A method for generating offshore wind power output scenarios considering the influence of typhoon

Abstract

The application relates to a typhoon-influenced offshore wind power output scene generation method, which comprises the following steps: considering the correlation between various meteorological factors and wind power output, adopting a Pearson correlation coefficient to reduce meteorological factors that influence wind power output under typhoon weather; according to the characteristics that an offshore wind power output curve is offset with the time sequence of typhoon, based on the characteristics of the reduced period factors, a K-means clustering based on DTW distance measurement is used to construct wind turbine output categories under different typhoon influences, and a labeled original data set is constructed; a W-ACGAN network is used to expand the labeled original data set, and a labeled offshore wind power output scene during a typhoon is constructed; according to wind power output prediction data during a typhoon, the constructed offshore wind power output scene during a typhoon is subjected to similarity screening, and a short-term offshore wind power output similar scene set under typhoon weather is obtained. Compared with the prior art, the application has the advantages of high accuracy of the generated scene.

Description

Technical Field

[0001] This invention relates to the field of offshore wind power generation technology, and in particular to a method for generating offshore wind power output scenarios that takes into account the impact of typhoons. Background Technology

[0002] Offshore wind power is one of the fastest-growing renewable energy sources in the power system. The coastal areas of my country where offshore wind power clusters are developing are also regions severely affected by tropical cyclones in the Northwest Pacific. Typhoon weather is one of the major natural disasters affecting offshore wind power. The randomness of typhoon intensity and path evolution makes it difficult to accurately describe offshore wind power output scenarios. Researching methods for generating offshore wind power output scenarios under typhoon weather is of great significance for quantifying the uncertainty of wind power output under complex offshore conditions and optimizing power system planning and operation.

[0003] Currently, there is considerable research on methods for generating wind power output scenarios, mainly including probabilistic model methods, classical scenario methods, and deep learning generation methods. Among these, deep learning generation methods have strong generalization and data representation capabilities, and offer advantages such as unsupervised learning and the ability to learn autonomously. While all of the above methods can effectively generate wind power output scenarios, methods for constructing offshore wind power output scenarios under the uncertain influence of typhoon weather require further research. Summary of the Invention

[0004] The purpose of this invention is to address the problem of insufficient data in describing wind power output scenarios under different typhoon impact levels, and to provide a method for generating offshore wind power output scenarios that takes into account the impact of typhoons. This method learns the characteristics of wind power output under typhoon weather and improves the accuracy of generating short-term offshore wind power output scenario sets.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A method for generating offshore wind power output scenarios that take into account the impact of typhoons includes the following steps:

[0007] Step 1) Consider the correlation between various meteorological factors and wind power output, and use the Pearson correlation coefficient to simplify the meteorological factors that affect wind power output during typhoon weather;

[0008] Step 2) In view of the characteristic that the offshore wind power output curve shifts with the time sequence of typhoons, based on the reduced term factor characteristics, K-means clustering with dynamic time planning distance metric is used to construct wind turbine output categories under different typhoon influences, and build a labeled original dataset.

[0009] Step 3) The W-ACGAN network with Wasserstein distance added is used to augment the labeled original dataset to construct a labeled offshore wind power output scenario during typhoons.

[0010] Step 4) Based on the wind power output prediction data during typhoons, perform similarity screening on the constructed offshore wind power output scenarios during typhoons to obtain a set of similar scenarios for short-term offshore wind power output under typhoon weather.

[0011] Further, step 1) specifically involves: analyzing the correlation between meteorological factors and wind power output using the Pearson correlation coefficient, and selecting meteorological factors with a strong correlation to wind power output to construct a dataset of relevant meteorological characteristics under typhoon weather, wherein the correlation between meteorological factors and wind power output is I. pcc (P,Q i ) is represented as:

[0012]

[0013] Among them, P and Q i These are wind power output and meteorological factors i and P, respectively. k and Q ik P and Q respectively i The k-th value in; and P and Q respectively i The mean of ; n is the total number.

[0014] Further, step 2) specifically involves: employing an improved k-means clustering method, utilizing dynamic time planning distance as a metric for inter-cluster spacing, and leveraging its ability to allow for temporal scaling during cluster matching to obtain sequences with the most similar morphology possible. The silhouette coefficient is then used as an evaluation method to select the optimal number of clusters. This process is applied to cluster the raw offshore wind power output data affected by typhoon timing.

