Method and device for generating typical wind power scene and storage medium
By generating typical wind power scenarios, this method addresses the shortcomings of traditional wind power forecasting methods in medium- and long-term high-dimensional time series simulation and uncertainty characterization, achieving efficient modeling and forecasting of wind power output and improving the stability of the power grid and the predictability of wind power generation.
Patent Information
- Application Number
- CN202510965560.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-14
- Publication Date
- 2025-10-31
AI Technical Summary
Traditional wind power forecasting methods are insufficient in medium- and long-term high-dimensional time series simulation and medium- and long-term uncertainty characterization. They are difficult to accurately model the probability distribution of wind speed and radiation, which leads to uncertainty in wind power and photovoltaic power output, posing a challenge to the stable operation of the power grid.
Historical wind speed data is acquired, expanded, and preprocessed before being input into the VAE-RealNVP model to generate a probability distribution. Combined with wind speed, installed capacity, and wind curtailment rate characteristics, a point-to-point power prediction model is constructed to generate typical wind power scenarios. Convolutional neural networks and gated recurrent units are used for time-series modeling, and representative scenarios are generated through hierarchical clustering reduction.
It enables efficient modeling and accurate prediction of wind power output, reduces data redundancy, improves the predictability of wind power generation and the stability of the power grid, and provides efficient and accurate data support for power grid dispatch.
Smart Images
Figure CN120879538A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind power scene generation technology, specifically to a method, apparatus, and storage medium for generating typical wind power scenes. Background Technology
[0002] Renewable energy sources such as wind and solar power are crucial components of the global energy transition. Driven by environmental protection and the low-carbon economy, wind and solar power generation are receiving increasing attention due to their cleanliness, renewability, and economic viability. However, a major challenge facing wind and solar power generation is the volatility and intermittency of their output power, a phenomenon strongly influenced by natural conditions such as wind speed and radiation intensity. Due to the fluctuations in wind speed and radiation, the power generation of wind and solar power exhibits significant uncertainty at different times and locations, posing a severe challenge to the stable operation of the power grid.
[0003] To address this issue, accurate modeling and simulation of the probability distributions of wind speed and radiation are needed. Accurate modeling of these natural factors' probability distributions can help predict potential trends in wind and solar energy, thus providing data support for grid dispatch. The probability distributions of wind speed and radiation not only affect the power generation dispatch of individual power plants but also involve the load balance of the grid and the stability of power supply over a larger scale. Through refined probability distribution models, the output fluctuations of wind and solar power generation can be more accurately assessed, providing a theoretical basis for grid load forecasting, power dispatch, and energy storage management, reducing waste in wind and solar power generation (such as wind and solar curtailment). However, traditional wind power forecasting methods have shortcomings in medium- and long-term high-dimensional time-series simulation, especially in characterizing medium- and long-term uncertainties. Summary of the Invention
[0004] In view of this, the purpose of the present invention is to provide a method, apparatus and storage medium for generating typical wind power scenarios, so as to solve the problem of the shortcomings of traditional wind power prediction methods in medium- and long-term high-dimensional time series simulation, especially in the characterization of medium- and long-term uncertainties.
[0005] According to a first aspect of the present invention, a method for generating a typical wind power scenario is provided, the method comprising:
[0006] Obtain wind speed data for a preset historical time period, construct an original dataset based on the wind speed data for the preset historical time period, divide the original dataset according to seasons, and expand the wind speed data for each season.
[0007] Preprocess the expanded wind speed data;
[0008] The preprocessed wind speed data is input into the pre-built VAE-RealNVP model to generate the probability distribution of the wind speed data;
[0009] Input features are selected by calculating the correlation between different features and the target variable. The input features include wind speed, installed capacity, and wind curtailment rate.
[0010] The installed capacity and wind curtailment rate for each month in the historical preset time period are obtained from the power grid data platform. The wind speed data for each month in the original dataset is also obtained. A training set is constructed based on the monthly wind speed data, installed capacity and wind curtailment rate.
[0011] The pre-built model architecture is iteratively trained using the training set to obtain a trained point-to-point power prediction model.
[0012] Based on the probability distribution of the wind speed data, multiple wind speed curves are obtained by sampling; the multiple wind speed curves are respectively input into the trained point-to-point power prediction model to generate the wind power scenario corresponding to each wind speed curve.
[0013] Multiple wind power scenarios are simplified to obtain a preset number of typical scenarios.
