Method and system for predicting a day-ahead wind power of wind farms

Patent metadata

Field	Value
Publication number	US-12159245-B2
Application number	US-202217652681-A
Country	US
Kind code	B2
Filing date	Feb 26, 2022
Priority date	Oct 29, 2021
Publication date	Dec 3, 2024
Grant date	Dec 3, 2024

Abstract

A method for predicting a day-ahead wind power of wind farms, comprising: constructing a raw data set based on a correlation between the to-be-predicted daily wind power, the numerical weather forecast meteorological feature and a historical daily wind power; obtaining a clustered data set and performing k-means clustering, obtaining a raw data set with cluster labels, and generating massive labeled scenes based on robust auxiliary classifier generative adversarial networks; determining the cluster label category of the to-be-predicted day based on the known historical daily wind power and numerical weather forecast meteorological feature, and screening out multiple scenes with high similarity to the to-be-predicted daily wind power based on the cluster label category; and obtaining the prediction results of the to-be-predicted daily wind power at a plurality of set times based on an average value, an upper limit value and a lower limit value of the to-be-predicted daily wind power.

First claim

What is claimed is:

1. A method for predicting a day-ahead wind power of a wind farm, comprising:
constructing a raw data set containing a numerical weather forecast meteorological feature and a to-be-predicted daily wind power based on a correlation between the to-be-predicted daily wind power and each of the numerical weather forecast meteorological feature and a historical daily wind power;
removing the to-be-predicted daily wind power in the raw data set, obtaining a clustered data set and performing k-means clustering, obtaining a raw data set with cluster labels, and generating massive labelled scenes based on robust auxiliary classifier generative adversarial networks;
determining a cluster label category of a to-be-predicted day based on the historical daily wind power and the numerical weather forecast meteorological feature, and screening out multiple scenes with high similarity to the to-be-predicted daily wind power from the massive labelled scenes based on the cluster label category, forming a similar scene set;
obtaining point prediction and interval prediction results of the to-be-predicted daily wind power at a plurality of set times based on an average value, an upper limit value and a lower limit value of the to-be-predicted daily wind power in the similar scene set; and
dispatching wind power of the wind farm stably and optimally based on the point prediction and interval prediction results.

2. The method of claim 1 , wherein the step of based on a correlation between the to-be-predicted daily wind power and each of the numerical weather forecast meteorological feature and a historical daily wind power comprises:
selecting a wind speed, a wind direction, a temperature, a humidity and a pressure as a first primary selection feature of the to-be-predicted daily wind power;
selecting a historical daily wind power with an absolute value of a Pearson correlation coefficient greater than a preset threshold as a second primary selection feature of the to-be-predicted daily wind power based on the correlation between the historical daily wind power and the to-be-predicted daily wind power;
calculating a second Pearson correlation coefficient between the to-be-predicted daily wind power and each of the first primary selection feature and the second primary selection feature; and
comparing an absolute value of the second Pearson correlation coefficient with the preset threshold to determine a correlation between the to-be-predicted daily wind power and each of the first primary selection feature and the second primary selection feature.

3. The method of claim 2 , wherein the calculation formula of a Pearson correlation coefficient is as follows:

PCC
⁡
(

x
,

y
i

)

=

1
-

∑

k
=
1

n

(

x
k

-

x
¯

)

⁢

(

y
ik

-

y
i

_

)

∑

k
=
1

n

(

x
k

-

x
¯

)

⁢

∑

k
=
1

n

(

y
ik

-

y
i

_

)

wherein, x is the to-be-predicted daily wind power, y i is an impact feature of the to-be-predicted daily wind power; x k and y ik are the k-th data in x and y i ; x and y i are average values of x and y i ; k and n are positive integers.

4. The method of claim 3 , wherein the step of constructing a raw data set containing a numerical weather forecast meteorological feature and a to-be-predicted daily wind power comprises:
removing primary selection features with very weak correlation or without correlation based on a correlation determination result between the to-be-predicted daily wind power and each of the first primary selection feature and the second primary selection feature;
taking remaining primary selection features as the impact features of the to-be-predicted daily wind power, and normalizing the features predicted by a data set corresponding to the impact features;
integrating the normalized features with the to-be-predicted daily wind power in turn based on a preset arrangement sequence, and forming a to-be-predicted daily raw data; and
constructing an annual raw data sample based on the to-be-predicted daily raw data and selecting several raw data from the annual raw data sample randomly to construct a raw data set.

5. The method of claim 4 , wherein the step of removing the to-be-predicted daily wind power in the raw data set, obtaining a clustered data set and performing k-means clustering, obtaining a raw data set with cluster labels comprises:
removing the to-be-predicted daily wind power in the raw data set, obtaining the normaliz

Assignees

• Economic And Tech Research Institute Of State Grid Liaoning Electric Power Co Ltd
• State Grid Corp China
• Univ Northeast Electric Power

Inventors

• Pan Xiao
• Zhang Mingli
• Zhao Lin
• Zhang Na
• Song Zhuoran
• Huang nantian
• Gao Jing
• Lv Xuming
• Li Hua
• Cheng Mengzeng
• Ji Xing
• Shang Wenying
• Hou Yixin
• Yang Suo
• Yang Bo
• Liu Yutong
• Man Linkun
• Xu Xilin
• Yang Haifeng
• Yang Fangyuan
• Liu Kai
• Li Jinqi
• Wang Zongyuan

Classifications

• G06F18/2133

  based on naturality criteria, e.g. with non-negative factorisation or negative correlation · CPC title

• G06F17/15

  Correlation function computation {including computation of convolution operations (arithmetic circuits for sum of products per se, e.g. multiply-accumulators G06F7/5443; digital filters, e.g. FIR, IIR, adaptive filters H03H17/00)} · CPC title

• G06F18/211

  Selection of the most significant subset of features · CPC title

• G06Q50/06

  Energy or water supply · CPC title

• G06N3/0475

  Generative networks · CPC title