A wind power prediction method considering wake effect in extreme scenarios

Abstract

This invention relates to the field of wind power prediction technology, and more specifically, to a wind power prediction method considering wake effects in extreme scenarios. The method includes: generating virtual data from sample data in extreme scenarios; the sample data and the virtual data constituting an expanded dataset; constructing a wake correction model based on the expanded dataset and outputting a dynamic hypergraph of wake associations; inputting the dynamic hypergraph into a pre-constructed wind power prediction model; extracting spatiotemporal correlation features; training the wind power prediction model based on the expanded dataset; and predicting wind power using the trained wind power prediction model. This invention reduces the amount of training data required for the model. Compared to conventional wake models, the wake correction model reduces the wake radius prediction error and improves power prediction accuracy. It is particularly suitable for predicting wind power in extreme scenarios such as typhoons and strong turbulence, and can serve as a basis for grid connection and dispatch decisions.

Description

Technical Field

[0001] This invention relates to the field of wind power prediction technology, and more specifically, to a wind power prediction method that considers wake effects in extreme scenarios. Background Technology

[0002] As wind power accounts for an increasing proportion of the energy mix, accurate wind power forecasting is crucial for the safe and stable operation of the power grid. Due to the wake effect, upstream wind turbines can influence downstream turbines. Ignoring the wake effect significantly reduces the accuracy of wind power forecasting, thus affecting grid dispatching and grid connection processes. Accurate wind power forecasting requires sufficient sample data. However, in some scenarios, sufficient sample data is unavailable. For example, in extreme weather conditions (such as typhoons and strong turbulence), wind farms often shut down due to equipment protection shutdowns, resulting in scarce SCADA (Supervisory Control and Data Acquisition) data. Newly commissioned wind farms, due to their short operating time, also face the "small sample problem" of insufficient wake effect-related data.

[0003] When there is insufficient sample size, the existing wind power prediction methods have poor generalization ability, and the absence of data may cause the model to fail. The model cannot learn the relationship between wake and power, resulting in insufficient accuracy in wind power prediction. Summary of the Invention

[0004] To overcome the problem of insufficient wind power prediction accuracy in extreme scenarios in the prior art, this invention provides a wind power prediction method that considers wake effects in extreme scenarios.

[0005] To address the aforementioned technical problems, the present invention employs the following technical solution: a wind power prediction method considering wake effects in extreme scenarios, comprising: generating virtual data from sample data under extreme scenarios, wherein the sample data and the virtual data constitute an expanded dataset; constructing a wake correction model based on the expanded dataset and outputting a dynamic hypergraph of wake association; inputting the dynamic hypergraph into a pre-constructed wind power prediction model; extracting spatiotemporal correlation features; training the wind power prediction model based on the expanded dataset; and predicting wind power using the trained wind power prediction model.

[0006] In the technical solution of this invention, a wind power prediction method considering wake effect in extreme scenarios is provided. A wake correction model is constructed based on sample data and virtual data, which reduces the amount of training data required for the model. It is especially suitable for power prediction in small sample scenarios. Compared with conventional wake models, the wake correction model reduces the wake radius prediction error and improves the power prediction accuracy. It is especially suitable for predicting wind power in extreme scenarios such as typhoons and strong turbulence, and can be used as a basis for grid connection and dispatch.

[0007] Furthermore, the method includes the following steps: S1: Obtain sample data from the wind farm. When the sample data belongs to an extreme scenario, input the sample data into the physical information to generate the adversarial network Physics-GAN and output virtual data. S2: The sample data and the virtual data constitute an expanded dataset. The conventional wake model is modified based on the expanded dataset to obtain a wake correction model. S3: Input the expanded dataset into the wake correction model, use the wake correction model to calculate the wake influence relationship between wind turbines, and construct a dynamic hypergraph characterizing the wake association accordingly; S4: Input the dynamic hypergraph into the pre-constructed wind power prediction model, extract the spatiotemporal correlation features, and train the wind power prediction model based on the expanded dataset to obtain the trained wind power prediction model. S5: Input the real-time data of the wind farm to be predicted into the trained wind power prediction model, and output the predicted wind power value.

[0008] Furthermore, the sample data includes meteorological data, wind turbine data, wake characteristic data, and basic wind farm data.

[0009] Furthermore, step S1 also includes determining whether an extreme scenario is being experienced based on wind speed and turbulence intensity.

