CN-122000881-A - Ultra-short-term wind power prediction method, device and medium for extreme weather scene

CN122000881ACN 122000881 ACN122000881 ACN 122000881ACN-122000881-A

Abstract

The invention discloses a method, equipment and medium for predicting ultra-short-term wind power in extreme weather scenes, and relates to the technical field of wind power prediction. The method comprises the steps of carrying out outlier identification in a physical constraint-based coarse screening and statistical analysis-based fine screening progressive fusion mode, constructing a multi-strategy processing space, carrying out weighted fusion outlier correction by using Bayesian optimization self-adaptive determination weight, further expanding an extreme weather sample by adopting a TimeGAN model, carrying out weather scene division by combining the correction sample, then correspondingly training a transducer model and a CNN model, and carrying out prediction by adaptively selecting a model corresponding to the training completion in a prediction stage. According to the method, different abnormal value generation reasons are comprehensively considered, a targeted identification and processing method is adopted, and the ultra-short-term wind power prediction of different weather scenes is carried out by adopting a multi-combination model, so that the prediction accuracy is improved.

Inventors

YANG MAO
SUN LIANJUN
JIANG YUXI

Assignees

东北电力大学

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (8)

1. The ultra-short-term wind power prediction method for the extreme weather scene is characterized by comprising the following steps of: Performing outlier identification on a basic training sample by adopting an outlier identification mode of progressive fusion of a coarse sieve based on physical constraint and a fine sieve based on statistical analysis; Constructing a processing space containing a plurality of correction strategies, adaptively optimizing and determining fusion weights of the strategies based on a Bayesian optimization framework, and carrying out weighted fusion abnormal value correction on the identified abnormal values; Carrying out small sample expansion on extreme weather by adopting a TimeGAN model, combining the expansion sample with a correction sample, dividing weather scenes of combined sample data, and correspondingly distributing the combined sample data into a CNN model and a fransformer model for training, wherein the TimeGAN model comprises a generator, a discriminator and a supervisor, the generated sample of the generator is subjected to antagonism with the real sample through the discriminator, the training discriminator distinguishes the real sample from the generated sample, and meanwhile, the training generator generates samples which are approximately in real distribution; Determining basic data of a wind power plant to be tested, wherein the basic data of the wind power plant comprise the capacity of a general assembly machine, rated capacity, wind speed at the height of a hub, wind direction at the height of the hub, atmospheric temperature and numerical weather forecast of atmospheric pressure; And extracting NWP data characteristics based on the wind farm basic data to divide weather scenes, and adaptively selecting a prediction model which is completed by corresponding training to predict ultra-short-term wind power, wherein the weather scenes comprise regular weather and extreme weather.
2. The ultra-short term wind power prediction method for extreme weather scenarios according to claim 1, wherein the coarse screening based on physical constraints comprises the following specific steps: Setting 6 types of physical constraint outlier judgment rules, which are expressed as: Wherein v represents wind speed, p represents power, p max represents maximum generated power, v cut-in represents cut-in wind speed, and v cut-out represents cut-out wind speed; when the training sample does not meet any rule requirement, the training sample is judged to be an abnormal value.
3. The ultra-short term wind power prediction method for extreme weather scenarios according to claim 1, wherein the fine screening based on statistical analysis comprises the following specific steps: and constructing polynomial expansion characteristics of a wind speed sequence for the data after coarse screening, calculating residual errors of real power and predicted power by utilizing a mapping relation between fitted wind speed and power of a gradient lifting regression model, determining a normal value interval based on a quarter bit distance of the residual errors, and judging the data of which the residual errors exceed the interval as abnormal values, wherein the predicted power is determined by the fitted mapping relation between wind speed and power.
4. The ultra-short term wind power prediction method for extreme weather scenarios according to claim 1, wherein the plurality of correction strategies comprises linear interpolation correction, local mean correction and weighted fusion correction based on a gradient lifting regression model.
5. The ultra-short term wind power prediction method for extreme weather scenarios according to claim 1, wherein the bayesian optimization framework comprises: And constructing a double-layer optimization problem to determine an optimal weight vector, wherein for inner-layer optimization, fusion correction is carried out on an abnormal value based on the current weight, and for outer-layer optimization, a proxy model of the weight and the error is established through Gaussian process regression with the aim of minimizing root mean square error of a prediction model on a verification set, and the weight is iteratively updated based on an expected lifting criterion.
6. The method for predicting ultra-short-term wind power in extreme weather according to claim 1, wherein the method for extracting NWP data features based on the wind farm basic data to perform weather scene division and adaptively selecting a prediction model corresponding to the training completion to perform ultra-short-term wind power prediction specifically comprises: Utilizing CEEMDAN to decompose and denoise the original power data in the correction data; And extracting NWP data characteristics based on the denoised data, judging a current weather scene according to the NWP data characteristics, carrying out ultra-short-term wind power prediction by adopting a trained CNN model if the weather is normal weather, and carrying out ultra-short-term wind power prediction by adopting a trained converter model if the weather is extreme weather.
7. An electronic device comprising a memory for storing a computer program and a processor that runs the computer program to cause the electronic device to perform the ultra-short term wind power prediction method for extreme weather scenarios according to any one of claims 1-6.
8. A computer-readable storage medium, characterized in that it stores a computer program, which when executed by a processor implements the ultra-short term wind power prediction method for extreme weather scenarios according to any one of claims 1-6.

