CN113962495A - Wind power plant power prediction method and device and electronic equipment - Google Patents

Wind power plant power prediction method and device and electronic equipment Download PDF

Info

Publication number
CN113962495A
CN113962495A CN202111568234.2A CN202111568234A CN113962495A CN 113962495 A CN113962495 A CN 113962495A CN 202111568234 A CN202111568234 A CN 202111568234A CN 113962495 A CN113962495 A CN 113962495A
Authority
CN
China
Prior art keywords
wind
power
attention
module
extraction module
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111568234.2A
Other languages
Chinese (zh)
Inventor
凡航
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huakong Tsingjiao Information Technology Beijing Co Ltd
Original Assignee
Huakong Tsingjiao Information Technology Beijing Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huakong Tsingjiao Information Technology Beijing Co Ltd filed Critical Huakong Tsingjiao Information Technology Beijing Co Ltd
Priority to CN202111568234.2A priority Critical patent/CN113962495A/en
Publication of CN113962495A publication Critical patent/CN113962495A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/06—Wind turbines or wind farms

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Mathematics (AREA)
  • Software Systems (AREA)
  • Operations Research (AREA)
  • General Business, Economics & Management (AREA)
  • Mathematical Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Marketing (AREA)
  • Mathematical Optimization (AREA)
  • Tourism & Hospitality (AREA)
  • Artificial Intelligence (AREA)
  • Pure & Applied Mathematics (AREA)
  • Public Health (AREA)
  • Primary Health Care (AREA)
  • General Health & Medical Sciences (AREA)
  • Water Supply & Treatment (AREA)
  • Geometry (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Computer Hardware Design (AREA)
  • Quality & Reliability (AREA)
  • Medical Informatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Algebra (AREA)

Abstract

The application discloses a method and a device for predicting power of a wind power plant and electronic equipment, wherein the method comprises the following steps: acquiring power characteristic data and wind speed characteristic data of a plurality of wind power plants included in a wind power plant cluster; the method comprises the steps of inputting power characteristic data and wind speed characteristic data of a plurality of wind power plants into a pre-established graph attention network model to obtain predicted power of the plurality of wind power plants, wherein the power characteristic data are used as model input data and are sequentially calculated by an attention characteristic extraction module and a convolution calculation module to obtain power characteristics, the wind speed characteristic data are used as model input data and are sequentially calculated by the attention characteristic extraction module and the convolution calculation module to obtain wind speed characteristics, the power characteristics and the wind speed characteristics are calculated by a characteristic fusion module to obtain predicted power of the plurality of wind power plants, and the attention characteristic extraction module comprises a space attention characteristic extraction module. By adopting the scheme, the accuracy of the wind power plant power prediction is improved.