[0014] Preferably,
[0015] The expansion of wind speed data for each season includes:
[0016] The time granularity of the wind speed data was adjusted to the hourly level. The window size and the distance of each translation were set. The sliding window method was used to translate the wind speed data for each season to obtain multiple overlapping samples.
[0017] Preferably,
[0018] The preprocessing of the expanded wind speed data includes:
[0019] Obtain the mean and standard deviation of the expanded wind speed data, set a filtering coefficient, and obtain a standard range based on the mean, standard deviation, and filtering coefficient;
[0020] Wind speed data exceeding the standard range are considered outliers, and these outliers are replaced with the mean or boundary value.
[0021] The K-nearest neighbor algorithm was used to fill in the missing values in the wind speed data after outlier replacement.
[0022] Preferably,
[0023] The step of inputting the preprocessed wind speed data into the pre-built VAE-RealNVP model to obtain the probability distribution of the wind speed data includes:
[0024] The preprocessed wind speed data is input into the encoder part of the VAE. The encoder part of the VAE maps the input wind speed data to the latent space to obtain the mean and variance of the input wind speed data.
[0025] A normal distribution is established based on the mean and variance of the wind speed data, and the latent variable Z0 is sampled from the normal distribution using a reparameterization method.
[0026] During the forward propagation of the model, the RealNVP model is used to perform an inverse transformation on the latent variable Z0 to obtain the latent variable Z;
[0027] The latent variable Z is input into the decoder, and the latent variable Z is mapped back to the original data space to obtain the probability distribution of the wind speed data.
[0028] Preferably,
[0029] The step of iteratively training the pre-built model architecture using a training set to obtain the trained point-to-point power prediction model includes:
[0030] Mean squared error is used as the loss function in the model training process. Monthly wind speed data, installed capacity rate, and wind curtailment rate are input into the pre-built model architecture, which outputs the wind power of the corresponding month. Through iterative training, the loss value of the loss function is continuously reduced until the loss value no longer decreases or reaches the preset number of iterations, thus obtaining the trained point-to-point power prediction model.
[0031] Preferably,
[0032] The pre-built model architecture includes:
[0033] First, a multi-channel one-dimensional convolutional layer is built, with the convolutional kernel sliding on the time axis to extract local features. Then, a multi-layer GRU network is connected to perform temporal modeling.
[0034] Preferably,
[0035] The reduction of multiple wind power scenarios to obtain a preset number of typical scenarios includes:
[0036] A 720-dimensional scene vector is constructed based on each wind power scenario and the corresponding wind speed data;
[0037] Obtain the Euclidean distance matrix between each scene vector;
[0038] Treat each scene as an independent cluster and initialize the probability distribution of each scene to be equal.
[0039] Based on the Euclidean distance matrix between scene vectors, select the scene vector pair with the smallest weighted distance for merging, and gradually adjust the probability distribution of the merged clusters until a preset number of typical scenes are reached.
[0040] Generate a combination of typical scenarios and their corresponding probabilities as the reduced result.
[0041] According to a second aspect of the present invention, an apparatus for generating typical wind power scenarios is provided, the apparatus comprising:
[0042] Historical wind speed data acquisition module: used to acquire wind speed data for a preset historical time period, construct an original dataset based on the wind speed data for the preset historical time period, divide the original dataset according to seasons, and expand the wind speed data for each season.
[0043] Preprocessing module: used to preprocess the expanded wind speed data;
[0044] Probability distribution acquisition module: used to input preprocessed wind speed data into a pre-built VAE-RealNVP model to generate the probability distribution of wind speed data;
[0045] Input feature filtering module: used to filter input features by calculating the correlation between different features and the target variable. The input features include wind speed, installed capacity rate, and wind curtailment rate.
[0046] Training set construction module: used to obtain the monthly installed capacity rate and wind curtailment rate of the historical preset time period from the power grid data platform, obtain the monthly wind speed data from the original dataset, and construct a training set based on the monthly wind speed data, installed capacity rate and wind curtailment rate;
[0047] Prediction model training module: used to iteratively train a pre-built model architecture using a training set to obtain a trained point-to-point power prediction model;
[0048] Scene generation module: Based on the probability distribution of the wind speed data, it obtains multiple wind speed curves by sampling; inputs the multiple wind speed curves into the trained point-to-point power prediction model to generate wind power scene corresponding to each wind speed curve;
[0049] Typical scenario acquisition module: used to reduce multiple wind power scenarios to obtain a preset number of typical scenarios.
[0050] According to a third aspect of the present invention, a storage medium is provided, the storage medium storing a computer program, which, when executed by a host controller, implements the steps of the above-described method.
