Two-stage wind power prediction method based on physical mechanism and data driving

Abstract

The application provides a two-stage wind power prediction method and system based on physical mechanism and data driving, a storage medium and an electronic device, and relates to the field of wind power prediction. In the application, the physical mechanism of the evolution of the meteorological variables in the time and space dimensions is solved to derive the physical evolution data of each meteorological variable at multiple time steps, respectively. The fusion mechanism of the physical information in the model input and output two stages is designed, including fusing the observation data of each meteorological variable at the current time step with the physical evolution data at the same time step, and taking the observation data as the input of the corresponding meteorological variable prediction model; and fusing the prediction data of each meteorological variable at the next time step with the physical evolution data at the same time step to obtain the final prediction results of each meteorological variable, respectively. The application effectively improves the prediction accuracy of the model, improves the adaptability to sudden wind conditions, balances the prediction accuracy, stability and physical consistency, and has wide application and promotion value.

Description

Technical Field

[0001] This invention relates to the field of wind power prediction, specifically to a two-stage wind power prediction method, system, storage medium, and electronic device that combines physical mechanisms and data-driven approaches. Background Technology

[0002] With the continuous expansion of new energy installed capacity, wind power, as an intermittent and highly volatile renewable energy source, poses a severe challenge to grid dispatch and system stability due to its large-scale integration. Meteorological variables, including wind speed, temperature, and air pressure, largely determine the upper limit of wind power forecasting and directly affect the accuracy of power prediction. High-quality wind power forecasting not only helps improve grid integration and absorption capacity but also significantly reduces dispatch reserve costs, enhancing the economy and reliability of the power system. Constructing a wind power forecasting model with high accuracy and strong robustness has become a core requirement for wind farm operation and power system dispatch.

[0003] In related technologies, existing wind power prediction methods mainly fall into three categories: physical models, data-driven models, and physical-data hybrid driven models. In recent years, physical-data fusion-driven models have become a research hotspot, attempting to introduce physical knowledge into machine learning models to improve the accuracy and robustness of predictions. For example, patent CN118300102A discloses a method for predicting wind power based on a neural network driven by a hybrid mechanism and data approach; patent CN116307282A discloses a wind power prediction method and system based on a mechanism model and data-driven approach; and patent CN119294580A discloses an ultra-short-term wind farm power prediction method based on a dual physical and data-driven approach.

[0004] Although some methods have attempted to integrate physical models with data-driven approaches for wind power prediction, there are still many technical limitations, such as the limited ways to introduce physical information. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a two-stage wind power prediction method, system, storage medium, and electronic device that combines physical mechanisms and data-driven approaches, thus solving the technical problem of limited physical information introduction methods in existing technologies.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A two-stage wind power prediction method that combines physical mechanisms and data-driven approaches includes:

[0010] Acquire and preprocess meteorological variable datasets and wind turbine power datasets; wherein, the meteorological variable dataset includes observation data of each meteorological variable at continuous time steps, and the wind turbine power dataset includes wind turbine output power values ​​at continuous time steps and corresponding historical meteorological sequences;

[0011] Solve the physical mechanism of the evolution of pre-constructed meteorological variables in the spatiotemporal dimension, so as to derive the physical evolution data of each meteorological variable at multiple time steps;

[0012] The observation data of each meteorological variable at the current time step are fused with the physical evolution data at the same time step and used as input to the corresponding meteorological variable prediction model, so as to independently output the prediction data of each meteorological variable at the next time step.

[0013] The forecast data of each meteorological variable in the next time step are fused with the physical evolution data in the same time step to obtain the final forecast results of each meteorological variable.

[0014] The predicted meteorological sequence containing the prediction results of each meteorological variable is concatenated with the historical meteorological sequence in the wind turbine power dataset and used as the input of the wind turbine power prediction model. Combined with the wind turbine output power value in the wind turbine power dataset, the final wind turbine power prediction result is obtained.

[0015] Preferably, the meteorological variables include wind speed vector, air temperature and air pressure, and the wind speed vector includes wind speed components in the x and y directions on the horizontal plane and wind speed components in the z direction on the vertical plane.