Description

Wind power plant power prediction method and device and electronic equipment
Technical Field
The application relates to the technical field of computers and wind farm power prediction, in particular to a wind farm power prediction method and device and electronic equipment.
Background
In the related technology of wind power plants, the power of the wind power plants is often required to be predicted, the prediction of the power of the wind power plants refers to the prediction of the total power of the wind power plants within a period of time in the future according to a certain time interval, and the prediction of the power of the wind power plants refers to the prediction of the total power of the wind power plants within a period of time in the future according to a certain time interval, for example, within 4 hours in the future, the prediction is performed according to a time interval not less than 15 minutes.
Accurate wind farm power prediction is of great significance to scheduling department for heliostat power generation plans, adjusting the operation modes of power systems and the like, and provides important information for trading of power spot markets.
The existing wind power plant power prediction can adopt a physical method, an accumulation method based on single wind power plant prediction, a spatial scale-up method and the like. The physical method has the characteristics of clear flow, strong interpretability of a prediction result and poor prediction precision. The accumulation method based on single wind farm prediction does not fully consider the correlation among wind farms in the prediction process. The spatial upscaling method is coarse in granularity when correlation among wind power plants is considered.
There are currently studies that propose using graph machine learning methods for power prediction of wind farms, but with the same weights applied to the input data at the time of prediction. However, the power of the electric field of the wind has a strong non-linear and complex pattern, and when ultra-short term prediction is performed, the attention on different parts of the input data should be different.
Therefore, various known wind power plant power prediction methods have the problem of low prediction accuracy to different degrees.
Disclosure of Invention
The embodiment of the application provides a wind power plant power prediction method, a wind power plant power prediction device and electronic equipment, and aims to solve the problem that the wind power plant power prediction precision is low in the prior art.
The embodiment of the application provides a wind power plant power prediction method which is characterized by comprising the following steps:
acquiring wind power of a plurality of wind power plants at a plurality of historical moments included in a wind power plant cluster as power characteristic data, and acquiring predicted wind speeds of the plurality of wind power plants at a plurality of future moments as wind speed characteristic data;
inputting the power characteristic data and the wind speed characteristic data of the wind power plants into a pre-established graph attention network model to obtain the predicted power of the wind power plants, wherein the graph attention network model comprises an attention feature extraction module, a convolution calculation module and a feature fusion module, the power characteristic data is used as model input data, and is sequentially calculated by the attention characteristic extraction module and the convolution calculation module to obtain power characteristics, the wind speed characteristic data is used as model input data, and wind speed characteristics are obtained after calculation of the attention characteristic extraction module and the convolution calculation module in sequence, and after the power characteristics and the wind speed characteristics are calculated by the characteristic fusion module, the predicted power of the wind power plants is obtained, and the attention characteristic extraction module comprises a space attention characteristic extraction module.
Further, the spatial attention feature extraction module performs calculation by using the following formula:
Figure 655068DEST_PATH_IMAGE001
；
Figure 960147DEST_PATH_IMAGE002
；
wherein the content of the first and second substances,
Figure 343111DEST_PATH_IMAGE003
spatial input data for a spatial attention feature extraction module, the spatial input data being the power feature data or the wind speed feature data,
Figure 809865DEST_PATH_IMAGE004
for the number of said plurality of wind farms,
Figure 924451DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 721637DEST_PATH_IMAGE006
the number of the plurality of time instants,
Figure 706911DEST_PATH_IMAGE007
，
Figure 344566DEST_PATH_IMAGE008
，
Figure 195716DEST_PATH_IMAGE009
，
Figure 514702DEST_PATH_IMAGE010
，
Figure 682378DEST_PATH_IMAGE011
is a parameter used for the learning of,
Figure 179349DEST_PATH_IMAGE012
is an activation function;
Figure 268528DEST_PATH_IMAGE013
for the spatial attention matrix calculated based on the spatial input data of the spatial attention feature extraction module,
Figure 187943DEST_PATH_IMAGE013
element (1) of
Figure 411724DEST_PATH_IMAGE014
Representing wind farms
Figure 656761DEST_PATH_IMAGE015
Wind farm
Figure 702077DEST_PATH_IMAGE016
The spatial correlation between the two signals is determined,
Figure 379177DEST_PATH_IMAGE017
the output data of the spatial attention feature extraction module.
Further, the attention feature extraction module further comprises a temporal attention feature extraction module, an output of the temporal attention feature extraction module being an input of the spatial attention feature extraction module.
Further, the time attention feature extraction module calculates by using the following formula:
Figure 521446DEST_PATH_IMAGE018
；
Figure 875067DEST_PATH_IMAGE019
；
Figure 922526DEST_PATH_IMAGE020
；
wherein the content of the first and second substances,
Figure 387005DEST_PATH_IMAGE003
time input data for a time attention feature extraction module, the time input data being the power feature data or the wind speed feature data,
Figure 587043DEST_PATH_IMAGE004
for the number of said plurality of wind farms,
Figure 111565DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 944523DEST_PATH_IMAGE006
the number of the plurality of time instants,
Figure 212693DEST_PATH_IMAGE021
，
Figure 50592DEST_PATH_IMAGE022
，
Figure 542754DEST_PATH_IMAGE023
，
Figure 49958DEST_PATH_IMAGE024
，
Figure 403710DEST_PATH_IMAGE025
is a parameter used for the learning of,
Figure 578340DEST_PATH_IMAGE026
is an activation function;
Figure 444665DEST_PATH_IMAGE027
for the time attention matrix calculated based on the time input data of the time attention feature extraction module,
Figure 688433DEST_PATH_IMAGE027
element (1) of
Figure 95144DEST_PATH_IMAGE028
Indicating the time of day
Figure 858700DEST_PATH_IMAGE015
And time of day
Figure 912238DEST_PATH_IMAGE016
The time dependency between the two is determined,
Figure 190773DEST_PATH_IMAGE029
extracting output data of a module for time attention features;
the spatial attention feature extraction module adopts the following formula to calculate:
Figure 604436DEST_PATH_IMAGE030
；
Figure 740276DEST_PATH_IMAGE002
；
wherein the content of the first and second substances,
Figure 948403DEST_PATH_IMAGE031
spatial input data for a spatial attention feature extraction module, said spatial input data being output data of said temporal attention feature extraction module,
Figure 714234DEST_PATH_IMAGE007
，
Figure 416742DEST_PATH_IMAGE008
，
Figure 951629DEST_PATH_IMAGE009
，
Figure 596237DEST_PATH_IMAGE010
，
Figure 301893DEST_PATH_IMAGE011
is a parameter used for the learning of,
Figure 791780DEST_PATH_IMAGE012
is an activation function;
Figure 384436DEST_PATH_IMAGE013
for the spatial attention matrix calculated based on the spatial input data of the spatial attention feature extraction module,
Figure 199945DEST_PATH_IMAGE013
element (1) of
Figure 691100DEST_PATH_IMAGE014
Representing wind farms
Figure 250258DEST_PATH_IMAGE015
Wind farm
Figure 697420DEST_PATH_IMAGE016
The spatial correlation between the two signals is determined,
Figure 467186DEST_PATH_IMAGE017
the output data of the spatial attention feature extraction module.
Further, the convolution calculation module comprises a graph convolution calculation module and a one-dimensional convolution calculation module, and the output of the graph convolution calculation module is used as the input of the one-dimensional convolution calculation module.
Further, the graph convolution calculating module calculates by using the following formula:
Figure 898167DEST_PATH_IMAGE032
；
Figure 729857DEST_PATH_IMAGE033
；
Figure 313416DEST_PATH_IMAGE034
-
Figure 1886DEST_PATH_IMAGE035
；
Figure 920164DEST_PATH_IMAGE036
；
wherein the content of the first and second substances,
Figure 804812DEST_PATH_IMAGE037
is a wind farm
Figure 23304DEST_PATH_IMAGE015
Wind farm
Figure 695725DEST_PATH_IMAGE016
The geographic location distance between the two or more,
Figure 570140DEST_PATH_IMAGE038
is that
Figure 312007DEST_PATH_IMAGE004
The variance of the geographical location distances of the individual wind farms,
Figure 853847DEST_PATH_IMAGE039
is a distance threshold, D is a node degree matrix for a wind farm,
Figure 87382DEST_PATH_IMAGE040
for the contiguous matrix of the plurality of wind farms,
Figure 465405DEST_PATH_IMAGE041
a Laplacian matrix for the plurality of wind farms;
Figure 36063DEST_PATH_IMAGE003
for the power characteristic data or the wind speed characteristic data,
Figure 416098DEST_PATH_IMAGE004
for the number of said plurality of wind farms,
Figure 86114DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 200700DEST_PATH_IMAGE006
the number of a plurality of moments;
parameter(s)
Figure 529045DEST_PATH_IMAGE042
Is a vector of coefficients of the polynomial,
Figure 45477DEST_PATH_IMAGE043
is a matrix
Figure 886394DEST_PATH_IMAGE041
Is determined by the maximum characteristic value of the image,
Figure 222697DEST_PATH_IMAGE044
which represents the Hadamard product of the two,
Figure 856197DEST_PATH_IMAGE045
an output for the graph convolution calculation module;
the iterative formula of the chebyshev polynomial is
Figure 227136DEST_PATH_IMAGE046
-
Figure 973375DEST_PATH_IMAGE047
Wherein
Figure 547707DEST_PATH_IMAGE048
。
Further, the model input data of the attention feature extraction module and the output of the convolution calculation module are connected by a residual error, and the formula is as follows:
Figure 732700DEST_PATH_IMAGE049
；
wherein the content of the first and second substances,
Figure 958145DEST_PATH_IMAGE050
the data is input for the model and,
Figure 124553DEST_PATH_IMAGE051
is the output of the convolution computation module and,
Figure 701028DEST_PATH_IMAGE052
is an input to the feature fusion module.
Further, the wind speed characterization data includes predicted wind speeds for a plurality of altitudes at a plurality of future times for the plurality of wind farms.
The embodiment of the present application further provides a wind farm power prediction device, including:
the data acquisition module is used for acquiring wind power of a plurality of wind power plants at a plurality of historical moments included in the wind power plant cluster as power characteristic data and predicted wind speeds of the plurality of wind power plants at a plurality of future moments as wind speed characteristic data;