[0051] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:
[0052] This application acquires historical wind speed data over a preset time period, expands and preprocesses the wind speed data, inputs the preprocessed wind speed data into a pre-built VAE-RealNVP model to obtain the probability distribution of the wind speed data, filters input features by calculating the correlation between different features and the target variable, and iteratively trains the pre-built model architecture by constructing a training set to obtain a trained point-to-point power prediction model. Based on the probability distribution of the wind speed data, multiple wind speed curves are obtained by sampling. These multiple wind speed curves are input into the trained point-to-point power prediction model to generate wind power scenarios corresponding to each wind speed curve. The multiple wind power scenarios are reduced to obtain a preset number of typical scenarios. This application, through accurate wind speed data modeling, wind power mapping, and efficient scenario reduction technology, solves the shortcomings of traditional wind power prediction methods in medium- and long-term high-dimensional time series simulation, especially in characterizing medium- and long-term uncertainties, and can better cope with the fluctuations and uncertainties of wind power output.
[0053] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description
[0054] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0055] Figure 1 This is a flowchart illustrating a method for generating a typical wind power scenario according to an exemplary embodiment;
[0056] Figure 2 This is a system schematic diagram of an apparatus for generating a typical wind power scenario, according to another exemplary embodiment;
[0057] In the attached diagram: 1-Historical wind speed data acquisition module, 2-Preprocessing module, 3-Probability distribution acquisition module, 4-Input feature filtering module, 5-Training set construction module, 6-Prediction model training module, 7-Scene generation module, 8-Typical scene acquisition module. Detailed Implementation
[0058] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.
[0059] Example 1
[0060] Figure 1 This is a flowchart illustrating a method for generating a typical wind power scenario according to an exemplary embodiment, such as... Figure 1 As shown, the method includes:
[0061] S1, Obtain wind speed data for a historical preset time period, construct an original dataset based on the wind speed data for the historical preset time period, divide the original dataset according to seasons, and expand the wind speed data for each season.
[0062] S2, preprocess the expanded wind speed data;
[0063] S3, input the preprocessed wind speed data into the pre-built VAE-RealNVP model to generate the probability distribution of the wind speed data;
[0064] S4. Input features are selected by calculating the correlation between different features and the target variable. The input features include wind speed, installed capacity, and wind curtailment rate.
[0065] S5. Obtain the monthly installed capacity and wind curtailment rate for each month in the historical preset time period from the power grid data platform, obtain the monthly wind speed data from the original dataset, and construct a training set based on the monthly wind speed data, installed capacity and wind curtailment rate.
[0066] S6. Iteratively train the pre-built model architecture using the training set to obtain a trained point-to-point power prediction model.
[0067] S7. Based on the probability distribution of the wind speed data, multiple wind speed curves are obtained by sampling; the multiple wind speed curves are respectively input into the trained point-to-point power prediction model to generate the wind power scenario corresponding to each wind speed curve.
[0068] S8 reduces multiple wind power scenarios to obtain a preset number of typical scenarios;
[0069] Understandably, this embodiment first uses a variational autoencoder (VAE) model to efficiently model wind speed data. By introducing the RealNVP flow model, this method can complicate the posterior distribution of wind speed data, thereby generating wind speed data that conforms to the actual climate change patterns. Specifically, the VAE model maps the original wind speed data to the latent space through an encoder, and then reconstructs the wind speed data through a decoder. The RealNVP flow model optimizes the posterior distribution, handles more complex wind speed data distributions, and ensures that the generated wind speed data has seasonal and temporal characteristics, effectively reflecting the dynamic changes of actual wind farms. The steps are as follows:
[0070] Raw wind speed data acquisition: Obtain wind speed data from historical time periods (generally more than one year) as raw wind speed data. Divide the raw dataset into four parts according to the season (spring, summer, autumn, winter), with each part corresponding to the data of one season.
[0071] Time granularity adjustment: Adjust the time granularity of the data to the hour level. Assume there are 30 days in a month and 24 hours in a day, for a total of 720 hours (24*30).
[0072] Data expansion: The sliding window technique is used to shift and expand the data for each season to increase the number of samples. For example, the window size is set to 720 hours, and the window is shifted by 24 hours each time to generate multiple overlapping samples. The processed dataset for each season has the shape of (x, 720), where x is the number of samples and 720 represents the number of hours for each sample.