[0016] Preferably, the physical mechanism of the evolution of meteorological variables in the spatiotemporal dimension refers to a set of partial differential equations under the framework of fluid dynamics.

[0017] Preferably, the partial differential equations are solved using the Fourier neural operator FNO, the graphical neural operator GNO, or the generalized integral neural operator GINO.

[0018] Preferably, the observation data of each meteorological variable at the current time step are linearly fused with the physical evolution data at the same time step, as shown below:

[0019]

[0020] Where Input represents the input, α1 and β1 are the first learnable parameters, and x t For any meteorological variable observation data at time step t, The physical evolution data of the same meteorological variable at time step t.

[0021] Preferably, the predicted meteorological variable data for the next time step are linearly fused with the physical evolution data for the same time step, as shown below:

[0022]

[0023] in, Let be the prediction result of any meteorological variable at time step t+1, and α2 and β2 be the second learnable parameters. For any meteorological variable prediction data at time step t+1, For the physical evolution data of the same meteorological variable at time step t+1;

[0024] Alternatively, the predicted meteorological variables for each time step can be nonlinearly fused with the physical evolution data for the same time step, as follows:

[0025]

[0026] Where MLP stands for Multilayer Perceptron, and [;] represents the stitching operation.

[0027] Preferably, both the meteorological variable prediction model and the wind turbine power prediction model are time series prediction models; wherein the meteorological variable prediction model or the wind turbine power prediction model is any one of a multilayer perceptron (MLP), a gated recurrent unit (GRU), a long short-term memory network (LSTM), or a Transformer model.

[0028] A wind power prediction system based on a two-stage fusion of physical mechanisms and data-driven approaches includes:

[0029] The data acquisition and preprocessing module is used to acquire and preprocess meteorological variable datasets and wind turbine power datasets; wherein, the meteorological variable dataset includes observation data of each meteorological variable at continuous time steps, and the wind turbine power dataset includes wind turbine output power values ​​at continuous time steps and corresponding historical meteorological sequences;

[0030] The physical deduction module is used to solve the physical mechanism of the evolution of pre-constructed meteorological variables in the spatiotemporal dimension, so as to derive the physical evolution data of each meteorological variable at multiple time steps.

[0031] The input fusion module is used to fuse the observation data of each meteorological variable at the current time step with the physical evolution data at the same time step, and use it as the input of the corresponding meteorological variable prediction model to independently output the prediction data of each meteorological variable at the next time step.

[0032] The output fusion module is used to fuse the predicted data of each meteorological variable in the next time step with the physical evolution data in the same time step to obtain the final prediction results of each meteorological variable.

[0033] The power prediction module concatenates the predicted meteorological sequence containing the prediction results of each meteorological variable with the historical meteorological sequence in the wind turbine power dataset, and uses this as the input to the wind turbine power prediction model. Combined with the wind turbine output power value in the wind turbine power dataset, the final wind turbine power prediction result is obtained.

[0034] A storage medium storing a computer program for wind power prediction based on a two-stage fusion of physical mechanisms and data-driven approaches, wherein the computer program causes a computer to execute the two-stage wind power prediction method as described above.

[0035] An electronic device, comprising:

[0036] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing the two-stage wind power prediction method as described above.

[0037] (III) Beneficial Effects

[0038] This invention provides a two-stage wind power prediction method, system, storage medium, and electronic device that combines physical mechanisms and data-driven approaches. Compared with existing technologies, it has the following advantages:

[0039] This invention solves for the physical mechanism of the spatiotemporal evolution of pre-constructed meteorological variables to derive the physical evolution data of each meteorological variable at multiple time steps. A physical information fusion mechanism is designed in both the input and output stages of the model to achieve collaborative modeling of physical information and data-driven models throughout the entire process. This includes fusing the observation data of each meteorological variable at the current time step with the physical evolution data at the same time step, and using this as input to the corresponding meteorological variable prediction model; and fusing the prediction data of each meteorological variable at the next time step with the physical evolution data at the same time step to obtain the final prediction results for each meteorological variable. This invention effectively improves the prediction accuracy of the model and enhances its adaptability to sudden wind conditions, balancing prediction accuracy, stability, and physical consistency, and has broad application and promotion value. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1A block diagram illustrating a two-stage wind power prediction method that combines physical mechanisms and data-driven approaches, as provided in an embodiment of the present invention.