[0015] DTW(A,B)={Dist(a,b)+min[D(i-1,j),D(i,j-1),D(i-1,j-1)]}

[0016] SC = (ba) / max{b,a}

[0017] Wherein, DTW(A,B) represents the dynamic time planning distance of the offshore wind power output time series under the influence of two typhoon time series, i,j=1,2,·····n are different time points; D(0,0)=0,D(i,0)=D(0,i)=+∞; b is the minimum average distance between the wind turbine output sample and other cluster samples; a is the average distance between the sample and samples within the cluster; the silhouette coefficient SC ranges from [-1,+1], and the closer SC is to 1, the better the clustering quality.

[0018] Further, step 3) specifically involves: putting the labeled original dataset obtained in step 2) into the W-ACGAN network framework for training, introducing a gradient penalty function into the discriminator function so that the discriminator can use the Wasserstein distance to calculate the difference between the generated data and the real data, thereby realizing data augmentation and generation of offshore wind power output scenarios during typhoons.

[0019] Furthermore, the Wasserstein distance is expressed as:

[0020]

[0021] Among them, W(P) r ,P g ) represents the Wasserstein distance between the actual output and the generated output data, ∏(P) r ,P g Actual output data distribution P r With the distribution of generated output data P g The set of joint probability distributions; inf denotes the infimum; y is the generated output sample; ||·| is the distance between the actual output sample and the generated output sample; E represents the expectation of the distance between the actual output distribution and the generated output distribution.

[0022] Furthermore, considering that the Wasserstein distance is difficult to calculate directly, the Kantorovich Rubinstein dual form of the Wasserstein distance is used to describe the distance between generated samples and real samples:

[0023]

[0024] Among them, W(P) r ,P g ) denotes the Kantorovich-Rubinstein dual form of the Wasserstein distance, sup denotes the supremum, and ‖f‖ L ≤1 indicates that the distribution difference functions f(x) and f(y) should satisfy 1-Lipschitz continuity, and the upper bound of the absolute value of their derivatives is 1.

[0025] Furthermore, the objective function of the discriminator, which incorporates a gradient penalty function, is:

[0026]

[0027] Among them, P x P represents the actual wind turbine output sample distribution. z To generate a sample distribution of wind turbine output, D represents the discriminator function, L S L represents the probability of classifying data as true.C This represents the probability of correctly classifying the data. The penalty function represents the gradient of the loss function of the discriminator D. Let λ denote the gradient, and λ be the penalty coefficient.

[0028] Furthermore, the wind power output prediction data during the typhoon is determined based on the wind speed during the typhoon and the active power output model of the wind turbine, wherein the active power output model of the wind turbine is modeled using a cubic function.

[0029] Furthermore, the active power output model of the wind turbine is expressed as follows:

[0030]

[0031] Among them, P w,j δ represents the active power output of the j-th wind turbine; w,j This represents the output coefficient of the j-th wind turbine. V represents the maximum active power output of the j-th wind turbine when it is operating at full capacity; V, V in V e V out These represent the actual wind speed, cut-in wind speed, rated wind speed, and cut-out wind speed of the wind turbine during a typhoon, respectively.

[0032] Further, step 4) specifically involves: based on the wind power output prediction data during the typhoon, performing typhoon impact degree label discrimination, and using PCC similarity to screen similar scenarios. In descending order, multiple samples with strong similarity to the wind power output during the typhoon to be predicted are selected from the corresponding label generated samples to form a set of similar scenarios for short-term offshore wind power output under typhoon weather.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] To address the problem of generating offshore wind power output scenarios under the uncertain influence of typhoon weather, this invention proposes a method for generating short-term offshore wind power output scenarios based on an improved auxiliary classification generative adversarial network. This method constructs an original dataset of output-related features under typhoon weather influence, divides the dataset according to the degree of typhoon impact on wind power output, and utilizes an improved auxiliary classification generative adversarial network for data augmentation and output scenario generation. This improves the accuracy of generating offshore wind power output scenarios under typhoon influence, providing a reference for quantifying the uncertainty of wind power output under complex offshore conditions and optimizing power system planning and operation.

[0035] This study first considers the correlation and coupling between various meteorological factors and wind power output. This paper uses the Pearson correlation coefficient to analyze the relevant meteorological factors and constructs a meteorological feature scene set under typhoon weather to improve the generation quality of subsequent power output scenes.

