive environmental data of the wind turbine location and the wind power at the wind turbine location respectively; constructing an adaptive weight adjustment method based on deep reinforcement learning flexible action evaluation, and adaptively adjusting the weights of the theoretical wind power and the predicted wind power to obtain the final predicted wind power.

[0009] As a preferred solution of the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in the present invention, the generating of synthetic historical weather data samples to expand the existing data set includes:

[0010] Use an embedding network to capture the characteristics of wind power time series data and map it to a low-dimensional latent space to retain the key information of the time series;

[0011] Construct the generator G and discriminator D, where G(z) represents the generator function that maps the latent space vector z sampled from the standard normal distribution to the data space, and D(x) represents the probability that the discriminator network output x comes from the training data rather than the generator;

[0012] The segmented data is input into DCGAN for training, the real wind power data label is 1, and the generated fake data label is 0 for adversarial training, and the performance of the discriminator is measured using binary cross entropy loss;

[0013] The distribution of generated data is compared with that of real data, evaluated using the Kullback-Leibler divergence, and the model is fine-tuned using real wind power data.

[0014] As a preferred solution of the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in the present invention, wherein: the empirical mode decomposition of the enhanced data set to obtain the intrinsic mode components of different time scales includes:

[0015] For an original wind power sequence x(i), all large values ​​are determined as the upper envelope, and all small values ​​are determined as the lower envelope. m(i) represents the mean value of the upper envelope and the lower envelope. The first component is h1(i) = x(i)-m(i);

[0016] In the next round of screening, h1(i) is regarded as the original data, m1(i) is the mean value of the upper and lower envelopes of h1(i), and the second component h2(i) is determined;

[0017] The screening process is repeated n times until h n (i) is an intrinsic mode function or residual component r n (i) is a monotonic function, terminating the decomposition process;

[0018] Specify q1(i)=h1(i), q2(i)=h2(i), …, q k (i) = h k (i),, x(i), is finally decomposed into n eigenmode components q t (i) and a residual component r n (i).

[0019] As a preferred solution of the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in the present invention, wherein: the variational mode decomposition of the enhanced data set to obtain the wind turbine power trend sequence and period sequence includes:

[0020] Assume that the wind power data f is decomposed into i, components, and the fraction-preserving decomposition sequence is a modal component with a limited bandwidth and an intermediate frequency. At the same time, the sum of the estimated bandwidths of each mode is minimized. The constraint condition is that the sum of all modes is equal to the original data, and the constrained variational problem is transformed into an unconstrained variational problem.

[0021] The alternating direction multiplier iteration algorithm is used to continuously update each component and its frequency to obtain the saddle point of the unconstrained model, that is, the optimal solution to the original problem. The algorithm re-estimates the repetition frequency according to the power spectrum repetition frequency of each component and initializes the parameters.

[0022] Execute cycle n=n+1, and when the frequency ω>0, update each component and repeat the steps until the stop condition is met.

[0023] As a preferred solution of the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in the present invention, wherein: the construction of the SwinLSTM network model includes:

[0024] The construction of the STB module includes adopting a window-based multi-head self-attention technique to reduce the number of elements involved in a single calculation, capturing contextual information by moving the position of the window at different levels, and inputting the input sequence into a normalization layer and a multi-layer perception mechanism to capture global spatial dependencies;

[0025] Building a SwinLSTM Cell based on the STB module includes capturing temporal dependencies by introducing convolution operators into the conversion from input to state and from state to state, calculating similarity scores between all positions through a self-attention mechanism to capture global spatial dependencies, and merging the input gate, forget gate, and output gate into a filter gate; combining the temporal dependencies and global spatial dependencies to achieve average updating of cell states and hidden states to capture long-term and short-term temporal dependencies;

[0026] The SwinLSTM network model is constructed with the SwinLSTM Cell as the core frequency, including dividing the wind power data at time t into non-overlapping data blocks, flattening the data blocks and inputting them into the patch embedding layer, and the SwinLSTM layer receives the converted image patches, hidden states H t-1 and cell status C t-1 To generate the hidden state H t and cell status C t , H t is copied into two copies, one for the reconstruction layer and the other for C t Together with the SwinLSTM layer for the next time step;

[0027] The SwinLSTM network model is used to process wind power data to obtain predicted wind turbine power generation.

[0028] As a preferred solution of the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in the present invention, it also includes:

[0029] Based on the wind power signal components obtained by variational mode decomposition, a standardized matrix of the original data is obtained, and the correlation coefficient matrix R of the standardized matrix is ​​calculated. ij ) m×n ;

[0030] Calculate the eigenvalue matrix and the eigenvector corresponding to the eigenvalue, and calculate the contribution rate T i and the cumulative contribution rate η i , according to the cumulative contribution rate η i Determine the number of principal components to be selected;

[0031] The reduced-dimensional data is obtained by selecting the eigenvalues ​​of the principal components and the corresponding eigenvectors.

[0032] As a preferred solution of the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in the present invention, the acquisition of the final predicted wind power includes:

[0033] Define basic parameters, including reward R, action value function Q(s,a), state value function V(s), and advantage function A; build a SAC network model to maximize the expected cumulative reward by optimizing the policy network and value function network, while maximizing the entropy of action distribution; build an Actor network model and a Critic network model, and formulate a Markov decision process to obtain a combined model prediction method based on deep reinforcement learning flexible action evaluation, and adaptively adjust the sub-model prediction weights;

[0034] Collecting data of actual physical systems related to the wind farm, and calculating theoretical wind turbine power generation according to the data predicted by the SwinLSTM network model;

[0035] The weights of the theoretical wind turbine power generation and the predicted wind turbine power generation are adaptively adjusted through the SAC algorithm to achieve prediction of future wind power.