the power prediction module is used for inputting the power characteristic data and the wind speed characteristic data of the wind power plants into a pre-established graph attention network model to obtain predicted power of the wind power plants, wherein the graph attention network model comprises an attention characteristic extraction module, a convolution calculation module and a characteristic fusion module, the power characteristic data is used as model input data, power characteristics are obtained after calculation of the attention characteristic extraction module and the convolution calculation module in sequence, the wind speed characteristic data is used as model input data, wind speed characteristics are obtained after calculation of the attention characteristic extraction module and the convolution calculation module in sequence, and the predicted power of the wind power plants are obtained after calculation of the power characteristics and the wind speed characteristics by the characteristic fusion module, the attention feature extraction module comprises a spatial attention feature extraction module.
Embodiments of the present application further provide an electronic device, including a processor and a machine-readable storage medium storing machine-executable instructions executable by the processor, the processor being caused by the machine-executable instructions to: the method for predicting the power of any wind power plant is realized.
An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the method for predicting the power of the wind farm is implemented.
Embodiments of the present application also provide a computer program product containing instructions, which when run on a computer, cause the computer to perform any of the wind farm power prediction methods described above.
The beneficial effect of this application includes:
according to the method provided by the embodiment of the application, a pre-established graph attention network model is used for power prediction of a wind power plant cluster comprising a plurality of wind power plants, wherein the wind power at a plurality of historical moments of the plurality of wind power plants and the predicted wind speed at a plurality of future moments of the plurality of wind power plants are used as model input data of the graph attention network model, the graph attention network model comprises an attention feature extraction module, and further the attention feature extraction module comprises a spatial attention feature extraction module, so that attention degrees of different parts of model input data can be distinguished, a spatial attention mechanism aiming at the model input data is realized, namely spatial correlation of wind power plant time is more fully considered in the power prediction process, and the accuracy of the power prediction of the wind power plants is improved.
Additional features and advantages of the application will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the application. The objectives and other advantages of the application may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The accompanying drawings are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application and not to limit the application. In the drawings:
FIG. 1 is a flowchart of a method for predicting power of a wind farm provided in an embodiment of the present application;
FIG. 2 is a schematic illustration of wind farm power prediction using a graphical attention network model in one embodiment of the present application;
FIG. 3 is a schematic illustration of wind farm power prediction using a graphical attention network model in another embodiment of the present application;
FIG. 4 is a schematic structural diagram of a wind farm power prediction device provided in an embodiment of the present application;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
In order to provide an implementation scheme for improving the wind farm power prediction accuracy, embodiments of the present application provide a wind farm power prediction method, a wind farm power prediction device, and an electronic device, and the following description is made in conjunction with the drawings of the specification for the preferred embodiments of the present application, it should be understood that the preferred embodiments described herein are only for illustrating and explaining the present application, and are not used to limit the present application. And the embodiments and features of the embodiments in the present application may be combined with each other without conflict.
The embodiment of the application provides a method for predicting power of a wind power plant, as shown in fig. 1, the method comprises the following steps:
step 11, acquiring wind power of a plurality of wind power plants at a plurality of historical moments included in a wind power plant cluster as power characteristic data, and acquiring predicted wind speeds of a plurality of wind power plants at a plurality of future moments as wind speed characteristic data;
step 12, inputting power characteristic data and wind speed characteristic data of a plurality of wind power plants into a pre-established graph attention network model to obtain predicted power of the plurality of wind power plants, wherein the graph attention network model comprises an attention characteristic extraction module, a convolution calculation module and a characteristic fusion module, the power characteristic data is used as model input data, power characteristics are obtained after calculation of the attention characteristic extraction module and the convolution calculation module in sequence, the wind speed characteristic data is used as model input data, wind speed characteristics are obtained after calculation of the attention characteristic extraction module and the convolution calculation module in sequence, predicted power of the plurality of wind power plants is obtained after calculation of the power characteristics and the wind speed characteristics through the characteristic fusion module, and the attention characteristic extraction module comprises a space attention characteristic extraction module.
In the method provided by the embodiment of the application, a pre-established graph attention network model is used for power prediction of a wind power plant cluster comprising a plurality of wind power plants, wherein the wind power at a plurality of historical moments of the plurality of wind power plants and the predicted wind speed at a plurality of future moments of the plurality of wind power plants are used as model input data of the graph attention network model, the graph attention network model comprises an attention feature extraction module, and further the attention feature extraction module comprises a spatial attention feature extraction module, so that attention degrees of different parts of model input data can be distinguished, a spatial attention mechanism aiming at the model input data is realized, namely spatial correlation of wind power plant time is more fully considered in a power prediction process, and the accuracy of the power prediction of the wind power plants is improved.
The method and apparatus provided herein are described in detail below with reference to the accompanying drawings using specific embodiments.
Fig. 2 is a schematic diagram of a power prediction of a wind farm by using an attention network model in an embodiment of the present application, where the attention network model includes an attention feature extraction module, a convolution calculation module, and a feature fusion module, which are connected in sequence.
The attention feature extraction module comprises a spatial attention feature extraction module (SAtt), and model input data of the attention network model is directly input into the spatial attention feature extraction module, namely the spatial input data of the spatial attention feature extraction module.
The convolution calculation module may include a graph convolution calculation module (GCN) and a one-dimensional convolution calculation module (Conv).
Furthermore, in the graph attention network model shown in fig. 2, the model input data of the attention feature extraction module and the output of the convolution calculation module may be connected by a residual error, that is, the space input data of the space attention feature extraction module and the output of the one-dimensional convolution calculation module are connected by a residual error, and a residual error connection structure is introduced, so that gradient dissipation caused by too complicated model can be avoided, and prediction accuracy is further improved.
When the wind farm power prediction is performed by using the graph attention power network model shown in fig. 2, the model input data may include two kinds of characteristic data, one is wind power at a plurality of historical moments of a plurality of wind farms included in the wind farm cluster as power characteristic data, and the other is predicted wind speeds at a plurality of future moments of the plurality of wind farms as wind speed characteristic data.
In the embodiment of the application, the number of the plurality of wind power plants included in the wind power plant cluster can be set according to the actual situation of the wind power plant cluster in actual prediction, and 20 wind power plants are taken as an example in the application for scheme description. The number of the plurality of historical time instants and the number of the plurality of future time instants can be flexibly selected and set based on actual needs, for example, the plurality of historical time instants can be selected as 40 historical time instants every 15 minutes, and the plurality of future time instants can be selected as 20 future time instants every 15 minutes.
In the embodiment of the present application, the wind power at the historical time may be obtained through actual measurement, the predicted wind speed at the future time may be obtained through wind speed Prediction, for example, the predicted wind speed is obtained through Numerical Weather Prediction (NWP), and in the Numerical Weather Prediction, a plurality of predictions at different heights are often performed, so in the embodiment of the present application, the predicted wind speeds at a plurality of heights, for example, the predicted wind speeds at 4 heights, may also be obtained.
When power prediction is performed, the two kinds of feature data are respectively input into the spatial attention feature extraction module, that is, as shown in fig. 2, the power feature data are used as model input data and are sequentially calculated by the spatial attention feature extraction module, the graph convolution calculation module and the one-dimensional convolution calculation module to obtain power features, and the wind speed feature data are used as model input data and are sequentially calculated by the spatial attention feature extraction module, the graph convolution calculation module and the one-dimensional convolution calculation module to obtain wind speed features.
In this embodiment of the application, the spatial attention feature extraction module may calculate by using the following formula:
Figure 424134DEST_PATH_IMAGE001
；
Figure 989238DEST_PATH_IMAGE002
；
wherein the content of the first and second substances,
Figure 405176DEST_PATH_IMAGE003