[0036] Secondly, considering the characteristic that the shape of the power output curve shifts with the timing of typhoons, the k-means clustering method with DTW distance metric is adopted. By leveraging its ability to allow for scaling of time features during class matching, sequences with the most similar shapes can be obtained. In this scenario, the inter-class features are significantly different after clustering, achieving better clustering results.

[0037] Finally, adding Wasserstein distance to the auxiliary classification generative adversarial network can effectively solve the training difficulties and pattern collapse problems, and can better fit the probability distribution, generating a large number of higher quality labeled output scenarios to supplement the imbalanced dataset for short-term output scenario analysis. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0039] Figure 2 This is a schematic diagram of a dynamic time-bending path;

[0040] Figure 3 This is a schematic diagram illustrating the training process of a W-ACGAN-based offshore wind power output scenario generation model.

[0041] Figure 4 A schematic diagram showing the SC values ​​for different numbers of clusters;

[0042] Figure 5 This is a typical wind turbine output scenario under the influence of a typhoon in one embodiment;

[0043] Figure 6 To compare the loss of the improved W-ACGAN with that of the original ACGAN, (6a) represents the change in the generation loss of the original ACGAN, (6b) represents the change in the classification loss of the original ACGAN, (6c) represents the change in the generation loss of the improved ACGAN, and (6d) represents the change in the classification loss of the improved ACGAN.

[0044] Figure 7 This is a set of similar scenarios for short-term offshore wind power output during typhoon weather in one embodiment.

[0045] Figure 8 The confidence intervals of W-ACGAN and MC in the generation output scenario are compared, where (8a) is the short-term output scenario set based on MC and (8b) is the short-term output scenario set based on W-ACGAN. Detailed Implementation

[0046] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0047] This embodiment provides a method for generating offshore wind power output scenarios that take into account the impact of typhoons, such as... Figure 1 As shown, it includes the following steps:

[0048] Step 1) Considering the correlation between various meteorological factors and wind power output, the Pearson correlation coefficient is used to simplify the meteorological factors that affect wind power output during typhoon weather.

[0049] Considering the correlation between various meteorological factors and wind power output, feature reduction of relevant meteorological factors is performed to improve the accuracy of subsequent typhoon impact classification and wind power output scenarios. The Pearson correlation coefficient (PCC) is used to analyze the correlation between meteorological factors and wind power output, and meteorological factors with strong correlation to wind power output are selected to construct a relevant feature dataset for typhoon weather.

[0050]

[0051] Formula (1) represents the correlation between wind power output and meteorological factors, where P and Q are the two factors. i These are wind power output and meteorological factors i and P, respectively. k and Q ik P and Q respectively i The k-th value in; and P and Q respectively i The mean of ; n is the total number.

[0052] Step 2) In view of the characteristic that the offshore wind power output curve shifts with the time sequence of typhoons, based on the reduced time factor characteristics, K-means clustering with dynamic time planning distance metric is used to construct wind turbine output categories under different typhoon influences, and a labeled original dataset is constructed.

[0053] An improved k-means clustering method is employed, utilizing Dynamic Time Warping (DTW) distance as a metric for inter-class spacing. By leveraging DTW's ability to scale temporal features during class matching, sequences with the most similar morphologies are obtained. A schematic diagram of the dynamic time warping path is shown below. Figure 2 As shown.

[0054] Considering that k-means clustering requires setting an appropriate number of clusters k, intra-cluster similarity increases with the number of clusters, but excessive clusters reduce inter-cluster differences, the silhouette coefficient (SC) is used as an evaluation method to select the optimal number of clusters, thereby clustering the raw offshore wind power output data affected by typhoon timing.

[0055] DTW(A,B)={Dist(a,b)+min[D(i-1,j),D(i,j-1),D(i-1,j-1)]} (2)

[0056] SC=(ba) / max{b,a} (3)

[0057] Formula (2) represents the DTW distance of the offshore wind power output time series under the influence of two typhoons. i,j=1,2,·····n are different time points; D(0,0)=0,D(i,0)=D(0,i)=+∞; b is the minimum average distance between the wind turbine output sample and other cluster samples; a is the average distance between the sample and the samples within the cluster; the silhouette coefficient SC ranges from [-1,+1], and the closer SC is to 1, the better the clustering quality.