[0036] In a second aspect, the present invention provides a LSTM-Transformer medium- and long-term wind power forecasting combined model system, comprising:

[0037] The data set processing module is used to obtain and preprocess the meteorological data of the wind farm, use the adversarial generative network to capture the intrinsic characteristics of the original data set, and generate synthetic historical weather data samples to expand the existing data set;

[0038] The signal decomposition module is used to perform empirical mode decomposition on the enhanced data set to obtain modal components of different time scales, and to perform variational mode decomposition on the enhanced data set to obtain the wind turbine power trend series and period series;

[0039] A training prediction module is used to construct a SwinLSTM network model, and train the SwinLSTM network model using the data obtained by signal decomposition to respectively predict the comprehensive environmental data of the wind turbine location and the wind power at the wind turbine location;

[0040] The result acquisition module is used to construct an adaptive weight adjustment method based on deep reinforcement learning flexible action evaluation, and to adaptively adjust the weights of the theoretical wind power and the predicted wind power to obtain the final predicted wind power.

[0041] In a third aspect, the present invention provides an electronic device, comprising:

[0042] Memory and processor;

[0043] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the LSTM-Transformer medium- and long-term wind power forecasting combined model method are implemented.

[0044] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the LSTM-Transformer medium- and long-term wind power forecasting combined model method.

[0045] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a LSTM-Transformer medium- and long-term wind power forecasting combined model method and system, which uses a deep convolutional generative adversarial network (DCGAN) to expand the original wind power data set to capture and retain its intrinsic characteristics, and at the same time, taking into account the complexity of the factors affecting wind power, the enhanced data set is processed by the empirical mode decomposition technology to extract the intrinsic mode components on different time scales. On this basis, the SwinLSTM network is used to predict the comprehensive environmental data and wind power of the wind turbine location respectively, effectively reducing the prediction error. Finally, the soft action evaluation (Soft Actor-Critic, SAC) method based on deep reinforcement learning is introduced to realize adaptive weight adjustment, dynamically optimize the prediction weights of each sub-model, and ensure the accuracy and flexibility of the prediction strategy. This study not only improves the technical level of wind power forecasting, but also provides strong support for the stable operation and energy management of the power system, and helps to promote the effective utilization and sustainable development of renewable energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0047] Figure 1 A schematic diagram of the overall process logic of the LSTM-Transformer medium- and long-term wind power forecasting combined model method according to an embodiment of the present invention;

[0048] Figure 2 A specific flow chart of the LSTM-Transformer medium- and long-term wind power forecasting combined model method according to an embodiment of the present invention;

[0049] Figure 3 A background block diagram of the LSTM-Transformer medium- and long-term wind power forecasting combined model method according to an embodiment of the present invention;

[0050] Figure 4 A structural block diagram of the LSTM-Transformer medium- and long-term wind power forecasting combined model method according to an embodiment of the present invention;

[0051] Figure 5 A schematic diagram of a DCGAN generator G of the LSTM-Transformer medium- and long-term wind power forecasting combined model method according to an embodiment of the present invention;

[0052] Figure 6 A schematic diagram of a SwinTransformer block of the LSTM-Transformer medium- and long-term wind power forecasting combined model method according to an embodiment of the present invention;

[0053] Figure 7 A schematic diagram of a SwinLSTM module of the LSTM-Transformer medium- and long-term wind power forecasting combined model method according to an embodiment of the present invention;

[0054] Figure 8 A schematic diagram of a SwinLSTM-B module of the LSTM-Transformer medium- and long-term wind power forecasting combined model method according to an embodiment of the present invention;

[0055] Fig. 9 A comparison chart of predicted values ​​and actual values ​​of the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in one embodiment of the present invention;

[0056] Fig.10A comparison chart of predicted values ​​and actual values ​​of the original data set using the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in one embodiment of the present invention;

[0057] Fig.11 A comparison diagram of the predicted value and the actual value without signal decomposition of the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in one embodiment of the present invention;

[0058] Fig.12 This is an error bar chart comparing various methods of the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0060] Example 1

[0061] Reference Figure 1-Figure 8 As an embodiment of the present invention, a LSTM-Transformer medium- and long-term wind power forecasting combined model method is provided. Figure 1 The specific steps shown include:

[0062] S100: Obtain meteorological data of the wind farm and aggregate and clean the data;

[0063] S200: Use a generative adversarial network to capture the intrinsic features of the original dataset and generate synthetic historical weather data samples to expand the existing dataset;

[0064] S300: performing empirical mode decomposition on the enhanced data set to obtain intrinsic mode components of different time scales, and performing variational mode decomposition on the enhanced data set to obtain a wind turbine power trend series and a period series;

[0065] S400: Build a SwinLSTM network model, use the data obtained from signal decomposition to train the SwinLSTM network model, and predict the comprehensive environmental data of the wind turbine location and the wind power at the wind turbine location respectively;

[0066] S500: PCA is used to reduce the dimensionality of input variables, which improves the prediction accuracy while maintaining the calculation speed of the SwinLSTM network model and overcoming the problem of overfitting.