the space input data of the space attention characteristic extraction module is power characteristic data or wind speed characteristic data,
Figure 937789DEST_PATH_IMAGE004
for the number of a plurality of wind farms,
Figure 920044DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 854502DEST_PATH_IMAGE006
the number of a plurality of moments; when the power characteristic data is the wind speed power of 40 historical moments of 20 wind farms,
Figure 441342DEST_PATH_IMAGE004
is a total of 20, and is,
Figure 477562DEST_PATH_IMAGE005
the number of the carbon atoms is 1,
Figure 745732DEST_PATH_IMAGE006
and 40, when the wind speed characterization data is the predicted wind speed for 4 altitudes at 20 future times for 20 wind farms,
Figure 534697DEST_PATH_IMAGE004
is a total of 20, and is,
Figure 495699DEST_PATH_IMAGE005
is the number of the carbon atoms in the carbon atoms to be 4,
Figure 48909DEST_PATH_IMAGE006
is 40.
Figure 120770DEST_PATH_IMAGE007
，
Figure 29821DEST_PATH_IMAGE008
，
Figure 646878DEST_PATH_IMAGE009
，
Figure 703696DEST_PATH_IMAGE010
，
Figure 48089DEST_PATH_IMAGE011
Are parameters used for learning, which will be trained when training the model,
Figure 329422DEST_PATH_IMAGE012
is an activation function;
Figure 163386DEST_PATH_IMAGE013
for the spatial attention matrix calculated based on the spatial input data of the spatial attention feature extraction module,
Figure 910762DEST_PATH_IMAGE013
element (1) of
Figure 809579DEST_PATH_IMAGE014
Representing wind farms
Figure 693222DEST_PATH_IMAGE015
Wind farm
Figure 166928DEST_PATH_IMAGE016
The spatial correlation between the two signals is determined,
Figure 182027DEST_PATH_IMAGE017
the output data of the spatial attention feature extraction module.
Figure 133802DEST_PATH_IMAGE017
The calculation of (c) can be understood as that 1 softmax function is adopted to ensure that the sum of attention mechanisms on each wind farm is 1.
In this embodiment of the present application, the graph convolution calculating module may calculate by using the following formula:
Figure 871951DEST_PATH_IMAGE032
；
Figure 798450DEST_PATH_IMAGE033
；
Figure 254839DEST_PATH_IMAGE034
-
Figure 10305DEST_PATH_IMAGE035
；
Figure 868540DEST_PATH_IMAGE036
；
wherein the content of the first and second substances,
Figure 682386DEST_PATH_IMAGE037
is a wind farm
Figure 422809DEST_PATH_IMAGE015
Wind farm
Figure 716387DEST_PATH_IMAGE016
The geographic location distance between the two or more,
Figure 179861DEST_PATH_IMAGE038
is that
Figure 900692DEST_PATH_IMAGE004
The variance of the geographical location distances of the individual wind farms,
Figure 331673DEST_PATH_IMAGE039
is a distance threshold, D is a node degree matrix for a wind farm,
Figure 428942DEST_PATH_IMAGE040
for a contiguous matrix of a plurality of wind farms,
Figure 42195DEST_PATH_IMAGE041
laplacian matrices for a plurality of wind farms; wherein the distance threshold value
Figure 199507DEST_PATH_IMAGE039
May be half of the average of the elements in the distance matrix of the plurality of wind farms.
Figure 852205DEST_PATH_IMAGE003
Inputting data for the model, namely power characteristic data or wind speed characteristic data,
Figure 35056DEST_PATH_IMAGE004
for the number of a plurality of wind farms,
Figure 722389DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 785023DEST_PATH_IMAGE006
the number of the plurality of moments is the same as the meaning of the formula adopted by the space attention feature extraction module;
parameter(s)
Figure 973952DEST_PATH_IMAGE042
Is a vector of coefficients of the polynomial,
Figure 678603DEST_PATH_IMAGE043
is a matrix
Figure 954864DEST_PATH_IMAGE041
Is determined by the maximum characteristic value of the image,
Figure 204711DEST_PATH_IMAGE044
which represents the Hadamard product of the two,
Figure 628739DEST_PATH_IMAGE045
output of the graph convolution calculation module;
the iterative formula of the chebyshev polynomial is
Figure 340343DEST_PATH_IMAGE046
-
Figure 251536DEST_PATH_IMAGE047
Wherein
Figure 718289DEST_PATH_IMAGE048
。
In the embodiment of the application, in the formula adopted by the graph convolution calculation module, in order to dynamically adjust the spatial correlation between the wind power plants, compared with the currently commonly used graph convolution calculation formula, a spatial attention mechanism is adopted to describe the geographical position relationship between the wind power plants
Figure 567297DEST_PATH_IMAGE053
Modified to obtain an improved matrix
Figure 98903DEST_PATH_IMAGE054
Wherein
Figure 412073DEST_PATH_IMAGE044
Representing a Hadamard product.
In the embodiment of the application, after the graph convolution calculation module with the attention mechanism is adopted to extract the time-space correlation of the wind power plant cluster, one-dimensional convolution can be performed in the time dimension, that is, the one-dimensional convolution calculation module can calculate by adopting the following formula:
Figure 252990DEST_PATH_IMAGE055
wherein
Figure 841491DEST_PATH_IMAGE056
A standard convolution operation is shown as one of the standard convolution operations,
Figure 426056DEST_PATH_IMAGE057
is a parameter of a time domain convolution kernel having an activation function of
Figure 593732DEST_PATH_IMAGE058
。
If the residual error connection structure is not adopted, the output of the one-dimensional convolution calculation module is the extracted feature, namely when the model input data is the power feature data, the power feature is output, and when the model input data is the wind speed feature data, the wind speed feature is output.
If the space input data of the space attention feature extraction module and the output of the one-dimensional convolution calculation module are connected by adopting a residual error and a residual error connection structure is introduced, calculating by adopting the following formula based on the residual error connection structure:
Figure 356283DEST_PATH_IMAGE049
；
wherein the content of the first and second substances,
Figure 914303DEST_PATH_IMAGE050
the data is input for the model and,
Figure 99297DEST_PATH_IMAGE051
is the output of the convolution calculation module, namely the output of the one-dimensional convolution calculation module,
Figure 574009DEST_PATH_IMAGE052
the extracted features can be understood as the input of the feature fusion module, that is, when the model input data is the power feature data, the power feature is output, and when the model input data is the wind speed feature data, the wind speed feature is output.
The feature fusion module performs feature fusion on the input power feature and the wind speed feature, and the specific feature fusion mode adopted may be various feasible modes, which are not described in detail herein by way of example.
After feature fusion, the obtained feature fusion result may be directly used as a prediction result of power prediction, that is, predicted power of the wind farms at a plurality of future times, or in the graph attention network model, a full connection layer (not shown in fig. 2) is added after the feature fusion module, and based on the number of future times expected to be included in the prediction result, dimension reduction processing is performed on the feature fusion result obtained by feature fusion by using the full connection layer to obtain a specified number of predicted power, for example, predicted power of the wind farms at 16 future times is obtained, and an interval between adjacent 16 future times is 15 minutes.
As can be seen from fig. 2, in the attention network model for predicting the power of the wind farm cluster in consideration of the attention mechanism, attention on the space and a graph convolution algorithm are considered to be combined to extract features on the space, convolution along a time axis is adopted to extract the features on the time, residual connection is adopted to fuse the extracted features and original data, multi-task learning is finally adopted to realize prediction of the power of a plurality of wind farms, and the total power of the wind farm cluster can be obtained by summing the predicted power of each wind farm.
It is worth mentioning that because the architecture shown in fig. 2 is adopted, the power prediction of a single wind farm can be performed by the power network model according to the data of the adjacent wind farms, and the prediction accuracy of the single wind farm can be improved while the total prediction power of the wind farm cluster is improved. Even if different wind power plants belong to different operation subjects, the method can be realized by means of privacy calculation of multi-party safety calculation by adopting the attention network model.
Fig. 3 is a schematic diagram of a wind farm power prediction using a graph attention network model in another embodiment of the present application, where the graph attention network model includes an attention feature extraction module, a convolution calculation module, and a feature fusion module, which are connected in sequence.
The attention feature extraction module comprises a temporal attention feature extraction module (TAtt) and a spatial attention feature extraction module (SAtt), and the model input data of the graph attention network model is directly input into the temporal attention feature extraction module, i.e. the temporal input data of the temporal attention feature extraction module is also input, and the output of the temporal attention feature extraction module is used as the input of the spatial attention feature extraction module.
The convolution calculation module may include a graph convolution calculation module (GCN) and a one-dimensional convolution calculation module (Conv).
Furthermore, in the attention network model shown in fig. 3, the model input data of the attention feature extraction module and the output of the convolution calculation module may be connected by a residual error, that is, the time input data of the time attention feature extraction module and the output of the one-dimensional convolution calculation module are connected by a residual error, and a residual error connection structure is introduced, so that gradient dissipation caused by too complicated model can be avoided, and prediction accuracy is further improved.
When the wind farm power prediction is performed by using the graph attention network model shown in fig. 3, the model input data may include two kinds of characteristic data, one is wind power at a plurality of historical moments of a plurality of wind farms included in the wind farm cluster as power characteristic data, and the other is predicted wind speeds at a plurality of future moments of the plurality of wind farms as wind speed characteristic data.
In the embodiment of the application, the number of the plurality of wind power plants included in the wind power plant cluster can be set according to the actual situation of the wind power plant cluster in actual prediction, and 20 wind power plants are taken as an example in the application for scheme description. The number of the plurality of historical time instants and the number of the plurality of future time instants can be flexibly selected and set based on actual needs, for example, the plurality of historical time instants can be selected as 40 historical time instants every 15 minutes, and the plurality of future time instants can be selected as 20 future time instants every 15 minutes.