[0073] Preprocessing: The n-sigma method is used to detect and process outliers in the data. Specifically, the following steps are taken: set the screening coefficient n, calculate the mean μ and standard deviation σ of the data, regard values that exceed the range of (μ-nσ, μ+nσ) as outliers and process them accordingly (such as replacing them with the mean or boundary values), and use the K-Nearest Neighbor (KNN) algorithm to impute missing values in the data to ensure the continuity and integrity of the data.
[0074] Encoder processing: The preprocessed data is input into the encoder part of the VAE. The encoder maps the input data x to the latent space, obtaining the mean μ and variance σ. 2 ;
[0075] Reparameterized sampling: based on mean μ and variance σ 2 Establish a normal distribution, and use the reparameterization technique to obtain the normal distribution q0(z0|x)=N(μ,σ) 2 Sample latent variable Z0 in )
[0076] Optimizing the posterior through a flow model: A flow model is introduced into the model to optimize the posterior distribution. During the forward propagation of the model, the RealNVP model is used to perform an inverse transformation on the latent variable Z0, resulting in a more complex, data-adaptive distribution of the latent variable Z. The specific formula is as follows: Subsequently, the complex posterior distribution q(z|x) and prior distribution p(z) obtained from the flow model will be used to calculate the KL divergence as part of the loss function to optimize the entire generation process;
[0077] Decoder processing: The latent variable Z after inverse transformation is input into the Decoder, and the latent variable Z is mapped back to the original data space to generate the probability distribution of wind speed data;
[0078] Generation and mapping of wind power output data:
[0079] This embodiment constructs a point-to-point power prediction model, combining important features such as wind speed, wind curtailment rate, and installed capacity. It employs a convolutional neural network (CNN) and gated recurrent unit (GRU) model to generate wind power output data. The CNN is used to process and extract local patterns from raw features such as wind speed, while the GRU further captures the time-varying patterns of wind power through time-series modeling. The model outputs accurate wind power output data and can effectively adapt to complex time-series data and nonlinear relationships. The implementation steps are as follows:
[0080] Feature selection: By calculating the correlation between each feature and the target variable (such as power output efficiency), it is determined that wind speed, installed capacity, and wind curtailment rate have a high correlation with the output value of the target variable. For example, wind speed and power output efficiency show a strong positive correlation, while wind curtailment rate and power output efficiency have a certain correlation. Based on these results, this embodiment uses wind speed, installed capacity, and wind curtailment rate as features of the final input model. These features are not only closely related to power output efficiency, but also physically closely related to the operating state of wind power generation, which helps the model to better predict power output efficiency and improve prediction accuracy.
[0081] Input / output construction: In the initial stage of data processing, the shape of the input tensor is [batch_size, seq_length, input_dim], where batch_size is the batch size of the input data, seq_length is 720, and input_dim includes features such as wind speed in a certain month (from the original dataset mentioned above), installed capacity, and wind curtailment rate (installed capacity and wind curtailment rate are obtained through the data platform inside the power grid). The shape of the output tensor is [batch_size, seq_length, 1], which is the power in that month.
[0082] Model setup: First, a multi-channel one-dimensional convolutional layer (1D CNN) is built. The convolutional kernel slides on the time axis to extract local features, such as short-term trends and fluctuations. Then, a multi-layer GRU network is connected to perform time series modeling.
[0083] Model Training: The loss function used in this embodiment is Mean Squared Error (MSE), a commonly used loss function in regression problems. It measures the difference between the model's predicted values and the actual values. MSE evaluates the model's performance by calculating the mean of the squares of the differences between the predicted and actual values. The formula for MSE is as follows:
[0084]
[0085] Where N represents the number of samples, y i It is the true value of the i-th sample. is the predicted value of the i-th sample. The smaller the MSE value, the closer the model's prediction result is to the true value and the better the prediction performance. The model parameters are iterated through backpropagation of the loss function until the loss value no longer decreases or the preset number of iterations is reached, and the trained point-to-point power prediction model is obtained.
[0086] Massive scene generation and correction:
[0087] Based on the probability distribution generated from the wind speed data obtained above, several wind speed curves are obtained through sampling, and these scenarios are input into the point-to-point power prediction model to generate corresponding wind power scenarios. The generated multiple prediction scenarios cover different power demands and weather conditions. These scenarios can not only simulate the output fluctuations of wind power under different weather conditions, but also provide important data support for subsequent grid dispatch and management.