[0067] S600: Construct an adaptive weight adjustment method based on deep reinforcement learning flexible action evaluation, and train the base module to convergence based on the enhanced data set;

[0068] S700: Adaptively adjust the weights of the theoretical wind power and the predicted wind power to obtain the final predicted wind power, so as to realize the prediction of future wind power.

[0069] It should be noted that the present invention provides a LSTM-Transformer medium- and long-term wind power forecasting combined model method and system, which uses a deep convolutional generative adversarial network (DCGAN) to expand the original wind power data set to capture and retain its intrinsic characteristics, while taking into account the complexity of factors affecting wind power, and processing the enhanced data set through the empirical mode decomposition technology to extract the intrinsic mode components on different time scales. On this basis, the SwinLSTM network is used to predict the comprehensive environmental data and wind power of the wind turbine position respectively, effectively reducing the prediction error. Finally, the soft action evaluation (Soft Actor-Critic, SAC) method based on deep reinforcement learning is introduced to realize adaptive weight adjustment, dynamically optimize the prediction weights of each sub-model, and ensure the accuracy and flexibility of the prediction strategy. This study not only improves the technical level of wind power prediction, but also provides strong support for the stable operation and energy management of the power system, which helps to promote the effective utilization and sustainable development of renewable energy.

[0070] The flow chart of the present invention is as follows Figure 2 As shown, a LSTM-Transformer medium- and long-term wind power forecasting combined model method based on signal decomposition, the background diagram of the present invention is as follows Figure 3 As shown, the overall network model structure diagram constructed by the present invention is as follows Figure 4 shown.

[0071] In the embodiment of the present application, the above step S100 includes the following sub-steps A1-A3;

[0072] In A1: set the outliers in the data set to 0, select continuous sequence data of different lengths to construct the training set and test set, where the outliers are set to the predicted values ​​of the test set;

[0073] In A2: Normalize the data set and scale the data to the range of [-1, 1] or [0, 1] so that the neural network can better process it. The formula is as follows:

[0074]

[0075] In A3: Noise data that obviously deviates from the normal range is identified and processed.

[0076] It should be noted that the above step S100 ensures that subsequent analysis and prediction are based on high-quality and accurate data by acquiring and cleaning the meteorological data of the wind farm. It not only removes noise and outliers and improves the reliability and availability of data, but also lays the foundation for generating more realistic synthetic data, thereby directly improving the performance and accuracy of the entire prediction system.

[0077] In the embodiment of the present application, the above step S200 includes the following sub-steps B1-B2;

[0078] In B1: an embedding network is used to capture the characteristics of wind power time series data and map it into a low-dimensional latent space to retain the key information of the time series;

[0079] In B2: Figure 5 As shown, the generator G and the discriminator D are constructed, G(z) represents the generator function that maps the latent space vector z sampled from the standard normal distribution to the data space, and D(x) represents the probability that the discriminator network output x comes from the training data rather than the generator;

[0080] In B3: The segmented data is input into DCGAN for training, the real wind power data label is 1, the generated fake data label is 0 for adversarial training, and the binary cross entropy loss is used to measure the performance of the discriminator:

[0081] l(x,y)=L={l1,...,l N} T ,l n =-[y n *logx n +(1-y n )*log(1-x n )]

[0082] In B4: Compare the distribution of generated data with real data, use Kullback-Leibler divergence for evaluation, fine-tune the model with real wind power data, and use DCGAN to generate the required historical weather time series data.

[0083] It should be noted that the above step S200 effectively expands the existing data set by using a generative adversarial network (GAN) to capture the intrinsic features of the original data set and generate synthetic historical weather data samples. This method not only increases the amount of data, but also introduces more diverse meteorological conditions, so that model training is no longer limited by the limitations of the original data and can learn a wider range of data distributions and patterns. The rich and diverse training data generated in this way helps to improve the generalization ability of the model, so that it can make more accurate and stable predictions under various meteorological conditions in actual operation, thereby enhancing the reliability and adaptability of the entire wind power prediction system.

[0084] In the embodiment of the present application, the above step S300 includes:

[0085] Specifically, performing empirical mode decomposition on the enhanced data set to obtain intrinsic mode components of different time scales includes:

[0086] For an original wind power sequence x(i), all large values ​​are determined as the upper envelope, and all small values ​​are determined as the lower envelope. m(i) represents the mean value of the upper envelope and the lower envelope. The first component is h1(i) = x(i)-m(i);

[0087] In the next round of screening, h1(i) is regarded as the original data, m1(i) is the mean value of the upper and lower envelopes of h1(i), and the second component h2(i) is determined;

[0088] The screening process is repeated n times until h n (i) is an intrinsic mode function or residual component r n (i) is a monotonic function, terminating the decomposition process;

[0089] Specify q1(i)=h1(i), q2(i)=h2(i), …, q k (i) = h k (i),, x(i), is finally decomposed into n eigenmode components q t (i) and a residual component r n (i) The formula is:

[0090]

[0091] Specifically, performing variational mode decomposition on the enhanced data set to obtain the wind turbine power trend series and period series includes:

[0092] Assume that the wind power data f is decomposed into i components, and the fraction-preserving decomposition sequence is a modal component with a limited bandwidth and an intermediate frequency. At the same time, the sum of the estimated bandwidths of each mode is minimized. The constraint condition is that the sum of all modes is equal to the original data. The corresponding constrained variational expression is as follows, where K is the number of modes to be decomposed, {u k},{ω k}, respectively corresponding to the kth modal component and the intermediate frequency after decomposition, δ(i), is the Lagrange function, and * is the convolution operator. Then solve the equation and introduce the Lagrange multiplication operator λ;

[0093]

[0094] Transforming the constrained variational problem into an unconstrained variational problem, the augmented Lagrange expression is obtained as follows, where ζ is a quadratic penalty factor, which is used to reduce the interference of Gaussian noise;

[0095]

[0096] The alternating direction multiplier iteration algorithm is used to continuously update each component and its frequency rate to obtain the saddle point of the unconstrained model, that is, the optimal solution of the original problem. All components can be obtained according to the rate space by the following formula, where ω represents the rate rate, They correspond to s(i),,τ(i),Rieter transform;

[0097]

[0098] ,yes , the residual after the dimensional filter, the algorithm re-estimates the repetition rate according to the power spectrum repetition rate of each component and initializes and n;

[0099] Execution cycle n=n+1, update when frequency ω>0 And update ω as follows k ,

[0100]

[0101] Repeat the above steps until the iteration stop condition is met.

[0102] It should be noted that the above step S300 can separate the long-term trend, periodic changes and other non-linear characteristics in the data, so as to better understand the dynamic characteristics of wind power. By decomposing the original signal into components with clear physical meanings, it not only helps to reveal the inherent laws hidden behind the data, but also provides more refined and targeted inputs for subsequent prediction models, so that the model can more accurately capture the change pattern of wind power, thereby improving the prediction accuracy and reliability.

[0103] In the embodiment of the present application, the above step S400 includes the following sub-steps D1-D4;

[0104] In D1: Figure 6 The STB module is constructed as shown;

[0105] Specifically, this module uses the window-based multi-head self-attention (W-MSA) technology, which effectively reduces the number of elements involved in a single calculation by dividing the input sequence into smaller windows and applying the self-attention mechanism independently within these windows, thereby reducing the computational burden;

[0106] Specifically, based on W-MSA, SW-MSA further moves the positions of these windows at different levels, allowing the model to capture broader contextual information while maintaining computational efficiency. This mechanism not only enhances the model's ability to express global spatial dependencies, but also enables the model to understand the spatial relationships in the image more carefully through the shift of local windows, making it possible to achieve more accurate dense predictions;

[0107] Specifically, the kernel frequency equation of the Swin Transformer block is as follows, where LN represents layer normalization (LayerNorm) and MLP is a multi-layer perceptron used to further process the data after the self-attention step. These equations describe how the Swin Transformer block processes the input sequence and captures the global spatial dependency through the self-attention mechanism;

[0108]

[0109] In D2: Figure 7 The SwinLSTM Cell is constructed based on the STB module.

[0110] Specifically, ConvLSTM overcomes the shortcomings of FC-LSTM in processing spatiotemporal data by introducing convolution operators into the transformation from input to state and from state to state. The key equation of ConvLSTM is as follows, where '*, represents convolution, I represents the Hadamard product. t ,,F,,O t , respectively input gate, forget gate and output gate;

[0111] I t =σ(W xi *X t +W hi *H t-1 +B i )

[0112] F t =σ(W xf *X t +W hf *H t-1 +B f )

[0113] O t =σ(W xo *X t +W ho *H t-1 +B o )

[0114]

[0115] Specifically, unlike the convolution operator that extracts local correlations, the self-attention mechanism captures global spatial dependencies by calculating the similarity scores between all positions. Therefore, we remove all weights W, and biases B, in the above equation and obtain the following equation, where I t =F t =O t Therefore, we divide the three gates: I t , F, O t Fusion into a gate, called filter gate R t ;

[0116] I t =F t =O t =σ(X t +H t-1 )

[0117]

[0118] Specifically, in SwinLSTM, the cell state C t , and the hidden state H t , the information level is updated evenly to capture long-term and short-term temporal dependencies. At the same time, the Swin Transformer block vertically learns global spatial dependencies. The key equation of SwinLSTM is as follows, where STB represents the Swin Transformer block and LP represents the linear projection;

[0119] R t =σ(STB(LP(X t ;H t-1 )))

[0120] C t =R t ⊙(tanh(STB(LP(X t ;H t-1 )))+C t-1 )

[0121] H t =R t ⊙tanh(C t )

[0122] In D3: Figure 8 As shown, the SwinLSTM-B network model is constructed with SwinLSTM Cell as the core frequency;

[0123] Specifically, the wind power data at time t is divided into non-overlapping data blocks with a length of δ, where δ is equal to 2 or 4. Therefore, the feature dimension of each patch is C·δ, where C represents the number of channels.

[0124] Specifically, the data blocks are flattened and input into the patch embedding layer, which maps the original features of the patch to an arbitrary dimension;

[0125] Specifically, the SwinLSTM layer receives the converted image patch and the hidden state H t-1 and cell status C t-1 To generate the hidden state H t and cell status C t .H t is copied into two copies, one for the reconstruction layer and the other for C t Together with the SwinLSTM layer for the next time step;

[0126] In D4: Use the SwinLSTM network model to process wind power data to obtain predicted wind turbine power generation.