[0042] Figure 2 The flowchart illustrates a two-stage wind power prediction method that combines physical mechanisms and data-driven approaches, as provided in this embodiment of the invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] This application provides a two-stage wind power prediction method, system, storage medium, and electronic device that combines physical mechanisms and data-driven approaches. This solves the technical problem of limited physical information introduction methods in the prior art and fully utilizes physical priors to improve the adaptability of the hybrid driving model.

[0045] The technical solution in this application is to solve the above-mentioned technical problems, and the general idea is as follows:

[0046] 1. Deep integration of physical information: Design a two-stage physical information introduction mechanism at the input and output ends to promote the integration of physical prior information with the entire process of time series modeling, and enhance the model's ability to characterize the evolution of variables.

[0047] 2. Improve physical modeling efficiency: By using neural operators to replace traditional partial differential equation numerical solvers, the function mapping relationship of meteorological variables from initial state to future evolution is learned, significantly reducing computational costs and meeting the real-time requirements of forecasting.

[0048] 3. Achieve learnable adjustment of fusion ratio: Introduce trainable fusion parameters to optimize the contribution of physical information and observation data in input and output, thereby enhancing the model's expressive power and interpretability.

[0049] 4. Enhance the generalization ability and adaptability of physical modeling: By introducing neural operators to model physical processes, the dependence on fixed model structures and boundary conditions is weakened, thereby improving generalization performance and application flexibility.

[0050] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0051] Example 1:

[0052] like Figure 1 As shown, this embodiment of the invention provides a two-stage wind power prediction method that combines physical mechanisms and data-driven approaches, characterized by comprising:

[0053] S1. Acquire and preprocess meteorological variable dataset and wind turbine power dataset; wherein, the meteorological variable dataset includes observation data of each meteorological variable at continuous time steps, and the wind turbine power dataset includes wind turbine output power values ​​at continuous time steps and corresponding historical meteorological sequences;

[0054] S2. Solve the physical mechanism of the evolution of pre-constructed meteorological variables in the spatiotemporal dimension, so as to derive the physical evolution data of each meteorological variable at multiple time steps respectively;

[0055] S3. The observation data of each meteorological variable at the current time step is fused with the physical evolution data at the same time step and used as the input of the corresponding meteorological variable prediction model to independently output the prediction data of each meteorological variable at the next time step.

[0056] S4. The predicted meteorological variables for each time step are fused with the physical evolution data for the same time step to obtain the final predicted results for each meteorological variable.

[0057] S5. The predicted meteorological sequence containing the prediction results of each meteorological variable is concatenated with the historical meteorological sequence in the wind turbine power dataset and used as the input of the wind turbine power prediction model. Combined with the wind turbine output power value in the wind turbine power dataset, the final wind turbine power prediction result is obtained.

[0058] This invention effectively improves the prediction accuracy of the model and enhances its adaptability to sudden wind conditions, taking into account prediction accuracy, stability and physical consistency, and has broad application and promotion value.

[0059] like Figure 2 As shown, Figure 2 A flowchart of a two-stage wind power prediction method that combines physical mechanisms and data-driven approaches is disclosed. Figure 2 In the middle, x and x ′ Let y be a meteorological variable dataset and a wind turbine power dataset, respectively, and let y and y be denoted as y. ′ The corresponding prediction results.

[0060] Next, we will combine Figure 2 The steps of the above scheme are described in detail:

[0061] In step S1, a meteorological variable dataset and a wind turbine power dataset are acquired and preprocessed; wherein, the meteorological variable dataset includes observation data of each meteorological variable at continuous time steps, and the wind turbine power dataset includes wind turbine output power values ​​at continuous time steps and corresponding historical meteorological sequences.