[0010] Furthermore, in step S1, the method further includes setting a first preset wind speed and a second preset wind speed. When the measured wind speed is between the first preset wind speed and the second preset wind speed, it is determined to be an extreme wind speed. A turbulence intensity threshold is set. When the measured turbulence intensity is greater than or equal to the turbulence intensity threshold, it is determined to be extreme turbulence.

[0011] Furthermore, in step S1, the method further includes establishing extreme scenario wake physics rules based on physical deviation terms. The physical deviation terms include at least one of typhoon scenario deviation rules, strong turbulence scenario deviation rules, and strong sandstorm scenario deviation rules. The extreme scenario wake physics rules are used by the Physics-GAN network to output the virtual data.

[0012] Furthermore, in step S2, the conventional wake model is modified in accordance with the wake physics rules for extreme scenarios.

[0013] Furthermore, in step S3, the process specifically includes treating each wind turbine in the wind farm as a node and the wake association as a hyperedge, thereby constructing a dynamic hypergraph that evolves over time.

[0014] Furthermore, the conventional wake model adopts the Jenson wake model.

[0015] Furthermore, in step S4, the wind power prediction model adopts a source domain pre-training-target domain fine-tuning transfer learning framework, which uses approximate mature wind farm data as the source domain to transfer to the target domain under extreme scenarios for transfer learning, and adjusts the target domain based on the virtual data.

[0016] Compared with the prior art, the beneficial effects of the present invention are: The wind power prediction method considering wake effect in extreme scenarios of the present invention constructs a wake correction model based on sample data and virtual data, which reduces the amount of training data required for the model and is especially suitable for power prediction in small sample scenarios. Compared with conventional wake models, the wake correction model reduces the wake radius prediction error and improves the power prediction accuracy. It is especially suitable for predicting wind power in extreme scenarios such as typhoons and strong turbulence, and can be used as a basis for grid connection and dispatch. Attached Figure Description

[0017] Figure 1 This is a flowchart of the wind power prediction method considering wake effect in extreme scenarios according to the present invention; Figure 2 This is a power prediction curve of a set of wind turbines according to an embodiment of the present invention. Detailed Implementation

[0018] The accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent. To better illustrate this embodiment, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings. The positional relationships described in the drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0019] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "long," and "short" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present patent. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0020] The technical solution of the present invention will be further described in detail below through specific embodiments and with reference to the accompanying drawings: Example 1 This embodiment discloses a wind power prediction method considering wake effects in extreme scenarios, comprising: generating virtual data from sample data in extreme scenarios, wherein the sample data and the virtual data constitute an expanded dataset; constructing a wake correction model based on the expanded dataset and outputting a wake-related output dynamic hypergraph; inputting the dynamic hypergraph into a pre-constructed wind power prediction model; extracting spatiotemporal correlation features; training the wind power prediction model based on the expanded dataset; and predicting wind power using the trained wind power prediction model.

[0021] In this embodiment, a wake correction model is constructed based on sample data and virtual data, reducing the amount of training data required for the model. This is particularly suitable for power prediction in scenarios with small sample sizes. Compared to conventional wake models, the wake correction model reduces the wake radius prediction error and improves power prediction accuracy. It is especially suitable for predicting wind power in extreme scenarios such as typhoons and strong turbulence, and can serve as a basis for grid connection and dispatch. The model accuracy of this solution remains consistently stable, and the physical rule base and dynamic hypergraph are visualized, allowing dispatchers to intuitively understand the physical basis of the prediction results, greatly improving the model's interpretability.

[0022] For more specific details, please refer to Figure 1 The wind power prediction method considering wake effect in an extreme scenario according to this embodiment includes the following steps: S1: Obtain sample data from the wind farm. When the sample data belongs to an extreme scenario, input the sample data into the physical information to generate the adversarial network Physics-GAN and output virtual data. S2: The sample data and the virtual data constitute an expanded dataset. Based on the expanded dataset, the conventional wake model is modified to obtain a wake correction model. S3: Input the expanded dataset into the wake correction model, use the wake correction model to calculate the wake influence relationship between wind turbines, and construct a dynamic hypergraph characterizing the wake association accordingly; S4: Input the dynamic hypergraph into the pre-constructed wind power prediction model, extract the spatiotemporal correlation features, and train the wind power prediction model based on the expanded dataset to obtain the trained wind power prediction model. S5: Input the real-time data of the wind farm to be predicted into the trained wind power prediction model, and output the predicted wind power value.