Description

Ultra-short-term wind power prediction method, device and medium for extreme weather scene Technical Field The invention relates to the technical field of wind power prediction, in particular to an ultra-short-term wind power prediction method, ultra-short-term wind power prediction equipment and ultra-short-term wind power prediction medium aiming at extreme weather scenes. Background According to the data released by the China energy agency in the recent days, the integrated power generation installed capacity of the China is 37.9 hundred million kilowatts by the end of 11 months in 2025, and the integrated power generation installed capacity is increased by 17.1% in the same ratio. The installed capacity of wind power breaks through 6 hundred million kilowatts and is increased by 22.4% in the same ratio. Wind power is used as a clean energy source, and the output power of the wind power is obviously influenced by weather conditions, especially wind speed and wind direction, and has randomness and fluctuation. The randomness and the volatility of wind power generation provide serious challenges for power grid dispatching, high-precision wind power prediction can provide reliable information for a power grid dispatching side, a power generation plan is reasonably arranged, the digestion capability of wind power is improved, and the safe and stable operation of a power system is ensured. At present, researchers build and analyze a prediction model by using a wind power data set as a whole, but the method does not fully consider that extreme weather can cause the wind power output to fluctuate severely, so that weather changes have obvious nonlinearity and irregularity, a large number of abnormal values appear in the data set, and the difficulty of wind power prediction is further increased. Therefore, in order to improve the accuracy of wind power prediction, a large number of abnormal values in the extreme weather scene must be accurately identified and effectively processed to ensure that the prediction model can still maintain higher prediction accuracy and reliability under actual complex meteorological conditions. Disclosure of Invention The invention aims to provide a method, equipment and medium for predicting ultra-short-term wind power in extreme weather scenes, and aims to solve or improve at least one of the technical problems. In order to achieve the above object, the present invention provides the following solutions: an ultra-short term wind power prediction method for extreme weather scenes comprises the following steps: Performing outlier identification on a basic training sample by adopting an outlier identification mode of progressive fusion of a coarse sieve based on physical constraint and a fine sieve based on statistical analysis; Constructing a processing space containing a plurality of correction strategies, adaptively optimizing and determining fusion weights of the strategies based on a Bayesian optimization framework, and carrying out weighted fusion abnormal value correction on the identified abnormal values; Carrying out small sample expansion on extreme weather by adopting a TimeGAN model, combining the expansion sample with a correction sample, dividing weather scenes of combined sample data, and correspondingly distributing the combined sample data into a CNN model and a fransformer model for training, wherein the TimeGAN model comprises a generator, a discriminator and a supervisor, the generated sample of the generator is subjected to antagonism with the real sample through the discriminator, the training discriminator distinguishes the real sample from the generated sample, and meanwhile, the training generator generates samples which are approximately in real distribution; Determining basic data of a wind power plant to be tested, wherein the basic data of the wind power plant comprise the capacity of a general assembly machine, rated capacity, wind speed at the height of a hub, wind direction at the height of the hub, atmospheric temperature and numerical weather forecast of atmospheric pressure; And extracting NWP data characteristics based on the wind farm basic data to divide weather scenes, and adaptively selecting a prediction model which is completed by corresponding training to predict ultra-short-term wind power, wherein the weather scenes comprise regular weather and extreme weather. Optionally, the coarse screening based on physical constraint comprises the following specific processes: Setting 6 types of physical constraint outlier judgment rules, which are expressed as: Wherein v represents wind speed, p represents power, p max represents maximum generated power, v cut-in represents cut-in wind speed, and v cut-out represents cut-out wind speed; when the training sample does not meet any rule requirement, the training sample is judged to be an abnormal value. Optionally, the fine screening based on statistical analysis comprises the following specific process