[0062] This step utilizes a sliding window mechanism to construct and acquire a dataset of meteorological variables and a dataset of wind turbine power. For example, the meteorological variables include wind speed vectors. The temperature T and air pressure p are given, and the wind speed vector includes wind speed components u and v in the x and y directions on the horizontal plane, and wind speed component w in the z direction on the vertical plane.

[0063] Accordingly, the meteorological variable dataset includes observation data of wind speed vector, temperature and air pressure at continuous time steps, and the wind turbine power dataset includes wind turbine output power values ​​at continuous time steps and corresponding historical meteorological sequences (historical observation sequences of wind speed vector, temperature and air pressure).

[0064] To ensure the stability of model training, this step also involves preprocessing the original data. This preprocessing includes a normalization process, where each variable is normalized using a maximum-minimum method, mapping it to the [0,1] interval. The normalization formula is as follows:

[0065]

[0066] Where, x max and x min These represent the maximum and minimum values ​​of the corresponding variables, respectively.

[0067] In step S2, the physical mechanism of the evolution of pre-constructed meteorological variables in the spatiotemporal dimension is solved to derive the physical evolution data of each meteorological variable at multiple time steps.

[0068] This invention pre-constructs the physical mechanism of the evolution of meteorological variables in the spatiotemporal dimension, specifically referring to a set of partial differential equations (PDEs) under the framework of fluid dynamics, including:

[0069]

[0070] Equation (2) represents the mass conservation under incompressible conditions, where u, v, and w are the wind speed components in the horizontal and vertical directions, respectively; Equation (3) is the momentum conservation equation, where ρ is the fluid density and μ is the dynamic viscosity. Equation (4) represents the advection evolution of temperature under adiabatic conditions; Equation (5) represents the pressure evolution under the quasi-static approximation.

[0071] Based on this, this step uses the Fourier neural operator FNO, the graph neural operator GNO, or the generalized integral neural operator GINO to solve the physical mechanism of the evolution of pre-constructed meteorological variables in the spatiotemporal dimension, so as to derive the physical evolution data of each meteorological variable at multiple time steps.

[0072] As mentioned above, this step uses neural operators to replace traditional numerical solvers for partial differential equations, learns the functional mapping relationship of meteorological variables from the initial state to the future evolution, significantly reduces computational costs, and balances physical consistency, modeling efficiency and scalability, making it suitable for integration with data-driven models.

[0073] Taking the Fourier Neural Operator (FNO) as an example, it can learn the mapping relationships in the function space. Its basic implementation process includes: projecting the input function into the Fourier frequency domain, using frequency domain convolution operations to capture the global correlation of the physical field, and then restoring it to the original space via inverse Fourier transform to generate the physical solution of the target variable at a future time step. This structure possesses stronger generalization ability and modeling efficiency. This step utilizes FNO to efficiently solve the aforementioned partial differential equation system, weakening the dependence on fixed model structure and boundary conditions, and improving generalization performance and application flexibility.

[0074] It should be noted that this step generates physical evolution values ​​for multiple consecutive time steps through an iterative approach. The predicted output of each step serves as the input for the next step, thus realizing the multi-step evolution of meteorological variables.

[0075] Subsequently, this embodiment of the invention designs a two-stage physical information introduction mechanism at both the input and output ends, promoting the integration of prior physical information with the entire temporal modeling process, and enhancing the model's ability to characterize the variable evolution process. See steps S3-S4 for details:

[0076] In step S3, the observation data of each meteorological variable at the current time step are fused with the physical evolution data at the same time step and used as the input of the corresponding meteorological variable prediction model to independently output the prediction data of each meteorological variable at the next time step.

[0077] like Figure 2 As shown, this step linearly fuses the observation data of each meteorological variable at the current time step with the physical evolution data at the same time step, as expressed as:

[0078]

[0079] Where Input represents the input, x t For any meteorological variable observation data at time step t, The fusion mechanism represents the physical evolution data of the same meteorological variable at time step t; α1 and β1 are the first learnable parameters, which can take negative values ​​to adjust the relative weights of observation and physical information. When the parameters are negative, this fusion mechanism can be regarded as a differential enhancement form, similar to differential coding, which can improve the flexibility and density of data representation.