[0058] Step 3) Use the W-ACGAN network with added Wasserstein distance to augment the labeled original dataset and construct a labeled offshore wind power output scenario during typhoons.

[0059] The labeled raw dataset obtained in step 2) is fed into the W-ACGAN network framework for training. The difference between generated and real data is calculated using Wasserstein distance, which effectively solves the training difficulties and mode collapse problem and can better fit the probability distribution. Considering that Wasserstein distance is difficult to calculate directly, its Kantorovich Rubinstein dual form is used to describe the distance between generated samples and real samples.

[0060]

[0061]

[0062] Wherein, formula (4) represents the Wasserstein distance between the actual output and the generated output data, ∏(P r ,P g Actual output data distribution P r With the distribution of generated output data P gThe set of joint probability distributions; inf denotes the infimum; y is the generated output sample; ‖·‖ is the distance between the actual output sample and the generated output sample; E represents the expectation of the distance between the actual output distribution and the generated output distribution. Formula (5) represents the Kantorovich Rubinstein dual form of the Wasserstein distance, sup denotes the supremum, ‖f‖ L ≤1 indicates that the distribution difference functions f(x) and f(y) should satisfy 1-Lipschitz continuity, and the upper bound of the absolute value of their derivatives is 1.

[0063] Introducing a gradient penalty function for function D within its domain ensures that the discriminator function approximately satisfies 1-Lipschitz continuity, enabling the discriminator loss function to effectively describe the Wasserstein distance. Based on this, an improved auxiliary classification adversarial generative adversarial network is used for data augmentation and generation of offshore wind power output scenarios during typhoons.

[0064]

[0065] Wherein, formula (6) is the objective function of W-ACGAN after adding the gradient penalty function D, P x P represents the actual wind turbine output sample distribution. z To generate a sample distribution of wind turbine output, D(·) represents the discriminator function, L S L represents the probability of classifying data as true. C This represents the probability of correctly classifying the data. The penalty function represents the gradient of the loss function of the discriminator D. Let λ represent the gradient, and λ be the penalty coefficient.

[0066] The training process of the W-ACGAN network is as follows: Figure 3 As shown.

[0067] Step 4) Based on the wind power output prediction data during typhoons, perform similarity screening on the constructed offshore wind power output scenarios during typhoons to obtain a set of similar scenarios for short-term offshore wind power output under typhoon weather.

[0068] The wind power output during a typhoon is obtained by combining the wind turbine active power output model with the wind speed V during the typhoon. A cubic function is used to model the active power output of the wind turbine. Based on the predicted wind power output during the typhoon, the degree of typhoon impact is labeled, and PCC is used to screen similar scenarios. Following the order from largest to smallest, N samples with strong similarity to the predicted wind power output during the typhoon are selected from the sample generated by this label, forming a set of similar scenarios for short-term offshore wind power output under the influence of a typhoon.

[0069]

[0070] Among them, P w,j δ represents the active power output of the j-th wind turbine; w,j This represents the output coefficient of the j-th wind turbine. V represents the maximum active power output of the j-th wind turbine when it is operating at full capacity; V, V in V e V out These represent the actual wind speed, cut-in wind speed, rated wind speed, and cut-out wind speed of the wind turbine during a typhoon, respectively. Considering actual conditions, this embodiment sets the cut-in wind speed, rated wind speed, and cut-out wind speed to 3, 14, and 25 m / s, respectively.

[0071] This embodiment uses the wind turbine output and meteorological data of 30 wind turbine units (2012-2017) of a certain offshore wind farm in my country during typhoons as the original dataset.

[0072] (1) Construction of the original dataset of relevant features under different typhoon impact levels

[0073] Correlation analysis of wind power output was conducted using meteorological factors wind speed, temperature, humidity, and pressure. The specific PEN correlation coefficients between wind power and each characteristic are shown in the table below.

[0074] Table 1. Correlation coefficients between wind power and various characteristics (person coefficients)

[0075] variable wind speed temperature humidity pressure Wind power output 0.820 -0.215 0.169 -0.235

[0076] The table shows that wind turbine output is correlated with various meteorological factors. Wind turbine output exhibits a strong positive correlation with wind speed, a very weak correlation with humidity, and a weak negative correlation with temperature and pressure. Wind speed, a meteorological factor strongly correlated with wind power output, was selected to construct a dataset of relevant features under typhoon conditions.