In the embodiment of the present application, the wind power at the historical time may be obtained through actual measurement, the predicted wind speed at the future time may be obtained through wind speed Prediction, for example, the predicted wind speed is obtained through Numerical Weather Prediction (NWP), and in the Numerical Weather Prediction, a plurality of predictions at different heights are often performed, so in the embodiment of the present application, the predicted wind speeds at a plurality of heights, for example, the predicted wind speeds at 4 heights, may also be obtained.
When power prediction is performed, the two kinds of feature data are respectively input into the time attention feature extraction module, that is, as shown in fig. 3, the power feature data are used as model input data, and are sequentially calculated by the time attention feature extraction module, the space attention feature extraction module, the graph convolution calculation module and the one-dimensional convolution calculation module to obtain power features, and the wind speed feature data are used as model input data, and are sequentially calculated by the time attention feature extraction module, the space attention feature extraction module, the graph convolution calculation module and the one-dimensional convolution calculation module to obtain wind speed features.
In this embodiment of the application, the time attention feature extraction module may calculate by using the following formula:
Figure 756729DEST_PATH_IMAGE018
；
Figure 67624DEST_PATH_IMAGE019
；
Figure 541462DEST_PATH_IMAGE020
；
wherein the content of the first and second substances,
Figure 355834DEST_PATH_IMAGE003
the time input data of the time attention characteristic extraction module is power characteristic data or wind speed characteristic data,
Figure 975035DEST_PATH_IMAGE004
for the number of a plurality of wind farms,
Figure 773226DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 286641DEST_PATH_IMAGE006
the number of a plurality of moments; characteristic of current powerWhen the data is the wind speed power of 40 historical moments of 20 wind farms,
Figure 486678DEST_PATH_IMAGE004
is a total of 20, and is,
Figure 276779DEST_PATH_IMAGE005
the number of the carbon atoms is 1,
Figure 578579DEST_PATH_IMAGE006
and 40, when the wind speed characterization data is the predicted wind speed for 4 altitudes at 20 future times for 20 wind farms,
Figure 846749DEST_PATH_IMAGE004
is a total of 20, and is,
Figure 947298DEST_PATH_IMAGE005
is the number of the carbon atoms in the carbon atoms to be 4,
Figure 377142DEST_PATH_IMAGE006
is 40.
Figure 149926DEST_PATH_IMAGE021
，
Figure 769258DEST_PATH_IMAGE022
，
Figure 678308DEST_PATH_IMAGE023
，
Figure 544633DEST_PATH_IMAGE024
，
Figure 539133DEST_PATH_IMAGE025
Are parameters used for learning, which will be trained when training the model,
Figure 451902DEST_PATH_IMAGE026
is an activation function;
Figure 481038DEST_PATH_IMAGE027
for the time attention matrix calculated based on the time input data of the time attention feature extraction module,
Figure 252684DEST_PATH_IMAGE027
element (1) of
Figure 547531DEST_PATH_IMAGE028
Indicating the time of day
Figure 695615DEST_PATH_IMAGE015
And time of day
Figure 625263DEST_PATH_IMAGE016
The time dependency between the two is determined,
Figure 833390DEST_PATH_IMAGE029
extracting output data of a module for time attention features;
Figure 802483DEST_PATH_IMAGE059
the calculation of (c) can be understood as that 1 softmax function is adopted to ensure that the sum of attention mechanisms on each wind farm is 1.
In this embodiment of the application, the spatial attention feature extraction module may calculate by using the following formula:
Figure 754259DEST_PATH_IMAGE030
；
Figure 243140DEST_PATH_IMAGE002
；
wherein the content of the first and second substances,
Figure 684486DEST_PATH_IMAGE031
the spatial input data of the spatial attention feature extraction module are output data of the temporal attention feature extraction module;
Figure 140875DEST_PATH_IMAGE007
，
Figure 148539DEST_PATH_IMAGE008
，
Figure 741194DEST_PATH_IMAGE009
，
Figure 556703DEST_PATH_IMAGE010
，
Figure 234809DEST_PATH_IMAGE011
are parameters used for learning, which will be trained when training the model,
Figure 341437DEST_PATH_IMAGE012
is an activation function;
Figure 788598DEST_PATH_IMAGE013
for the spatial attention matrix calculated based on the spatial input data of the spatial attention feature extraction module,
Figure 821014DEST_PATH_IMAGE013
element (1) of
Figure 986416DEST_PATH_IMAGE014
Representing wind farms
Figure 880423DEST_PATH_IMAGE015
Wind farm
Figure 198403DEST_PATH_IMAGE016
The spatial correlation between the two signals is determined,
Figure 90136DEST_PATH_IMAGE017
the output data of the spatial attention feature extraction module.
Figure 742834DEST_PATH_IMAGE017
The calculation of (c) can be understood as that 1 softmax function is adopted to ensure that the sum of attention mechanisms on each wind farm is 1.
In this embodiment of the present application, the graph convolution calculating module may calculate by using the following formula:
Figure 692729DEST_PATH_IMAGE032
；
Figure 380062DEST_PATH_IMAGE033
；
Figure 708275DEST_PATH_IMAGE034
-
Figure 582690DEST_PATH_IMAGE035
；
Figure 569232DEST_PATH_IMAGE036
；
wherein the content of the first and second substances,
Figure 111072DEST_PATH_IMAGE037
is a wind farm
Figure 344607DEST_PATH_IMAGE015
Wind farm
Figure 17903DEST_PATH_IMAGE016
The geographic location distance between the two or more,
Figure 260665DEST_PATH_IMAGE038
is that
Figure 657011DEST_PATH_IMAGE004
The variance of the geographical location distances of the individual wind farms,
Figure 77760DEST_PATH_IMAGE039
is a distance threshold, D is a node degree matrix for a wind farm,
Figure 723505DEST_PATH_IMAGE040
for a contiguous matrix of a plurality of wind farms,
Figure 769958DEST_PATH_IMAGE041
laplacian matrices for a plurality of wind farms; wherein the distance threshold value
Figure 538587DEST_PATH_IMAGE039
May be half of the average of the elements in the distance matrix of the plurality of wind farms.
Figure 113925DEST_PATH_IMAGE003
Inputting data for the model, namely power characteristic data or wind speed characteristic data,
Figure 450229DEST_PATH_IMAGE004
for the number of a plurality of wind farms,
Figure 582264DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 953202DEST_PATH_IMAGE006
the number of the plurality of moments is the same as the meaning of the formula adopted by the space attention feature extraction module;
parameter(s)
Figure 699441DEST_PATH_IMAGE042
Is a vector of coefficients of the polynomial,
Figure 788620DEST_PATH_IMAGE043
is a matrix
Figure 957302DEST_PATH_IMAGE041
Is determined by the maximum characteristic value of the image,
Figure 182747DEST_PATH_IMAGE044
which represents the Hadamard product of the two,
Figure 912937DEST_PATH_IMAGE045
output of the graph convolution calculation module;
the iterative formula of the chebyshev polynomial is
Figure 223832DEST_PATH_IMAGE046
-
Figure 884621DEST_PATH_IMAGE047
Wherein
Figure 230152DEST_PATH_IMAGE048
。
In the embodiment of the application, in the formula adopted by the graph convolution calculation module, in order to dynamically adjust the spatial correlation between the wind power plants, compared with the currently commonly used graph convolution calculation formula, a spatial attention mechanism is adopted to describe the geographical position relationship between the wind power plants
Figure 644426DEST_PATH_IMAGE053
Modified to obtain an improved matrix
Figure 442618DEST_PATH_IMAGE054
Wherein
Figure 172677DEST_PATH_IMAGE044
Representing a Hadamard product.
In the embodiment of the application, after the graph convolution calculation module with the attention mechanism is adopted to extract the time-space correlation of the wind power plant cluster, one-dimensional convolution can be performed in the time dimension, that is, the one-dimensional convolution calculation module can calculate by adopting the following formula:
Figure 123446DEST_PATH_IMAGE055
wherein
Figure 382389DEST_PATH_IMAGE056
A standard convolution operation is shown as one of the standard convolution operations,
Figure 667877DEST_PATH_IMAGE057
is a parameter of a time domain convolution kernel having an activation function of
Figure 247632DEST_PATH_IMAGE058
。
If the residual error connection structure is not adopted, the output of the one-dimensional convolution calculation module is the extracted feature, namely when the model input data is the power feature data, the power feature is output, and when the model input data is the wind speed feature data, the wind speed feature is output.
If the space input data of the space attention feature extraction module and the output of the one-dimensional convolution calculation module are connected by adopting a residual error and a residual error connection structure is introduced, calculating by adopting the following formula based on the residual error connection structure:
Figure 302175DEST_PATH_IMAGE049
；
wherein the content of the first and second substances,
Figure 997599DEST_PATH_IMAGE050
the data is input for the model and,
Figure 504804DEST_PATH_IMAGE051
is the output of the convolution calculation module, namely the output of the one-dimensional convolution calculation module,
Figure 327397DEST_PATH_IMAGE052
the extracted features can be understood as the input of the feature fusion module, that is, when the model input data is the power feature data, the power feature is output, and when the model input data is the wind speed feature data, the wind speed feature is output.
The feature fusion module performs feature fusion on the input power feature and the wind speed feature, and the specific feature fusion mode adopted may be various feasible modes, which are not described in detail herein by way of example.
After feature fusion, the obtained feature fusion result may be directly used as a prediction result of power prediction, that is, predicted power of the wind farms at a plurality of future times, or in the graph attention network model, a full connection layer (not shown in fig. 2) is added after the feature fusion module, and based on the number of future times expected to be included in the prediction result, dimension reduction processing is performed on the feature fusion result obtained by feature fusion by using the full connection layer to obtain a specified number of predicted power, for example, predicted power of the wind farms at 16 future times is obtained, and an interval between adjacent 16 future times is 15 minutes.
As can be seen from fig. 3, in the graph attention network model for predicting the power of the wind farm cluster considering the attention mechanism, the attention in time and the attention in space are considered, the feature in space is extracted by combining with the graph convolution algorithm, the feature in time is extracted by convolution along the time axis, residual connection is adopted, the extracted feature and original data are fused, finally, multi-task learning is adopted to predict the power of a plurality of wind farms, and the total power of the wind farm cluster can be obtained by summing the predicted power of each wind farm.