[0080] It should be noted that each meteorological variable (wind speed vector, temperature, air pressure) is fused independently here to form multiple fused feature sequences, which provide a basis for subsequent independent predictions.

[0081] Next, this step models the fused meteorological variable series to independently output the predicted data for each meteorological variable at the next time step. For example, the meteorological variable prediction model here uses a time series prediction model, which can be any of the following: Multilayer Perceptron (MLP), Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), or Transformer model.

[0082] Taking MLP as an example, perform single-step or multi-step variable prediction:

[0083] h (l) =σ(W (l) h (l-1) +b (l) (7)

[0084] Among them, h (l) W represents the output of the l-th layer. (l) and b (l) Let represent the weights and biases of the l-th layer, respectively, and σ be a non-linear activation function, such as ReLU.

[0085] More specifically, the MLP model can adopt a two-layer hidden layer structure (Input→64→64→Output, ReLU activation), where Input is the fusion result of formula (6), which maps the historical time steps of the input to obtain the predicted value.

[0086] It should be noted that since wind speed vector, air pressure and temperature differ in numerical scale and variation characteristics, three separate models are created to predict the different fused feature sequences obtained here, in order to improve accuracy and stability.

[0087] In step S4, the predicted data of each meteorological variable in the next time step are fused with the physical evolution data in the same time step to obtain the final prediction results of each meteorological variable.

[0088] like Figure 2 As shown, this step introduces the physical evolution prediction results generated by the neural operator into the model output. Compared with network model prediction results To integrate.

[0089] Understandably, variables Physical evolution data of meteorological variables at time step t+1 obtained from step S2 To ensure consistent parameter representation, it is replaced with...

[0090] Furthermore, the output fusion can specifically employ any of the following fusion methods:

[0091] The predicted meteorological variables for the next time step are linearly fused with the physical evolution data for the same time step, as shown below:

[0092]

[0093] in, For any meteorological variable predicted at time step t+1, For any meteorological variable prediction data at time step t+1, The physical evolution data of the same meteorological variable at time step t+1; α2 and β2 are the second learnable parameters, whose values ​​directly represent the weight of the prediction proportion.

[0094] Alternatively, the predicted meteorological variables for each time step can be nonlinearly fused with the physical evolution data for the same time step, as follows:

[0095]

[0096] Where MLP stands for Multilayer Perceptron, and [;] represents the stitching operation.

[0097] In addition, the meteorological variable forecast results are inversely normalized:

[0098] y = y norm ·(x max -x min )+x min (10)

[0099] In step S5, the predicted meteorological sequence containing the prediction results of each meteorological variable is concatenated with the historical meteorological sequence in the wind turbine power dataset and used as the input of the wind turbine power prediction model. Combined with the wind turbine output power value in the wind turbine power dataset, the final wind turbine power prediction result is obtained.

[0100] like Figure 2 As shown, this step first concatenates historical meteorological sequences with predicted meteorological sequences (including normalized predictions of wind speed vectors, temperature, and air pressure at one or more time steps), and uses this as input to the wind turbine power prediction model. For example, the meteorological variable prediction model here also uses a time series prediction model, which can also be any of the following: Multilayer Perceptron (MLP), Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), or Transformer model. Taking MLP as an example, this MLP model uses a three-layer hidden layer structure (Input→64→64→32→Output, ReLU activation). The concatenation result of the historical and predicted meteorological sequences is used to obtain the final wind turbine power prediction result.

[0101] It is understood that the training processes of the three meteorological variable prediction models in this embodiment of the invention are independent of each other. In an optional implementation, the training process of each model uses early stopping and learning rate scheduling to ensure convergence and performance stability, and adopts the same configuration as follows:

[0102] Loss function: Calculate mean squared error (MSE); Optimizer: Adam optimizer; Early stopping: Terminate training early when the validation set loss does not decrease significantly within 30 consecutive batches; Learning rate scheduling: The initial learning rate is set to 1e-3, and cosine annealing is used to dynamically adjust it during training.