[0023] In this embodiment, the sample data includes meteorological data, wind turbine data, wake characteristic data, and basic wind farm data.

[0024] In step S1, sample data can be collected using existing monitoring equipment at the wind farm and preprocessed. In some embodiments, for severe sandstorm scenarios, if the wind farm is not equipped with a dedicated sandstorm concentration sensor, the sandstorm concentration Cs can be calculated using empirical formulas based on visibility data from a meteorological station, or the real-time sandstorm monitoring data released by the local meteorological bureau can be directly accessed. The data preprocessing method includes: removing outliers and filtering valid data based on wind power physics rules; mapping meteorological data, wind turbine data, and wake data to the same spatiotemporal coordinate system for data alignment; and performing Min-Max normalization on numerical features such as wind speed, power, and wake radius to normalize the data to the [0~1] interval.

[0025] In this embodiment, step S1 further includes determining whether it is an extreme scenario based on wind speed and turbulence intensity.

[0026] In this embodiment, step S1 further includes setting a first preset wind speed and a second preset wind speed. When the measured wind speed is between the first preset wind speed and the second preset wind speed, it is determined to be an extreme wind speed. A turbulence intensity threshold is set. When the measured turbulence intensity is greater than or equal to the turbulence intensity threshold, it is determined to be extreme turbulence.

[0027] For example, the first preset wind speed is 15 m / s, and the second preset wind speed is 25 m / s. The wind speed *v* is obtained from the sample data. When 15 m / s ≤ *v* ≤ 25 m / s, the fan is in a high wind speed range. In this case, the fan will not stop but will operate at reduced power, which is considered an extreme wind speed. When the wind speed exceeds the second preset wind speed, it needs to be shut down for safety. The turbulence intensity threshold is 0.3. When the turbulence intensity threshold is greater than or equal to 0.3, it is considered extreme turbulence.

[0028] It should be noted that the typhoon scenario defined in this invention does not refer to the scenario when the eye of a typhoon passes through, but rather to the situation where the wind speed of the wind turbine increases under the influence of a typhoon but does not reach the cut-out wind speed. In this range, the wind turbine actively reduces its power through active pitch control to protect the equipment, but still generates electricity by connecting to the grid.

[0029] In this embodiment, step S1 further includes establishing extreme scenario wake physics rules based on physical deviation terms. The physical deviation terms include at least one of typhoon scenario deviation rules, strong turbulence scenario deviation rules, and strong sandstorm scenario deviation rules. The extreme scenario wake physics rules are used by the Physics-GAN to output the virtual data.

[0030] Among them, the physical deviation term is derived from fluid mechanics theory and a small amount of CFD simulation data. The CFD simulation data is obtained by computational fluid dynamics simulation, which uses numerical methods to perform high-fidelity simulation and analysis of fluid flow, heat transfer, mass transfer and related physicochemical processes.

[0031] The deviation rules for typhoon scenarios, strong turbulence scenarios, and strong sandstorm scenarios in this embodiment will be explained in detail below.

[0032] Typhoon scenario deviation rule: The direction of wake diffusion shifts with the angular velocity ω of the typhoon vortex, and the shift amount is... θ is positively correlated with ω, where ω is measured by the weather station. The wake attenuation coefficient k is corrected from the conventional land value of 0.075 to 0.12~0.15.

[0033] Deviation rules for strong turbulence scenarios: The wake attenuation coefficient k' increases with the increase of turbulence intensity I, and the wake radius growth rate increases by 20%~30%. The correction formula is as follows: , k'=k×(1+λ1I), where I is the turbulence intensity, x is the distance between the upstream and downstream wind turbines, λ1 is the turbulence correction factor, r is the wake radius at the downstream distance x, and r0 is the initial wake radius. In the calculation, r0 can be taken as the radius of the wind turbine rotor.