It is worth mentioning that because the framework shown in fig. 3 is adopted, the graph notes that the power prediction of a single wind farm can be performed by the power network model by using data of adjacent wind farms for reference, and the prediction accuracy of the single wind farm can be improved while the total predicted power of the wind farm cluster is improved. Even if different wind power plants belong to different operation subjects, the method can be realized by means of privacy calculation of multi-party safety calculation by adopting the attention network model.
Based on the same inventive concept, according to the wind farm power prediction method provided in the foregoing embodiment of the present application, correspondingly, another embodiment of the present application further provides a wind farm power prediction device, a schematic structural diagram of which is shown in fig. 4, and specifically includes:
the data acquisition module 41 is configured to acquire wind power at a plurality of historical moments of a plurality of wind farms included in the wind farm cluster as power characteristic data, and acquire predicted wind speeds at a plurality of future moments of the plurality of wind farms as wind speed characteristic data;
the power prediction module 42 is configured to input power feature data and wind speed feature data of a plurality of wind power plants into a pre-established graph attention network model to obtain predicted power of the plurality of wind power plants, where the graph attention network model includes an attention feature extraction module, a convolution calculation module and a feature fusion module, the power feature data is used as model input data, power features are obtained after calculation of the attention feature extraction module and the convolution calculation module in sequence, the wind speed feature data is used as model input data, wind speed features are obtained after calculation of the attention feature extraction module and the convolution calculation module in sequence, the power features and the wind speed features are calculated by the feature fusion module to obtain predicted power of the plurality of wind power plants, and the attention feature extraction module includes a spatial attention feature extraction module.
Further, the spatial attention feature extraction module performs calculation by using the following formula:
Figure 33185DEST_PATH_IMAGE001
；
Figure 899510DEST_PATH_IMAGE002
；
wherein the content of the first and second substances,
Figure 942946DEST_PATH_IMAGE003
the space input data of the space attention characteristic extraction module is power characteristic data or wind speed characteristic data,
Figure 818498DEST_PATH_IMAGE004
for the number of a plurality of wind farms,
Figure 582055DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 370013DEST_PATH_IMAGE006
the number of the plurality of time instants,
Figure 851810DEST_PATH_IMAGE007
，
Figure 62212DEST_PATH_IMAGE008
，
Figure 929542DEST_PATH_IMAGE009
，
Figure 403249DEST_PATH_IMAGE010
，
Figure 169080DEST_PATH_IMAGE011
is a parameter used for the learning of,
Figure 871588DEST_PATH_IMAGE012
is an activation function;
Figure 406474DEST_PATH_IMAGE013
for the spatial attention matrix calculated based on the spatial input data of the spatial attention feature extraction module,
Figure 785503DEST_PATH_IMAGE013
element (1) of
Figure 290827DEST_PATH_IMAGE014
Representing wind farms
Figure 46293DEST_PATH_IMAGE015
Wind farm
Figure 638949DEST_PATH_IMAGE016
The spatial correlation between the two signals is determined,
Figure 939611DEST_PATH_IMAGE017
the output data of the spatial attention feature extraction module.
Further, the attention feature extraction module further comprises a temporal attention feature extraction module, an output of which is used as an input of the spatial attention feature extraction module.
Further, the time attention feature extraction module adopts the following formula to calculate:
Figure 883296DEST_PATH_IMAGE018
；
Figure 239191DEST_PATH_IMAGE019
；
Figure 201200DEST_PATH_IMAGE020
；
wherein the content of the first and second substances,
Figure 922031DEST_PATH_IMAGE003
the time input data of the time attention characteristic extraction module is power characteristic data or wind speed characteristic data,
Figure 884171DEST_PATH_IMAGE004
for the number of a plurality of wind farms,
Figure 732172DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 299420DEST_PATH_IMAGE006
the number of the plurality of time instants,
Figure 253470DEST_PATH_IMAGE021
，
Figure 158365DEST_PATH_IMAGE022
，
Figure 793746DEST_PATH_IMAGE023
，
Figure 277817DEST_PATH_IMAGE024
，
Figure 356762DEST_PATH_IMAGE025
is a parameter used for the learning of,
Figure 496757DEST_PATH_IMAGE026
is an activation function;
Figure 670249DEST_PATH_IMAGE027
for the time attention matrix calculated based on the time input data of the time attention feature extraction module,
Figure 212089DEST_PATH_IMAGE027
element (1) of
Figure 757208DEST_PATH_IMAGE028
Indicating the time of day
Figure 118920DEST_PATH_IMAGE015
And time of day
Figure 909152DEST_PATH_IMAGE016
The time dependency between the two is determined,
Figure 305499DEST_PATH_IMAGE029
extracting output data of a module for time attention features;
the spatial attention feature extraction module adopts the following formula to calculate:
Figure 975514DEST_PATH_IMAGE030
；
Figure 824522DEST_PATH_IMAGE002
；
wherein the content of the first and second substances,
Figure 111454DEST_PATH_IMAGE031
the spatial input data of the spatial attention feature extraction module is the output data of the temporal attention feature extraction module,
Figure 159044DEST_PATH_IMAGE007
，
Figure 734382DEST_PATH_IMAGE008
，
Figure 86997DEST_PATH_IMAGE009
，
Figure 468300DEST_PATH_IMAGE010
，
Figure 573659DEST_PATH_IMAGE011
is a parameter used for the learning of,
Figure 365903DEST_PATH_IMAGE012
is an activation function;
Figure 455082DEST_PATH_IMAGE013
for the spatial attention matrix calculated based on the spatial input data of the spatial attention feature extraction module,
Figure 577759DEST_PATH_IMAGE013
element (1) of
Figure 553936DEST_PATH_IMAGE014
Representing wind farms
Figure 798973DEST_PATH_IMAGE015
Wind farm
Figure 109868DEST_PATH_IMAGE016
The spatial correlation between the two signals is determined,
Figure 22854DEST_PATH_IMAGE017
the output data of the spatial attention feature extraction module.
Further, the convolution calculation module comprises a graph convolution calculation module and a one-dimensional convolution calculation module, and the output of the graph convolution calculation module is used as the input of the one-dimensional convolution calculation module.
Further, the graph convolution calculating module calculates by using the following formula:
Figure 368385DEST_PATH_IMAGE032
；
Figure 518743DEST_PATH_IMAGE033
；
Figure 67668DEST_PATH_IMAGE034
-
Figure 532147DEST_PATH_IMAGE035
；
Figure 528922DEST_PATH_IMAGE036
；
wherein the content of the first and second substances,
Figure 302712DEST_PATH_IMAGE037
is a wind farm
Figure 588199DEST_PATH_IMAGE015
Wind farm
Figure 387528DEST_PATH_IMAGE016
The geographic location distance between the two or more,
Figure 192804DEST_PATH_IMAGE038
is that
Figure 888228DEST_PATH_IMAGE004
The variance of the geographical location distances of the individual wind farms,
Figure 661012DEST_PATH_IMAGE039
is a distance threshold, D is a node degree matrix for a wind farm,
Figure 516229DEST_PATH_IMAGE040
for a contiguous matrix of a plurality of wind farms,
Figure 690858DEST_PATH_IMAGE041
laplacian matrices for a plurality of wind farms;
Figure 88341DEST_PATH_IMAGE003
for the power characteristic data or the wind speed characteristic data,
Figure 99154DEST_PATH_IMAGE004
for the number of a plurality of wind farms,
Figure 443547DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 472683DEST_PATH_IMAGE006
the number of a plurality of moments;
parameter(s)
Figure 821494DEST_PATH_IMAGE042
Is a vector of coefficients of the polynomial,
Figure 303291DEST_PATH_IMAGE043
is a matrix
Figure 716955DEST_PATH_IMAGE041
Is determined by the maximum characteristic value of the image,
Figure 335018DEST_PATH_IMAGE044
which represents the Hadamard product of the two,
Figure 356195DEST_PATH_IMAGE045
output of the graph convolution calculation module;
the iterative formula of the chebyshev polynomial is
Figure 59708DEST_PATH_IMAGE046
-
Figure 11484DEST_PATH_IMAGE047
Wherein
Figure 64147DEST_PATH_IMAGE048
。
Furthermore, residual errors are connected between the model input data of the attention feature extraction module and the output of the convolution calculation module, and the adopted formula is as follows:
Figure 708755DEST_PATH_IMAGE049
；
wherein the content of the first and second substances,
Figure 165144DEST_PATH_IMAGE050
the data is input for the model and,
Figure 671343DEST_PATH_IMAGE051
is the output of the convolution calculation module and,
Figure 998419DEST_PATH_IMAGE052
is the input to the feature fusion module.
Further, the wind speed characterization data includes predicted wind speeds for a plurality of altitudes at a plurality of future times for a plurality of wind farms.
The functions of the above modules may correspond to the corresponding processing steps in the flows shown in fig. 1 to fig. 3, and are not described herein again.
The wind farm power prediction device provided by the embodiment of the application can be realized by a computer program. It should be understood by those skilled in the art that the above-mentioned module division is only one of many module division, and if the division is performed into other modules or not, it is within the scope of the present application as long as the wall painting and printing apparatus has the above-mentioned functions.
Based on the same inventive concept, according to the wind farm power prediction method provided in the foregoing embodiment of the present application, correspondingly, another embodiment of the present application further provides an electronic device, a schematic structural diagram of which is shown in fig. 5, and specifically includes: a processor 51 and a machine-readable storage medium 52, the machine-readable storage medium 52 storing machine-executable instructions executable by the processor 51, the processor 51 being caused by the machine-executable instructions to: the method for predicting the power of any wind power plant is realized.
An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the method for predicting the power of the wind farm is implemented.
Embodiments of the present application also provide a computer program product containing instructions, which when run on a computer, cause the computer to perform any of the wind farm power prediction methods described above.
The machine-readable storage medium in the electronic device may include a Random Access Memory (RAM) and a Non-Volatile Memory (NVM), such as at least one disk Memory. Optionally, the memory may also be at least one memory device located remotely from the processor.
The Processor may be a general-purpose Processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but also Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components.
All the embodiments in the present specification are described in a related manner, and the same and similar parts among the embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the apparatus, the electronic device, the computer-readable storage medium, and the computer program product embodiment, since they are substantially similar to the method embodiment, the description is relatively simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the spirit and scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.