[0088] Generation and reduction of typical scenarios:
[0089] The reduction method used in this embodiment is based on hierarchical clustering technology, combined with a custom iterative merging strategy. This method can effectively simplify a large number of original scenarios into a few representative scenarios (i.e., typical scenarios). These typical scenarios not only reduce computational complexity but also preserve the probability distribution characteristics and volatility of wind power output data to the greatest extent, providing efficient and accurate input data for power system scheduling, energy storage management, and load forecasting. The specific implementation steps are as follows:
[0090] Data preparation: Based on the massive number of scenes and their corresponding meteorological characteristics such as wind speed, a 720-dimensional scene vector was constructed.
[0091] Distance calculation: Calculate the Euclidean distance matrix between all scene vectors to measure the similarity between scenes.
[0092] Initial settings: Treat each scene as an independent cluster and initialize the probability distribution of each scene to be equal.
[0093] Iterative merging: Based on the distance matrix, select the scene pairs with the smallest weighted distance for merging, and gradually adjust the probability distribution of the merged clusters until a preset number of typical scenes are reached.
[0094] Output result: Finally, a combination of typical scenarios and their corresponding probabilities is generated as the reduced result.
[0095] This embodiment can generate high-quality wind power output data and reduce data redundancy through scenario simplification technology, thereby optimizing the efficiency of wind power prediction. Based on factors such as wind speed, wind curtailment rate, and installed capacity, this method can generate representative wind power output data, providing strong support for grid dispatch and wind farm operation and management. In addition, this invention has strong adaptability and can generate wind power output data according to different regions and seasonal changes, further improving the predictability of wind power generation and the stability of the power system.
[0096] Example 2
[0097] Figure 2 This is a system schematic diagram illustrating an apparatus for generating a typical wind power scenario according to another exemplary embodiment, the apparatus comprising:
[0098] Historical wind speed data acquisition module 1: used to acquire wind speed data for a preset historical time period, construct an original dataset based on the wind speed data for the preset historical time period, divide the original dataset according to seasons, and expand the wind speed data for each season.
[0099] Preprocessing module 2: Used to preprocess the expanded wind speed data;
[0100] Probability distribution acquisition module 3: used to input the preprocessed wind speed data into the pre-built VAE-RealNVP model to generate the probability distribution of the wind speed data;
[0101] Input feature filtering module 4: used to filter input features by calculating the correlation between different features and the target variable, the input features including wind speed, installed capacity rate and wind curtailment rate;
[0102] Training set construction module 5: used to obtain the monthly installed capacity rate and wind curtailment rate of the historical preset time period from the power grid data platform, obtain the monthly wind speed data in the original dataset, and construct a training set based on the monthly wind speed data, installed capacity rate and wind curtailment rate.
[0103] Prediction model training module 6: Used to iteratively train the pre-built model architecture using the training set to obtain the trained point-to-point power prediction model;
[0104] Scene generation module 7: Based on the probability distribution of the wind speed data, it obtains multiple wind speed curves by sampling; inputs the multiple wind speed curves into the trained point-to-point power prediction model to generate a wind power scene corresponding to each wind speed curve;
[0105] Typical Scenarios Acquisition Module 8: Used to reduce multiple wind power scenarios to obtain a preset number of typical scenarios.
[0106] Example 3:
[0107] This embodiment provides a storage medium storing a computer program, which, when executed by a host controller, implements the various steps in the above method.
[0108] It is understood that the storage medium mentioned above can be a read-only memory, a hard disk, or an optical disk, etc.
[0109] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.
[0110] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means at least two.
[0111] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.
[0112] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0113] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.
[0114] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.
[0115] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.
[0116] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0117] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A method for generating typical wind power scenarios, characterized in that, The method includes: Obtain wind speed data for a preset historical time period, construct an original dataset based on the wind speed data for the preset historical time period, divide the original dataset according to seasons, and expand the wind speed data for each season. Preprocess the expanded wind speed data; The preprocessed wind speed data is input into the pre-built VAE-RealNVP model to generate the probability distribution of the wind speed data; Input features are selected by calculating the correlation between different features and the target variable. The input features include wind speed, installed capacity, and wind curtailment rate. The installed capacity and wind curtailment rate for each month in the historical preset time period are obtained from the power grid data platform. The wind speed data for each month in the original dataset is also obtained. A training set is constructed based on the monthly wind speed data, installed capacity and wind curtailment rate. The pre-built model architecture is iteratively trained using the training set to obtain a trained point-to-point power prediction model. Based on the probability distribution of the wind speed data, multiple wind speed curves are obtained by sampling; the multiple wind speed curves are respectively input into the trained point-to-point power prediction model to generate the wind power scenario corresponding to each wind speed curve. Multiple wind power scenarios are simplified to obtain a preset number of typical scenarios.