[0077] The relevant feature dataset under typhoon weather was clustered using k-means clustering with DTW distance added, and the optimal number of clusters was selected based on the silhouette coefficient. Figure 4 The values ​​of SC represent different numbers of clusters. The results show that the meteorological type clustering effect is most suitable when K=4. Figure 5 To achieve the optimal number of clusters, four typical wind turbine output scenarios under the influence of typhoons are presented.

[0078] (2) Analysis of the generation effect of wind power output scenario

[0079] The dataset is normalized, and four days of data with a resolution of 10 minutes are used to form a 24*24*2 window. This window is then concatenated with the corresponding conditional labels. The concatenated data is then input into the trained generator G to generate a large number of conditionally labeled typhoon power output datasets.

[0080] The simulation uses the GPU within the TensorFlow framework to train the network model using CUDA parallel computing. The network optimization algorithm employs the RMSProp optimizer, with a learning rate of 0.0002 and a maximum of 1000 training iterations.

[0081] Figure 6 To compare the losses of the improved W-ACGAN with those of the original ACGAN, it can be seen from (6a) and (6c) that, in terms of generation loss, the original ACGAN tends to stabilize after about 200 iterations, while the improved ACGAN reaches stability after about 100 iterations and has less volatility. It can be seen from (6b) and (6d) that, in terms of classification loss, both frameworks stabilize to near 0 after about 300 iterations, but the original ACGAN then shows a large fluctuation, while the improved ACGAN is more stable.

[0082] Figure 7 The dashed line represents the short-term wind power output scenario under typhoon weather generated by the improved ACGAN. It can be seen that the generated wind power scenario has no obvious disordered fluctuations, the output trend is consistent with the predicted value, and the measured value can be well contained within the scenario set range.

[0083] (3) Effectiveness analysis of power output scenarios during typhoons

[0084] The effectiveness of power output scenarios during typhoons is evaluated using two metrics: scenario coverage and average power range width. Higher scenario coverage indicates greater reliability of the generated scenario set; a narrower average power range width indicates that the generated scenario is closer to reality and performs better.

[0085]

[0086]

[0087] Wherein, formulas (8) and (9) represent the scene coverage rate and the average width of the power range, respectively; T′ represents the number of times when the actual wind power output falls into the scene set; T represents the total number of times in the wind power output sequence; and W represents the average width of the power range of the wind power output scene. This represents the upper bound of the wind power output scenario set at time t; This represents the lower bound of the wind power output scenario set at time t.

[0088] Figure 8It can be seen that within the 100% confidence interval, both the improved W-ACGAN and Markov-Coupla scene generation methods (MC method) can cover the actual wind turbine output curves well. However, the wind power output interval of the improved W-ACGAN method is narrower, and the distribution of its generated scenes is more concentrated, which better reflects the wind turbine output characteristics under actual typhoon weather.

[0089] The calculation results of scene generation index of the improved W-ACGAN and MC method under different confidence intervals are shown in the table below:

[0090] Table 2 Calculation Results of Scene Generation Indicators

[0091]

[0092] As shown in Table 2, both the improved ACGAN and the MC method have high coverage. However, the average width of the overall power range is about 33% smaller for the improved ACGAN method than for the MC method, indicating that the scenarios generated by the improved ACGAN method can more accurately describe the uncertainty of wind turbine output during typhoon weather compared to the MC method.

[0093] This case demonstrates that the method proposed in this invention is effective and feasible, and can provide a reference for generating offshore wind power output scenarios under the influence of typhoons.

[0094] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for generating offshore wind power output scenarios considering the impact of typhoons, characterized in that, Includes the following steps: Step 1) Consider the correlation between various meteorological factors and wind power output, and use the Pearson correlation coefficient to simplify the meteorological factors that affect wind power output during typhoon weather; Step 2) In view of the characteristic that the offshore wind power output curve shifts with the time sequence of typhoons, based on the reduced term factor characteristics, K-means clustering with dynamic time planning distance metric is used to construct wind turbine output categories under different typhoon influences, and build a labeled original dataset. Step 3) Use the W-ACGAN network with added Wasserstein distance to augment the labeled original dataset and construct a labeled offshore wind power output scenario during typhoons. Step 4) Based on the wind power output prediction data during typhoons, perform similarity screening on the constructed offshore wind power output scenarios during typhoons to obtain a set of similar scenarios for short-term offshore wind power output under typhoon weather. The wind power output prediction data during the typhoon is determined based on the wind speed during the typhoon and the active power output model of the wind turbine. The active power output model of the wind turbine is modeled using a cubic function. The active power output model of the wind turbine is represented as follows: in, Indicates the first j Active power output of each wind turbine unit; Indicates the first j The output coefficient of each wind turbine unit; Indicates the first j The maximum active power output of a wind turbine unit when it is running at full capacity; These represent the actual wind speed, cut-in wind speed, rated wind speed, and cut-out wind speed of the wind turbine during a typhoon, respectively.