Claims (11)

1. A wind farm power prediction method is characterized by comprising the following steps:
acquiring wind power of a plurality of wind power plants at a plurality of historical moments included in a wind power plant cluster as power characteristic data, and acquiring predicted wind speeds of the plurality of wind power plants at a plurality of future moments as wind speed characteristic data;
inputting the power characteristic data and the wind speed characteristic data of the wind power plants into a pre-established graph attention network model to obtain the predicted power of the wind power plants, wherein the graph attention network model comprises an attention feature extraction module, a convolution calculation module and a feature fusion module, the power characteristic data is used as model input data, and is sequentially calculated by the attention characteristic extraction module and the convolution calculation module to obtain power characteristics, the wind speed characteristic data is used as model input data, and wind speed characteristics are obtained after calculation of the attention characteristic extraction module and the convolution calculation module in sequence, and after the power characteristics and the wind speed characteristics are calculated by the characteristic fusion module, the predicted power of the wind power plants is obtained, and the attention characteristic extraction module comprises a space attention characteristic extraction module.
2. The method of claim 1, wherein the spatial attention feature extraction module calculates using the following formula:
Figure 8226DEST_PATH_IMAGE001
；
Figure 702644DEST_PATH_IMAGE002
；
wherein the content of the first and second substances,
Figure 527380DEST_PATH_IMAGE003
spatial input data for a spatial attention feature extraction module, the spatial input data being the power feature data or the wind speed feature data,
Figure 20547DEST_PATH_IMAGE004
for the number of said plurality of wind farms,
Figure 6958DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 703518DEST_PATH_IMAGE006
the number of the plurality of time instants,
Figure 551520DEST_PATH_IMAGE007
，
Figure 384347DEST_PATH_IMAGE008
，
Figure 807238DEST_PATH_IMAGE009
，
Figure 977712DEST_PATH_IMAGE010
，
Figure 878672DEST_PATH_IMAGE011
is a parameter used for the learning of,
Figure 831585DEST_PATH_IMAGE012
is an activation function;
Figure 910530DEST_PATH_IMAGE013
for the spatial attention matrix calculated based on the spatial input data of the spatial attention feature extraction module,
Figure 316104DEST_PATH_IMAGE013
element (1) of
Figure 755176DEST_PATH_IMAGE014
Representing wind farms
Figure 297015DEST_PATH_IMAGE015
Wind farm
Figure 45397DEST_PATH_IMAGE016
The spatial correlation between the two signals is determined,
Figure 938267DEST_PATH_IMAGE017
the output data of the spatial attention feature extraction module.
3. The method of claim 1, wherein the attention feature extraction module further comprises a temporal attention feature extraction module, an output of the temporal attention feature extraction module being an input to the spatial attention feature extraction module.
4. The method of claim 3, wherein the temporal attention feature extraction module calculates using the formula:
Figure 181030DEST_PATH_IMAGE018
；
Figure 593687DEST_PATH_IMAGE019
；
Figure 529282DEST_PATH_IMAGE020
；
wherein the content of the first and second substances,
Figure 909448DEST_PATH_IMAGE003
time input data for a time attention feature extraction module, the time input data being the power feature data or the wind speed feature data,
Figure 473678DEST_PATH_IMAGE004
for the number of said plurality of wind farms,
Figure 990110DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 96606DEST_PATH_IMAGE006
the number of the plurality of time instants,
Figure 183642DEST_PATH_IMAGE021
，
Figure 768207DEST_PATH_IMAGE022
，
Figure 185151DEST_PATH_IMAGE023
，
Figure 728128DEST_PATH_IMAGE024
，
Figure 82886DEST_PATH_IMAGE025
is a parameter used for the learning of,
Figure 221874DEST_PATH_IMAGE026
is an activation function;
Figure 712898DEST_PATH_IMAGE027
for the time attention matrix calculated based on the time input data of the time attention feature extraction module,
Figure 161197DEST_PATH_IMAGE027
element (1) of
Figure 989869DEST_PATH_IMAGE028
Indicating the time of day
Figure 916237DEST_PATH_IMAGE015
And time of day
Figure 261768DEST_PATH_IMAGE016
The time dependency between the two is determined,
Figure 631700DEST_PATH_IMAGE029
extracting output data of a module for time attention features;
the spatial attention feature extraction module adopts the following formula to calculate:
Figure 429892DEST_PATH_IMAGE030
；
Figure 425530DEST_PATH_IMAGE002
；
wherein the content of the first and second substances,
Figure 874835DEST_PATH_IMAGE031
spatial input data for a spatial attention feature extraction module, said spatial input data being output data of said temporal attention feature extraction module,
Figure 664936DEST_PATH_IMAGE007
，
Figure 950424DEST_PATH_IMAGE008
，
Figure 500485DEST_PATH_IMAGE009
，
Figure 555029DEST_PATH_IMAGE010
，
Figure 250452DEST_PATH_IMAGE011
is a parameter used for the learning of,
Figure 23236DEST_PATH_IMAGE012
is an activation function;
Figure 359014DEST_PATH_IMAGE013
for the spatial attention matrix calculated based on the spatial input data of the spatial attention feature extraction module,
Figure 533643DEST_PATH_IMAGE013
element (1) of
Figure 665547DEST_PATH_IMAGE014
Representing wind farms
Figure 660048DEST_PATH_IMAGE015
Wind farm
Figure 286332DEST_PATH_IMAGE016
The spatial correlation between the two signals is determined,
Figure 315468DEST_PATH_IMAGE017
the output data of the spatial attention feature extraction module.
5. The method of any of claims 1-4, wherein the convolution computation module comprises a graph convolution computation module and a one-dimensional convolution computation module, an output of the graph convolution computation module being an input to the one-dimensional convolution computation module.
6. The method of claim 5, wherein the graph convolution computation module computes using the following equation:
Figure 618274DEST_PATH_IMAGE032
；
Figure 614917DEST_PATH_IMAGE033
；
Figure 28581DEST_PATH_IMAGE034
-
Figure 177803DEST_PATH_IMAGE035
；
Figure 402242DEST_PATH_IMAGE036
；
wherein the content of the first and second substances,
Figure 371335DEST_PATH_IMAGE037
is a wind farm
Figure 854269DEST_PATH_IMAGE015
Wind farm
Figure 110194DEST_PATH_IMAGE016
The geographic location distance between the two or more,
Figure 754802DEST_PATH_IMAGE038
is that
Figure 476770DEST_PATH_IMAGE004
The variance of the geographical location distances of the individual wind farms,
Figure 232237DEST_PATH_IMAGE039
is a distance threshold, D is a node degree matrix for a wind farm,
Figure 841204DEST_PATH_IMAGE040
for the contiguous matrix of the plurality of wind farms,
Figure 391134DEST_PATH_IMAGE041
a Laplacian matrix for the plurality of wind farms;
Figure 600398DEST_PATH_IMAGE003
for the power characteristic data or the wind speed characteristic data,
Figure 408823DEST_PATH_IMAGE004
for the number of said plurality of wind farms,
Figure 387144DEST_PATH_IMAGE005
for the number of features of each wind farm,
Figure 639133DEST_PATH_IMAGE006
the number of a plurality of moments;
parameter(s)
Figure 555268DEST_PATH_IMAGE042
Is a vector of coefficients of the polynomial,
Figure 652537DEST_PATH_IMAGE043
is a matrix
Figure 750943DEST_PATH_IMAGE041
Is determined by the maximum characteristic value of the image,
Figure 908255DEST_PATH_IMAGE044
which represents the Hadamard product of the two,
Figure 78729DEST_PATH_IMAGE045
an output for the graph convolution calculation module;
the iterative formula of the chebyshev polynomial is
Figure 979689DEST_PATH_IMAGE046
-
Figure 932602DEST_PATH_IMAGE047
Wherein
Figure 11547DEST_PATH_IMAGE048
。
7. The method of any one of claims 1-4, wherein a residual connection is used between the model input data of the attention feature extraction module and the output of the convolution computation module, using the formula:
Figure 417121DEST_PATH_IMAGE049
；
wherein the content of the first and second substances,
Figure 856193DEST_PATH_IMAGE050
the data is input for the model and,
Figure 912879DEST_PATH_IMAGE051
is the output of the convolution computation module and,
Figure 677573DEST_PATH_IMAGE052
is an input to the feature fusion module.
8. A method according to any of claims 1-4, wherein said wind speed characteristic data comprises predicted wind speeds at a plurality of altitudes at a plurality of future times of said plurality of wind farms.
9. A wind farm power prediction device, comprising:
the data acquisition module is used for acquiring wind power of a plurality of wind power plants at a plurality of historical moments included in the wind power plant cluster as power characteristic data and predicted wind speeds of the plurality of wind power plants at a plurality of future moments as wind speed characteristic data;
the power prediction module is used for inputting the power characteristic data and the wind speed characteristic data of the wind power plants into a pre-established graph attention network model to obtain predicted power of the wind power plants, wherein the graph attention network model comprises an attention characteristic extraction module, a convolution calculation module and a characteristic fusion module, the power characteristic data is used as model input data, power characteristics are obtained after calculation of the attention characteristic extraction module and the convolution calculation module in sequence, the wind speed characteristic data is used as model input data, wind speed characteristics are obtained after calculation of the attention characteristic extraction module and the convolution calculation module in sequence, and the predicted power of the wind power plants are obtained after calculation of the power characteristics and the wind speed characteristics by the characteristic fusion module, the attention feature extraction module comprises a spatial attention feature extraction module.
10. An electronic device comprising a processor and a machine-readable storage medium storing machine-executable instructions executable by the processor, the processor being caused by the machine-executable instructions to: carrying out the method of any one of claims 1 to 8.
11. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method of any one of claims 1 to 8.
CN202111568234.2A 2021-12-21 2021-12-21 Wind power plant power prediction method and device and electronic equipment Pending CN113962495A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111568234.2A CN113962495A (en) 2021-12-21 2021-12-21 Wind power plant power prediction method and device and electronic equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111568234.2A CN113962495A (en) 2021-12-21 2021-12-21 Wind power plant power prediction method and device and electronic equipment