2. The method according to claim 1, characterized in that, The expansion of wind speed data for each season includes: The time granularity of the wind speed data was adjusted to the hourly level. The window size and the distance of each translation were set. The sliding window method was used to translate the wind speed data for each season to obtain multiple overlapping samples.
3. The method according to claim 2, characterized in that, The preprocessing of the expanded wind speed data includes: Obtain the mean and standard deviation of the expanded wind speed data, set a filtering coefficient, and obtain a standard range based on the mean, standard deviation, and filtering coefficient; Wind speed data exceeding the standard range are considered outliers, and these outliers are replaced with the mean or boundary value. The K-nearest neighbor algorithm was used to fill in the missing values in the wind speed data after outlier replacement.
4. The method according to claim 3, characterized in that, The step of inputting the preprocessed wind speed data into the pre-built VAE-RealNVP model to obtain the probability distribution of the wind speed data includes: The preprocessed wind speed data is input into the encoder part of the VAE. The encoder part of the VAE maps the input wind speed data to the latent space to obtain the mean and variance of the input wind speed data. A normal distribution is established based on the mean and variance of the wind speed data, and the latent variable Z0 is sampled from the normal distribution using a reparameterization method. During the forward propagation of the model, the RealNVP model is used to perform an inverse transformation on the latent variable Z0 to obtain the latent variable Z; The latent variable Z is input into the decoder, and the latent variable Z is mapped back to the original data space to obtain the probability distribution of the wind speed data.
5. The method according to claim 4, characterized in that, The step of iteratively training the pre-built model architecture using a training set to obtain the trained point-to-point power prediction model includes: Mean squared error is used as the loss function in the model training process. Monthly wind speed data, installed capacity rate, and wind curtailment rate are input into the pre-built model architecture, which outputs the wind power of the corresponding month. Through iterative training, the loss value of the loss function is continuously reduced until the loss value no longer decreases or reaches the preset number of iterations, thus obtaining the trained point-to-point power prediction model.
6. The method according to claim 5, characterized in that, The pre-built model architecture includes: First, a multi-channel one-dimensional convolutional layer is built, with the convolutional kernel sliding on the time axis to extract local features. Then, a multi-layer GRU network is connected to perform temporal modeling.
7. The method according to claim 6, characterized in that, The reduction of multiple wind power scenarios to obtain a preset number of typical scenarios includes: A 720-dimensional scene vector is constructed based on each wind power scenario and the corresponding wind speed data; Obtain the Euclidean distance matrix between each scene vector; Treat each scene as an independent cluster and initialize the probability distribution of each scene to be equal. Based on the Euclidean distance matrix between scene vectors, select the scene vector pair with the smallest weighted distance for merging, and gradually adjust the probability distribution of the merged clusters until a preset number of typical scenes are reached. Generate a combination of typical scenarios and their corresponding probabilities as the reduced result.
8. An apparatus for generating typical wind power scenarios, characterized in that, The device includes: Historical wind speed data acquisition module: used to acquire wind speed data for a preset historical time period, construct an original dataset based on the wind speed data for the preset historical time period, divide the original dataset according to seasons, and expand the wind speed data for each season. Preprocessing module: used to preprocess the expanded wind speed data; Probability distribution acquisition module: used to input preprocessed wind speed data into a pre-built VAE-RealNVP model to generate the probability distribution of wind speed data; Input feature filtering module: used to filter input features by calculating the correlation between different features and the target variable. The input features include wind speed, installed capacity rate, and wind curtailment rate. Training set construction module: used to obtain the monthly installed capacity rate and wind curtailment rate of the historical preset time period from the power grid data platform, obtain the monthly wind speed data from the original dataset, and construct a training set based on the monthly wind speed data, installed capacity rate and wind curtailment rate; Prediction model training module: used to iteratively train a pre-built model architecture using a training set to obtain a trained point-to-point power prediction model; Scene generation module: Based on the probability distribution of the wind speed data, it obtains multiple wind speed curves by sampling; inputs the multiple wind speed curves into the trained point-to-point power prediction model to generate wind power scene corresponding to each wind speed curve; Typical scenario acquisition module: used to reduce multiple wind power scenarios to obtain a preset number of typical scenarios.
9. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by the main controller, implements the various steps of the method for generating a typical wind power scenario as described in any one of claims 1-7.