[0034] Severe dust storm scenario: Taking the air temperature at 15℃ as an example, the air density ρ air The value was revised from 1.225 kg / m³ to 1.35~1.45 kg / m³, and the wind turbine thrust coefficient C... T Reduce by 5%~8%, correct formula ,in This refers to the concentration of dust, measured in mg / m³. Regarding air density ρ... air The incremental value indicates that in a severe sandstorm scenario, the fluid flowing through the wind turbine blades is no longer pure air, but a gas-solid mixture of air and sand particles. According to the gas-solid two-phase flow theory, the equivalent density of the mixture is equal to the sum of the air phase density and the sand phase density. In specific calculations, the base air density should be calculated based on real-time temperature and air pressure, and then the density increment due to the sand concentration should be added. ρ. The essence of the correction is the superposition of the mass of sand and dust.

[0035] In this embodiment, step S1 may specifically include the following steps: S11: Obtain sample data from the wind farm and calculate the corresponding physical deviation term under extreme scenarios.

[0036] S12: Construct a three-module GAN architecture consisting of a generator, a discriminator, and a physics constraint unit. The generator uses a one-dimensional U-Net network structure. Its inputs are the preprocessed sample data from step S1 and extreme weather parameters. The outputs are wake feature data and the corresponding wind turbine power sequence. The wake feature data includes the wake radius, wind speed attenuation rate, and wake influence range. The generator loss function incorporates a physics constraint term, Lphysics, with the formula: L... total =L GAN +λ2L physics , where L GAN The adversarial loss is λ2=0.6, the physical constraint weights are λ2=0.6, and Lphysics is the physical deviation term between the virtual data and step S21. The discriminator uses a CNN (convolutional neural network) to distinguish between the sample data in step S1 and the virtual data output by the generator, with the goal of minimizing the classification error. The physical constraint term is embedded in step S21 to verify the physical rationality of the generated data. The training iteration uses the Adam optimizer with a learning rate of 0.001 and 500-800 iterations until the physical pass rate of the generated data is ≥90%, at which point training stops.

[0037] S13: Select virtual data that the physical constraint judges as "qualified" from the generator output, and merge them with the sample data measured in step S1 to form an expanded dataset of sample data and virtual data, in which sample data accounts for 10%~30%, virtual data accounts for 70%~90%, and the dataset size is ≥1000 records.

[0038] In this embodiment, in step S2, the conventional wake model is modified in conjunction with the wake physics rules for extreme scenarios. Specifically, this includes: modifying the wake model for extreme scenarios, based on the Jensen wake model and incorporating the wake physics rules for extreme scenarios, to modify the wake wind speed attenuation formula as follows: , where C T’ denoted as the thrust coefficient corrected for extreme scenarios, k' as the wake attenuation coefficient corrected for extreme scenarios, u as the corrected actual wind speed at the downstream turbine, u0 as the incoming wind speed at the upstream turbine, and x as the straight-line distance between the upstream and downstream turbines.

[0039] Example 2 This embodiment is similar to Embodiment 1, see reference. Figure 1 The wind power prediction method considering wake effect in an extreme scenario according to this embodiment includes the following steps: S1: Obtain sample data from the wind farm. When the sample data belongs to an extreme scenario, input the sample data into the physical information to generate the adversarial network Physics-GAN and output virtual data. S2: The sample data and the virtual data constitute an expanded dataset. The conventional wake model is modified based on the expanded dataset to obtain a wake correction model. S3: Input the expanded dataset into the wake correction model, use the wake correction model to calculate the wake influence relationship between wind turbines, and construct a dynamic hypergraph characterizing the wake association accordingly; S4: Input the dynamic hypergraph into the pre-constructed wind power prediction model, extract the spatiotemporal correlation features, and train the wind power prediction model based on the expanded dataset to obtain the trained wind power prediction model. S5: Input the real-time data of the wind farm to be predicted into the trained wind power prediction model, and output the predicted wind power value.

[0040] In this embodiment, step S3 specifically includes treating each wind turbine in the wind farm as a node and the wake association as a hyperedge, constructing a dynamic hypergraph that evolves over time. More specifically, step S3 includes the following steps: S31: Each wind turbine in the wind farm is considered a node. The node feature matrix W includes: historical power of the turbine, corrected wake wind speed u, and extreme weather parameters. The historical power of the turbine is taken from the first three time points, and the extreme weather parameters include wind speed and turbulence intensity.