[0127] It should be noted that the above step S400 combines the excellent global spatial dependency capture capability of Swin Transformer and the powerful performance of LSTM in processing time series data, so that complex spatiotemporal relationships can be modeled more accurately. By predicting environmental data and wind power separately, the model can not only reduce the prediction error, but also understand the various factors affecting wind power output in more detail. In addition, using data preprocessed by empirical mode decomposition and variational mode decomposition as input further enhances the model's ability to learn features of different time scales, ensuring the accuracy and stability of the prediction results.

[0128] In the embodiment of the present application, the above step S500 performs dimensionality reduction processing on the input variables through PCA, while improving the prediction accuracy, maintaining the calculation speed of the SwinLSTM network model, and overcoming the problem of overfitting, including:

[0129] Based on the wind power signal components obtained by variational mode decomposition, the standardized matrix of the original data is obtained, and the correlation coefficient matrix R = (r ij ) m×n ;

[0130] Calculate the eigenvalue matrix and the eigenvector corresponding to the eigenvalue, and calculate the contribution rate T i and the cumulative contribution rate η i , according to the cumulative contribution rate η i Determine the number of principal components to be selected, where the contribution rate T i and the cumulative contribution rate η i The calculation formula is:

[0131]

[0132] The reduced-dimensional data is obtained by selecting the eigenvalues ​​of the principal components and the corresponding eigenvectors.

[0133] It should be noted that the above step S500 effectively reduces the dimension of the data, eliminates redundant information and noise, and enables the model to focus on the most representative and explanatory features, thereby enhancing the generalization ability of the model. The reduced-dimensional data not only speeds up the training and prediction speed, but also ensures the efficiency and responsiveness of the model when processing large-scale data sets, while reducing the complexity, avoiding overfitting due to too many parameters, and ensuring the stable performance and accuracy of the model on new data.

[0134] In the embodiment of the present application, the above step S600 includes the following sub-steps F1-F5;

[0135] In F1: define basic parameters;

[0136] Specifically, reward R (Reward), each time (each step) interacts with the environment will get a reward, and the more rewards the better;

[0137] Specifically, the following is the action value function Q(s,a), whose input is the previous state and the action to be performed, and the output is how much value the action can bring. Therefore, a greedy method is to choose the action that can maximize Q(s,a). The dimension of Q(s,a) is equal to the dimension of the action space;

[0138]

[0139] Specifically, the following is the state value function V(s), which is the expectation of the Q function. Because the expected integral action eliminates the action A, the state value function V can be used to intuitively reflect the quality of a state. It is actually the weighted average of Q(s,a) for different a;

[0140]

[0141] Specifically, the following is the advantage function A, whose value is equal to the action value function minus the state value function, which is equivalent to the action value Q(s,a) minus its baseline;

[0142] A π (s,a)=Q π (s,a)-V π (S)

[0143] In F2: Construct a SAC network model to maximize the expected cumulative reward by optimizing the policy network and the value function network, while maximizing the entropy of the action distribution. The optimization objectives of the SAC algorithm are as follows:

[0144]

[0145] Among them, α is a regularization coefficient used to control the importance of entropy. The larger α is, the stronger the exploration is, which helps to accelerate the subsequent strategy learning; on the contrary, the smaller α is, the greater the possibility of the strategy falling into a worse local optimum. π represents the state distribution of policy π, where policy entropy It is expressed as follows:

[0146]

[0147] In F3: build the Actor network model;

[0148] Specifically, the Actor network consists of a fully connected DNN network. Its input is the system state s observed by the Agent. t , according to the state s t Output the value and standard deviation of the Gaussian distribution and perform sampling. The Actor network loss function obtained by KL divergence can be expressed as follows, where v represents the network parameters of the Actor network;

[0149]

[0150] Specifically, since the process of sampling actions according to Gaussian distribution is not differentiable, the expectation of the action is rewritten as the expectation of the noise by reparameterization, so the loss function of the Actor network is rewritten as follows, where represents the expected value obtained by sampling, b mean and b std are the value and variance of the Actor network output, and ⊙ represents the Hadamard product:

[0151]

[0152] In F4: Build the Critic network model;

[0153] Specifically, the Critic network consists of two Q networks and two target Q networks. The Critic network receives the action-state pairs (s t ,a t) as input, and the fully connected DNN network outputs an expected value as the evaluation of the action-state pair. The Critic network will select a network with a smaller Q value each time it uses the Q network, thereby alleviating the problem of too high Q value. The target Q network setting is used to improve the stability of the training frequency. The loss function of any Q network can be expressed as follows, where w represents the network parameters of the Q network, w - represents the network parameters of the target Q network, Β represents the experience replay pool, and γ is the discount factor:

[0154]

[0155] Specifically, the parameters of each target Q network are updated using the exponential moving average (EMA), which is given by the following formula, where 0 < ε < < 1 is an update factor. The EMA method allows the target Q network to follow the Q network in an evenly sliding manner, which helps to reduce the difference in updates and effectively avoid unstable fluctuations in the learning process:

[0156] w - ←εw+(1-ε)w -

[0157] In F5: Formulate a Markov Decision Process (MDP);

[0158] Specifically, the state space model is constructed. Before the agent makes a decision, it will first observe the meteorological data of the weather forecast and the actual meteorological data in the wind farm, including the wind direction and speed of the weather forecast and the wind direction and speed actually measured in the wind farm. The state set is defined as follows, where Respectively represent the wind speed and wind direction of the weather forecast. Respectively represent the actual wind speed and wind direction in the wind farm:

[0159]

[0160] Specifically, to construct the action space model, the agent’s decision is to assign weights to the prediction results of the sub-model. The action is defined as follows, where w 1 ,w 2 Represents the weight distribution of the prediction results of the two sub-models:

[0161] A={w 1 ,w 2}

[0162] Specifically, a reward model is constructed. In the proposed method, the reward is defined to reflect the effect of the weight. The prediction performance of each sub-model in the combined model is measured according to the value square error. The reward is defined as follows, where p 1 ,p 2Represents the prediction results of the two sub-models, p * Indicates the actual predicted power generation:

[0163]

[0164] It should be noted that the above step S600 constructs an adaptive weight adjustment method based on deep reinforcement learning soft actor-critic (SAC). By training the base module on the enhanced data set to convergence, it can dynamically optimize the prediction weights of each sub-model to ensure the flexibility and accuracy of the prediction strategy. The advantage of this method is that it allows the model to adaptively adjust according to the changing data environment, so as to better cope with the uncertainty and complexity in wind power prediction. Automatically learning the optimal weight distribution through deep reinforcement learning not only improves the prediction accuracy, but also enhances the robustness and adaptability of the system, so that the prediction model can maintain high performance under different conditions, effectively improving the reliability and practicality of wind power prediction.

[0165] In the embodiment of the present application, the above step S700 includes the following sub-steps G1-G3;

[0166] In G1: collect data on the actual physical system related to the wind farm, including wind turbine operation data, meteorological data (such as wind speed, wind direction, temperature, humidity, etc.), terrain data, etc.;

[0167] In G2: The theoretical wind turbine power generation is calculated based on the wind speed, air pressure, temperature and other data predicted by SwinLSTM. The motor power calculation method is as follows, where the air volume Q is in m 3 / h, the unit of total pressure P is Pa, the unit of power N is kW, η is the total pressure efficiency of the fan, and K is the motor capacity coefficient;

[0168] N=(Q / 3600)*P / (1000*η)*K

[0169] In G3: The weights of theoretical wind turbine power generation and predicted wind turbine power generation are adaptively adjusted through the SAC algorithm to predict future wind power generation.

[0170] It should be noted that the above step S700 not only improves the accuracy of the prediction, but also enhances the adaptability of the model to different environmental conditions, making the prediction more reliable and stable. In addition, adaptive weight adjustment helps to balance the relationship between model complexity and prediction performance, avoid overfitting or underfitting problems, thereby providing strong support for the stable operation and energy management of the power system, and ensuring the effective utilization and sustainable development of wind power resources.

[0171] The above is a schematic scheme of an LSTM-Transformer medium- and long-term wind power forecast combination model method of this embodiment. It should be noted that the technical solution of the LSTM-Transformer medium- and long-term wind power forecast combination model system and the technical solution of the above-mentioned LSTM-Transformer medium- and long-term wind power forecast combination model method belong to the same concept. The technical solution of the LSTM-Transformer medium- and long-term wind power forecast combination model system in this embodiment is not described in detail. For details, please refer to the description of the technical solution of the above-mentioned LSTM-Transformer medium- and long-term wind power forecast combination model method.

[0172] This embodiment also provides an LSTM-Transformer medium- and long-term wind power forecasting combined model system, including:

[0173] The data set processing module is used to obtain and preprocess the meteorological data of the wind farm, use the adversarial generative network to capture the intrinsic characteristics of the original data set, and generate synthetic historical weather data samples to expand the existing data set;

[0174] The signal decomposition module is used to perform empirical mode decomposition on the enhanced data set to obtain modal components of different time scales, and to perform variational mode decomposition on the enhanced data set to obtain the wind turbine power trend series and period series;

[0175] The training prediction module is used to build a SwinLSTM network model, and train the SwinLSTM network model using the data obtained from signal decomposition to predict the comprehensive environmental data of the wind turbine location and the wind power at the wind turbine location;

[0176] The result acquisition module is used to construct an adaptive weight adjustment method based on deep reinforcement learning flexible action evaluation, and to adaptively adjust the weights of the theoretical wind power and the predicted wind power to obtain the final predicted wind power.

[0177] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.

[0178] This embodiment also provides an electronic device, which includes a processor, a memory, a communication interface, a display screen and an input device connected through a system master. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a certain or non-certain manner, and the non-certain manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a LSTM-Transformer medium- and long-term wind power forecasting combined model method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0179] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method proposed in the above embodiment is implemented.

[0180] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0181] Through the above description of the implementation mode, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and can also be implemented by hardware, but in many cases the former is a better implementation mode. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer's floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform the methods of various embodiments of the present invention.

[0182] Example 2

[0183] Reference Figures 9 to 12Based on the previous embodiment, this embodiment provides an application example of an LSTM-Transformer medium- and long-term wind power forecasting combined model method and system, in order to verify and illustrate the technical effects adopted in this method.