[0103] Thus, this embodiment of the invention completes the entire process of a two-stage wind power prediction method that combines physical mechanisms and data-driven approaches. This method has significant advantages over existing technologies, and experimental results further verify its superiority, including improved single-step prediction accuracy, enhanced adaptability to sudden wind changes and maximum wind conditions, improved multi-step prediction accuracy, and improved power prediction accuracy.

[0104] Example 2:

[0105] This invention provides a wind power prediction system based on a two-stage fusion of physical mechanisms and data-driven approaches, comprising:

[0106] The data acquisition and preprocessing module is used to acquire and preprocess meteorological variable datasets and wind turbine power datasets; wherein, the meteorological variable dataset includes observation data of each meteorological variable at continuous time steps, and the wind turbine power dataset includes wind turbine output power values ​​at continuous time steps and corresponding historical meteorological sequences;

[0107] The physical deduction module is used to solve the physical mechanism of the evolution of pre-constructed meteorological variables in the spatiotemporal dimension, so as to derive the physical evolution data of each meteorological variable at multiple time steps.

[0108] The input fusion module is used to fuse the observation data of each meteorological variable at the current time step with the physical evolution data at the same time step, and use it as the input of the corresponding meteorological variable prediction model to independently output the prediction data of each meteorological variable at the next time step.

[0109] The output fusion module is used to fuse the predicted data of each meteorological variable in the next time step with the physical evolution data in the same time step to obtain the final prediction results of each meteorological variable.

[0110] The power prediction module concatenates the predicted meteorological sequence containing the prediction results of each meteorological variable with the historical meteorological sequence in the wind turbine power dataset, and uses this as the input to the wind turbine power prediction model. Combined with the wind turbine output power value in the wind turbine power dataset, the final wind turbine power prediction result is obtained.

[0111] Example 3:

[0112] This invention provides a storage medium storing a computer program for wind power prediction based on a two-stage fusion of physical mechanisms and data-driven approaches. The computer program causes a computer to execute the two-stage wind power prediction method as described in Embodiment 1.

[0113] Example 4:

[0114] This invention provides an electronic device, comprising:

[0115] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing a two-stage wind power prediction method as described in Example 1.

[0116] It is understood that the wind power prediction system, storage medium and electronic device based on the fusion of physical mechanism and data-driven dual-stage provided in the embodiments of the present invention correspond to the wind power prediction method based on the synergy of physical mechanism and data-driven dual-stage provided in the embodiments of the present invention. The explanation, examples and beneficial effects of the relevant contents can be referred to the corresponding parts of the method, and will not be repeated here.

[0117] In summary, compared with existing technologies, it has the following beneficial effects:

[0118] 1. A lightweight physical extrapolation method for predicting meteorological variables is proposed. It uses Fourier neural operators to efficiently approximate the solution of partial differential equations, taking into account physical consistency, modeling efficiency and scalability, and is suitable for integration with data-driven models.

[0119] 2. A two-stage physical fusion architecture is proposed, which fuses the PDE physical evolution results of the corresponding time steps at the input and output ends respectively, so as to align physical information with observation data in the time dimension and improve modeling capabilities.

[0120] 3. Design learnable fusion weights to adjust the ratio of physical and observational data, and achieve quantitative analysis of the relative contributions of physical information and data-driven models at different time steps, thereby enhancing model interpretability. This fusion mechanism and its interpretation method should be protected.

[0121] 4. To address the differences in scale and evolution of different meteorological variables, a fusion strategy of multivariate independent modeling and differentiation is proposed to improve forecast accuracy.

[0122] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0123] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A two-stage wind power prediction method that combines physical mechanisms and data-driven approaches, characterized in that, include: Acquire and preprocess meteorological variable datasets and wind turbine power datasets; wherein, the meteorological variable dataset includes observation data of each meteorological variable at continuous time steps, and the meteorological variables include wind speed vector, temperature and air pressure; the wind turbine power dataset includes wind turbine output power values ​​at continuous time steps and corresponding historical meteorological sequences; The physical mechanism of the evolution of pre-constructed meteorological variables in the spatiotemporal dimension is solved by using neural operators to derive lightweight physical evolution data of each meteorological variable at multiple time steps. The observation data of each meteorological variable at the current time step are fused with the physical evolution data at the same time step and used as input to the corresponding meteorological variable prediction model, so as to independently output the prediction data of each meteorological variable at the next time step. The forecast data of each meteorological variable in the next time step are fused with the physical evolution data in the same time step to obtain the final forecast results of each meteorological variable. The predicted meteorological sequence containing the prediction results of each meteorological variable is concatenated with the historical meteorological sequence in the wind turbine power dataset and used as the input of the wind turbine power prediction model. Combined with the wind turbine output power value in the wind turbine power dataset, the final wind turbine power prediction result is obtained.