2. The method for generating offshore wind power output scenarios considering typhoon impact according to claim 1, characterized in that, Step 1) specifically involves: using Pearson correlation coefficient analysis to determine the correlation between meteorological factors and wind power output, and selecting meteorological factors with a strong correlation to wind power output to construct a dataset of relevant meteorological characteristics under typhoon weather. The correlation between meteorological factors and wind power output is then analyzed. Represented as: in, and These are wind power output and meteorological factors, respectively. i ; and They are respectively and The first in k One value; and They are respectively and The mean; n This represents the total number.

3. The method for generating offshore wind power output scenarios considering typhoon impact according to claim 1, characterized in that, Step 2) specifically involves: employing an improved k-means clustering method, utilizing dynamic time planning distance as a metric for inter-cluster spacing, and leveraging its ability to allow for temporal scaling during cluster matching to obtain sequences with the most similar morphology possible. The silhouette coefficient is then used as an evaluation method to select the optimal number of clusters. This process is applied to cluster the raw offshore wind power output data affected by typhoon timing. in, DTW ( A , B () represents the dynamic time-planning distance of the offshore wind power output time series under the influence of two typhoon sequences. For different time points; b represents the minimum average distance between the wind turbine output sample and other cluster samples. a The average distance between this sample and other samples within the cluster is denoted as SC; the silhouette coefficient SC ranges from -1 to +1.

4. The method for generating offshore wind power output scenarios considering typhoon impact according to claim 1, characterized in that, Step 3) specifically involves: putting the labeled original dataset obtained in step 2) into the W-ACGAN network framework for training, introducing a gradient penalty function into the discriminator function so that the discriminator can use the Wasserstein distance to calculate the difference between the generated data and the real data, thereby realizing data augmentation and generation of offshore wind power output scenarios during typhoons.

5. The method for generating offshore wind power output scenarios considering the impact of typhoons according to claim 4, characterized in that, The Wasserstein distance is expressed as: in, This represents the Wasserstein distance between the actual output and the generated output data. Distribution of actual output data Distribution of generated output data The set of joint probability distributions; inf Indicates the infimum; y To generate output samples; The distance between the actual output sample and the generated output sample; E It represents the expected distance between the actual power output distribution and the generated power output distribution.

6. The method for generating offshore wind power output scenarios considering the impact of typhoons according to claim 5, characterized in that, Considering that Wasserstein distance is difficult to calculate directly, the Kantorovich-Rubinstein dual form of Wasserstein distance is used to describe the distance between generated samples and real samples: in, The Kantorovich Rubinstein dual form representing the Wasserstein distance. Indicates the supremacy. Represents a function that measures the difference in distributions. , It should satisfy 1-Lipschitz continuity, with an upper bound of 1 for the absolute value of its derivative.

7. The method for generating offshore wind power output scenarios considering typhoon impact according to claim 6, characterized in that, The objective function of the discriminator that introduces a gradient penalty function is: in, This represents the actual power output sample distribution of wind turbines. To generate a sample distribution of wind turbine output, This represents the discriminator function. This represents the probability of classifying a data point as true. This represents the probability of correctly classifying the data. The penalty function represents the gradient of the loss function of the discriminator D. Represents the gradient. This is the penalty coefficient.

8. The method for generating offshore wind power output scenarios considering the impact of typhoons according to claim 1, characterized in that, Step 4) specifically involves: based on the wind power output prediction data during the typhoon, identifying the typhoon impact level label, and using PCC similarity to screen similar scenarios. In descending order, multiple samples with strong similarity to the wind power output during the typhoon to be predicted are selected from the corresponding label generated samples to form a set of similar scenarios for short-term offshore wind power output under typhoon weather.

Images