[0184] In order to verify the effectiveness of the proposed algorithm, this embodiment was verified on a wind farm data set in Germany. The prediction time length was one month, the prediction event resolution was 15 minutes, and a probabilistic prediction was made for the wind power generation data for the next month. Fig. 9 The proposed LSTM-Transformer based on signal decomposition compares the predicted value and actual value of medium- and long-term wind power. Then the performance is compared with the mean absolute error (MAE) and root mean square error (RMSE) evaluation indicators. The LSTM-Transformer medium- and long-term wind power prediction combined model method based on signal decomposition is the algorithm proposed in this paper. As a comparison, here is provided Fig.10 The prediction results are obtained by training the dataset without DCGAN data augmentation.

[0185] From this, we can see that by using DCGAN to capture the intrinsic characteristics of the original data set, and then generating synthetic historical weather data samples to repair and expand the existing data set, the amount and diversity of data in the data set are greatly increased, which improves the randomness of sample learning, thereby improving the prediction accuracy and has a great impact on model training.

[0186] As a comparison, here is the prediction result diagram without signal decomposition and direct training, Fig.11 It can be seen that after the EMD and VMD decomposition, the environmental data and wind power data are predicted respectively, and then the weights are configured to calculate the final prediction results, which can enhance the robustness of the prediction and be more accurate when predicting that the wind power reaches the peak, thereby improving the prediction accuracy and having a great impact on model training.

[0187] In order to verify the prediction accuracy of the method of the present invention, the present invention is compared with four methods, and the evaluation indicators selected are RMSE and MAE. The experimental results of the comparison chart on the original data set are shown in Table 1:

[0188] Table 1: Performance table of each model method

[0189] method RMA % MSE% LSTM 22.87 20.07 Transformer 21.31 19.69 GRU 20.45 19.25 LSTM-CNN 21.29 18.96 LSTM-Transformer-Conb 19.95 17.11

[0190] From the results in Table 1, we can see that the proposed LSTM-Transformer medium- and long-term wind power forecasting combined model method based on signal decomposition has achieved the best performance, which is better than the single network optimization methods such as Transformer, GRU, and LSTM. The model method combining LSTM and CNN is also compared. The proposed combined model forecasting method can make near-optimal weight decisions and has better forecasting accuracy because it combines the two-aspect forecasting method after signal decomposition and reinforcement learning training. The error bar graphs of the five schemes are shown as follows: Fig.12 shown.

[0191] Taking the wind power scenario into consideration, this paper combines deep learning with signal decomposition technology, and uses the DCGAN module to repair and expand the existing data set. Through signal decomposition, starting from environmental data and wind power data, a LSTM-Transformer medium- and long-term wind power forecasting combined model method based on signal decomposition is proposed. This method aims to extract the intrinsic characteristics and patterns of wind power and corresponding environmental data through signal decomposition, and use deep learning algorithms to train the model prediction weights to optimize the prediction weights, and obtain a prediction model of wind power and wind power environment. In addition, an adaptive weight adjustment method based on deep reinforcement learning flexible action evaluation (SAC) is constructed, and the module is trained to convergence based on the enhanced data set. Finally, the theoretical wind power and the predicted wind power are adaptively adjusted to obtain the final predicted wind power. Thereby, accurate prediction of wind power is achieved. The research in this paper not only promotes the development of wind power forecasting technology, but also provides strong support for the stable operation and energy management of power systems.

[0192] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, and their values ​​should be included in the scope of the claims of the present invention.

[0193] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0194] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0195] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0196] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0197] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0198] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A LSTM-Transformer medium- and long-term wind power forecasting combined model method, characterized in that: include: Obtain and preprocess the meteorological data of the wind farm, use the generative adversarial network to capture the intrinsic characteristics of the original dataset, and generate synthetic historical weather data samples to expand the existing dataset; Perform empirical mode decomposition on the enhanced data set to obtain the intrinsic mode components of different time scales, and perform variational mode decomposition on the enhanced data set to obtain the wind turbine power trend series and period series; A SwinLSTM network model is constructed, and the SwinLSTM network model is trained using the data obtained by signal decomposition to respectively predict the comprehensive environmental data of the wind turbine location and the wind power at the wind turbine location; An adaptive weight adjustment method based on deep reinforcement learning flexible action evaluation is constructed, and the theoretical wind power and predicted wind power are adaptively adjusted to obtain the final predicted wind power.

2. The LSTM-Transformer medium- and long-term wind power forecasting combined model method according to claim 1, characterized in that: The generating of synthetic historical weather data samples to expand the existing data set includes: Use an embedding network to capture the characteristics of wind power time series data and map it to a low-dimensional latent space to retain the key information of the time series; Construct the generator G and discriminator D, where G(z) represents the generator function that maps the latent space vector z sampled from the standard normal distribution to the data space, and D(x) represents the probability that the discriminator network output x comes from the training data rather than the generator; The segmented data is input into DCGAN for training, the real wind power data label is 1, and the generated fake data label is 0 for adversarial training, and the performance of the discriminator is measured using binary cross entropy loss; The distribution of generated data is compared with that of real data, evaluated using the Kullback-Leibler divergence, and the model is fine-tuned using real wind power data.

3. The LSTM-Transformer medium- and long-term wind power forecasting combined model method according to claim 2, characterized in that: The step of performing empirical mode decomposition on the enhanced data set to obtain intrinsic mode components of different time scales includes: For an original wind power sequence x(i), all the large value points are determined as the envelope, and all the large and small points are determined as the lower envelope. m(i) represents the mean point of the envelope and the lower envelope. The first component is h1(i) = x(i)-m(i); In the next round of screening, h1(i) is regarded as the original data, m1(i) is the mean point of the lower envelope of h1(i), and is then determined as the second component h2(i); The screening process is repeated n times until h n (i) is an intrinsic mode function or residual component r n (i) is a monotonic function, terminating the decomposition process; It is represented by q1(i)=h1(i), q2(i)=h2(i),…,q k (i) = h k (i),, x(i), is finally decomposed into n eigenmode components q t (i) and a residual component r n (i).