2. The two-stage wind power prediction method as described in claim 1, characterized in that, The wind speed vector includes wind speed components in the x and y directions on the horizontal plane and wind speed components in the z direction on the vertical plane.

3. The two-stage wind power prediction method as described in claim 1, characterized in that, The physical mechanism of the evolution of meteorological variables in the spatiotemporal dimension refers to the partial differential equations under the framework of fluid dynamics.

4. The two-stage wind power prediction method as described in claim 3, characterized in that, The partial differential equations are solved using Fourier neural operator FNO, graphical neural operator GNO, or generalized integral neural operator GINO.

5. The two-stage wind power prediction method as described in claim 1, characterized in that, The observation data of each meteorological variable at the current time step are linearly fused with the physical evolution data at the same time step, as shown below: in, Indicates input, , As the first learnable parameter, For any meteorological variable observation data at time step t, The physical evolution data of the same meteorological variable at time step t.

6. The two-stage wind power prediction method as described in claim 1, characterized in that, The predicted meteorological variables for the next time step are linearly fused with the physical evolution data for the same time step, as shown below: in, For any meteorological variable predicted at time step t+1, , As the second learnable parameter, For any meteorological variable prediction data at time step t+1, For the physical evolution data of the same meteorological variable at time step t+1; Alternatively, the predicted meteorological variables for each time step can be nonlinearly fused with the physical evolution data for the same time step, as follows: in, It is a multilayer perceptron. This is for splicing operations.

7. The two-stage wind power prediction method as described in claim 1, characterized in that, Both the meteorological variable prediction model and the wind turbine power prediction model adopt time series prediction models; wherein the meteorological variable prediction model or the wind turbine power prediction model is any one of the following: multilayer perceptron (MLP), gated recurrent unit (GRU), long short-term memory network (LSTM), or Transformer model.

8. A wind power prediction system based on a dual-stage fusion of physical mechanisms and data-driven approaches, characterized in that, include: The data acquisition and preprocessing module is used to acquire and preprocess meteorological variable datasets and wind turbine power datasets; wherein, the meteorological variable dataset includes observation data of each meteorological variable at continuous time steps, and the meteorological variables include wind speed vector, temperature and air pressure; the wind turbine power dataset includes wind turbine output power values ​​at continuous time steps and corresponding historical meteorological sequences; The physics derivation module is used to solve the physical mechanism of the evolution of pre-constructed meteorological variables in the spatiotemporal dimension using neural operators, so as to derive lightweight physical evolution data of each meteorological variable at multiple time steps. The input fusion module is used to fuse the observation data of each meteorological variable at the current time step with the physical evolution data at the same time step, and use it as the input of the corresponding meteorological variable prediction model to independently output the prediction data of each meteorological variable at the next time step. The output fusion module is used to fuse the predicted data of each meteorological variable in the next time step with the physical evolution data in the same time step to obtain the final prediction results of each meteorological variable. The power prediction module concatenates the predicted meteorological sequence containing the prediction results of each meteorological variable with the historical meteorological sequence in the wind turbine power dataset, and uses this as the input to the wind turbine power prediction model. Combined with the wind turbine output power value in the wind turbine power dataset, the final wind turbine power prediction result is obtained.

9. A storage medium, characterized in that, It stores a computer program for wind power prediction based on a two-stage fusion of physical mechanisms and data-driven approaches, wherein the computer program causes the computer to execute the two-stage wind power prediction method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing the two-stage wind power prediction method as described in any one of claims 1 to 7.

Images