[0041] S32: Definition and weight allocation of the superedge. If the downstream wind turbine is within the wake influence range of the upstream wind turbine, the wake influence range is calculated using the modified Jensen wake model. The formula for the wake influence range is: , C is the threshold for wake influence intensity. T’ Here, k' is the thrust coefficient corrected for extreme scenarios, and k' is the wake attenuation coefficient corrected for extreme scenarios. The upstream wind turbine and all downstream wind turbines affected by its wake are grouped into a single superedge, and weights are assigned based on the wake influence intensity. The formula for calculating the wake influence intensity is: When α∈[0.4~1], the weight is 1.0, which is defined as a strong association; when α∈[0.1~0.4], the weight is 0.6, which is defined as a medium association; and when α∈[0.08~0.1], the weight is 0.3, which is defined as a weak association.

[0042] S33: The hyperedge structure is adjusted every 15 minutes based on real-time wind direction and turbulence intensity. If the downstream wind turbine enters or leaves the wake influence range of the upstream wind turbine, the corresponding hyperedge connection relationship and weight are updated to form a dynamic hypergraph sequence.

[0043] In this embodiment, the conventional wake model adopts the Jensen wake model. The Jensen wake model is a simplified physical model of the wake effect of wind turbines widely used in the field of wind energy engineering. It is used to quickly estimate the impact of the wake generated by upstream wind turbines on the power generation of downstream wind turbines in a wind farm. This invention obtains a wake correction model based on the optimization of the Jensen wake model, which can improve the accuracy of wake effect calculation, and is especially suitable for extreme scenarios.

[0044] Example 3 This embodiment is similar to Embodiment 2, please refer to... Figure 1 The wind power prediction method considering wake effect in an extreme scenario according to this embodiment includes the following steps: S1: Obtain sample data from the wind farm. When the sample data belongs to an extreme scenario, input the sample data into the physical information to generate the adversarial network Physics-GAN and output virtual data. S2: The sample data and the virtual data constitute an expanded dataset. The conventional wake model is modified based on the expanded dataset to obtain a wake correction model. S3: Input the expanded dataset into the wake correction model, use the wake correction model to calculate the wake influence relationship between wind turbines, and construct a dynamic hypergraph characterizing the wake association accordingly; S4: Input the dynamic hypergraph into the pre-constructed wind power prediction model, extract the spatiotemporal correlation features, and train the wind power prediction model based on the expanded dataset to obtain the trained wind power prediction model. S5: Input the real-time data of the wind farm to be predicted into the trained wind power prediction model, and output the predicted wind power value.

[0045] In this embodiment, in step S4, the wind power prediction model adopts a source domain pre-training-target domain fine-tuning transfer learning framework, which uses approximate mature wind farm data as the source domain to transfer to the target domain under extreme scenarios for transfer learning, and adjusts the target domain based on the virtual data.

[0046] More specifically, step S4 includes the following steps: S41: Source Domain Model Pre-training: Select mature wind farm data similar to the target domain wind farm type, including wake features, turbine power, and extreme scenario meteorological parameters; construct a DHGCN-CNN-BIGRU prediction model, inputting the dynamic hypergraph information constructed in S3 into the DHGCN module for feature extraction, extracting dynamic wake correlation features between wind turbines under extreme scenarios. CNN is used to extract wind power sequences, and both processes are performed simultaneously to extract spatiotemporal feature changes. The spatiotemporal features output by DHGCN + the power time-series features extracted by CNN are input into the BIGRU module for prediction training, fitting the power change pattern. The model is trained using source domain data, freezing the bottom convolutional layers of DHGCN, training for 300 epochs, with the loss function being RMSE, until the source domain prediction RMSE ≤ 2.5%.

[0047] S42: Target Domain Small Sample Fine-tuning: An augmented dataset composed of virtual data and sample data is used as fine-tuning data. By unfreezing the top layer of DHGCN and the BiGRU layer, only the parameters of these two layers are updated. The learning rate is 0.001 and iterates for 50 to 100 rounds to avoid overfitting. L2 regularization and early stopping mechanism are introduced. Training is stopped if the validation set loss does not decrease for 5 consecutive rounds to ensure the model's generalization ability under small sample conditions.

[0048] In this embodiment, the DHGCN-CNN-BiGRU training model is adopted. DHGCN-CNN-BiGRU is a hybrid neural network model that integrates dynamic heterogeneous graph convolution, convolutional neural network and bidirectional gated recurrent unit. The DHGCN dynamic heterogeneous graph convolutional network module is used to extract high-order and dynamic spatial dependency features from complex, variable and multi-type association systems. The CNN convolutional neural network module is used to extract local spatial features in regular gridded data. The BiGRU bidirectional gated recurrent unit is used to extract temporal dynamic features from sequence data.