4. The LSTM-Transformer medium- and long-term wind power forecasting combined model method according to claim 3, characterized in that: The step of performing variational mode decomposition on the enhanced data set to obtain a wind turbine power trend sequence and a periodic sequence comprises: Assume that the wind power data f is decomposed into i, components, and the fraction-preserving decomposition sequence is a modal component with a finite bandwidth and an intermediate frequency. At the same time, the sum of the estimated bandwidths of each mode is minimized. The constraint condition is that the sum of all modes is equal to the original data, and the constrained variational problem is transformed into an unconstrained variational problem. The alternating direction multiplier iteration algorithm is used to continuously update each component and its frequency to obtain the saddle point of the unconstrained model, that is, the optimal solution to the original problem. The algorithm re-estimates the repetition frequency according to the power spectrum repetition frequency of each component and initializes the parameters. Execute cycle n=n+1, and when the frequency ω>0, update each component and repeat the steps until the stop condition is met.

5. The LSTM-Transformer medium- and long-term wind power forecasting combined model method according to claim 4, characterized in that: The construction of the SwinLSTM network model includes: The construction of STB module includes adopting window-based multi-head self-attention technology to reduce the number of elements involved in a single calculation, capturing the information of the packet context by moving the window position at different levels, and inputting the input sequence into the normalization layer and multi-layer perception mechanism to capture the global spatial dependency; Building a SwinLSTM Cell based on the STB module includes capturing time dependency by introducing convolution operators into the input-to-state and state-to-state conversions, calculating similarity scores between all positions through a self-attention mechanism to capture global spatial dependency, and merging the input gate, forget gate, and output gate into a filter gate; combining the time dependency and the global spatial dependency to achieve average updating of the cell state and the hidden state to capture the long-term and short-term dependencies of the time packets; The SwinLSTM network model is constructed with the SwinLSTM Cell as the core frequency, including dividing the wind power data at time t into non-overlapping data blocks, flattening the data blocks and inputting them into the patch embedding layer, and the SwinLSTM layer receives the converted image patches, hidden states H t-1 and cell status C t-1 To generate the hidden state H t and cell status C t , H t is copied into two copies, one for the reconstruction layer and the other for C t Together with the SwinLSTM layer for the next time step; The SwinLSTM network model is used to process wind power data to obtain predicted wind turbine power generation.

6. The LSTM-Transformer medium- and long-term wind power forecasting combined model method according to claim 5, characterized in that: Also includes: Based on the wind power signal components obtained by variational mode decomposition, a standardized matrix of the original data is obtained, and the correlation coefficient matrix R of the standardized matrix is ​​calculated. ij ) m×n ; Calculate the feature point matrix and the eigenvectors corresponding to the feature points, and calculate the contribution rate T i and the cumulative contribution rate η i , according to the cumulative contribution rate η i is the number of principal components to be selected; The reduced-dimensional data is obtained by selecting the characteristic points of the principal components and the corresponding characteristic vectors.

7. The LSTM-Transformer medium- and long-term wind power forecasting combined model method according to claim 6, characterized in that: The acquisition of the final predicted wind power includes: Define basic parameters, including reward R, action price point function Q(s,a), state price point function V(s), and advantage function A; construct a SAC network model, maximize the expected cumulative reward by optimizing the strategy network and point function network, and maximize the entropy of action distribution; construct an Actor network model and a Critic network model, and formulate them as a Markov decision process to obtain a combined model prediction method based on deep reinforcement learning flexible action evaluation, and adaptively adjust the prediction weights of sub-models; Collecting data of actual physical systems related to the wind farm, and calculating theoretical wind turbine power generation according to the data predicted by the SwinLSTM network model; The weights of the theoretical wind turbine power generation and the predicted wind turbine power generation are adaptively adjusted through the SAC algorithm to achieve prediction of future wind power.

8. A system using the LSTM-Transformer medium- and long-term wind power forecasting combined model method as described in any one of claims 1 to 7, characterized in that: include: The data set processing module is used to obtain and preprocess the meteorological data of the wind farm, use the adversarial generative network to capture the intrinsic characteristics of the original data set, and generate synthetic historical weather data samples to expand the existing data set; The signal decomposition module is used to perform empirical mode decomposition on the enhanced data set to obtain modal components of different time scales, and to perform variational mode decomposition on the enhanced data set to obtain the wind turbine power trend series and period series; A training prediction module is used to construct a SwinLSTM network model, and train the SwinLSTM network model using the data obtained by signal decomposition to respectively predict the comprehensive environmental data of the wind turbine location and the wind power at the wind turbine location; The result acquisition module is used to construct an adaptive weight adjustment method based on deep reinforcement learning flexible action evaluation, and to adaptively adjust the weights of the theoretical wind power and the predicted wind power to obtain the final predicted wind power.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the LSTM-Transformer medium- and long-term wind power forecasting combined model method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the LSTM-Transformer medium- and long-term wind power forecasting combined model method according to any one of claims 1 to 7.

Images