[0049] Compared with the prior art, the beneficial effects of the present invention are: The wind power prediction method considering wake effect in extreme scenarios of the present invention constructs a wake correction model based on sample data and virtual data, which reduces the amount of training data required for the model and is especially suitable for power prediction in small sample scenarios. Compared with conventional wake models, the wake correction model reduces the wake radius prediction error and improves the power prediction accuracy. It is especially suitable for predicting wind power in extreme scenarios such as typhoons and strong turbulence, and can be used as a basis for grid connection and dispatch.

[0050] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A wind power prediction method considering wake effect in extreme scenarios, characterized in that: The process includes generating virtual data from sample data under extreme scenarios, the sample data and the virtual data constituting an expanded dataset, constructing a wake correction model based on the expanded dataset and outputting a dynamic hypergraph of wake association, inputting the dynamic hypergraph into a pre-constructed wind power prediction model, extracting spatiotemporal correlation features, training the wind power prediction model based on the expanded dataset, and predicting wind power through the trained wind power prediction model.

2. The wind power prediction method considering wake effect in extreme scenarios according to claim 1, characterized in that: The method includes the following steps: S1: Obtain sample data from the wind farm. When the sample data belongs to an extreme scenario, input the sample data into the physical information to generate the adversarial network Physics-GAN and output virtual data. S2: The sample data and the virtual data constitute an expanded dataset. The conventional wake model is modified based on the expanded dataset to obtain a wake correction model. S3: Input the expanded dataset into the wake correction model, use the wake correction model to calculate the wake influence relationship between wind turbines, and construct a dynamic hypergraph characterizing the wake association accordingly; S4: Input the dynamic hypergraph into the pre-constructed wind power prediction model, extract the spatiotemporal correlation features, and train the wind power prediction model based on the expanded dataset to obtain the trained wind power prediction model. S5: Input the real-time data of the wind farm to be predicted into the trained wind power prediction model, and output the predicted wind power value.

3. The wind power prediction method considering wake effect in extreme scenarios according to claim 1, characterized in that: The sample data includes meteorological data, wind turbine data, wake characteristic data, and basic data of wind farms.

4. The wind power prediction method considering wake effect in extreme scenarios according to claim 2, characterized in that: Step S1 also includes determining whether an extreme scenario is being experienced based on wind speed and turbulence intensity.

5. The wind power prediction method considering wake effect in extreme scenarios according to claim 4, characterized in that: In step S1, a first preset wind speed and a second preset wind speed are set. When the measured wind speed is between the first preset wind speed and the second preset wind speed, it is determined to be an extreme wind speed. A turbulence intensity threshold is set. When the measured turbulence intensity is greater than or equal to the turbulence intensity threshold, it is determined to be extreme turbulence.

6. The wind power prediction method considering wake effect in extreme scenarios according to claim 2, characterized in that: In step S1, the method further includes establishing extreme scenario wake physics rules based on physical deviation terms. The physical deviation terms include at least one of typhoon scenario deviation rules, strong turbulence scenario deviation rules, and strong sandstorm scenario deviation rules. The extreme scenario wake physics rules are used by the Physics-GAN network to output the virtual data.

7. The wind power prediction method considering wake effect in extreme scenarios according to claim 6, characterized in that: In step S2, the conventional wake model is modified in accordance with the wake physics rules for extreme scenarios.

8. The wind power prediction method considering wake effect in extreme scenarios according to claim 2, characterized in that: In step S3, the specific steps include treating each wind turbine in the wind farm as a node and the wake association as a hyperedge, and constructing a dynamic hypergraph that evolves over time.

9. The wind power prediction method considering wake effect in extreme scenarios according to claim 2, characterized in that: The conventional wake model uses the Jenson wake model.

10. The wind power prediction method considering wake effect in extreme scenarios according to claim 1, characterized in that: The wind power prediction model adopts a source domain pre-training-target domain fine-tuning transfer learning framework, which uses approximate mature wind farm data as the source domain to transfer to the target domain under extreme scenarios for transfer learning, and adjusts the target domain based on the virtual data.

Images