Publications (1)

Publication Number Publication Date
CN113962495A true CN113962495A (en) 2022-01-21

Family

ID=79473351

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111568234.2A Pending CN113962495A (en) 2021-12-21 2021-12-21 Wind power plant power prediction method and device and electronic equipment

Country Status (1)

Country Link
CN (1) CN113962495A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114118638A (en) * 2022-01-28 2022-03-01 华控清交信息科技(北京)有限公司 Wind power plant power prediction method, GBDT model transverse training method and device

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111292562A (en) * 2020-05-12 2020-06-16 北京航空航天大学 Aviation flow prediction method
CN112529282A (en) * 2020-12-03 2021-03-19 清华大学 Wind power plant cluster short-term power prediction method based on space-time graph convolutional neural network
CN113705880A (en) * 2021-08-25 2021-11-26 杭州远眺科技有限公司 Traffic speed prediction method and device based on space-time attention diagram convolutional network

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111292562A (en) * 2020-05-12 2020-06-16 北京航空航天大学 Aviation flow prediction method
CN112529282A (en) * 2020-12-03 2021-03-19 清华大学 Wind power plant cluster short-term power prediction method based on space-time graph convolutional neural network
CN113705880A (en) * 2021-08-25 2021-11-26 杭州远眺科技有限公司 Traffic speed prediction method and device based on space-time attention diagram convolutional network

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114118638A (en) * 2022-01-28 2022-03-01 华控清交信息科技(北京)有限公司 Wind power plant power prediction method, GBDT model transverse training method and device
CN114118638B (en) * 2022-01-28 2022-04-19 华控清交信息科技(北京)有限公司 Wind power plant power prediction method, GBDT model transverse training method and device

Similar Documents

Publication Publication Date Title
Feng et al. A data-driven multi-model methodology with deep feature selection for short-term wind forecasting
Ahmed et al. An intelligent framework for short-term multi-step wind speed forecasting based on Functional Networks
US11835677B2 (en) Systems and methods for selecting global climate simulation models for training neural network climate forecasting models
US20230128989A1 (en) Systems and methods of data preprocessing and augmentation for neural network climate forecasting models
Mu et al. Hourly and daily urban water demand predictions using a long short-term memory based model
Bamisile et al. Comparison of machine learning and deep learning algorithms for hourly global/diffuse solar radiation predictions
CN110686633B (en) Landslide displacement prediction method and device and electronic equipment
Pourhabib et al. Short-term wind speed forecast using measurements from multiple turbines in a wind farm
Xue et al. Parameter uncertainty analysis of surface flow and sediment yield in the Huolin Basin, China
Jiang et al. Global horizontal radiation forecast using forward regression on a quadratic kernel support vector machine: Case study of the Tibet Autonomous Region in China
Schefzik et al. Ensemble postprocessing methods incorporating dependence structures
CN114330935A (en) New energy power prediction method and system based on multiple combined strategy integrated learning
CN111815043B (en) Flood flow prediction method and system based on storm characteristics
Yousuf et al. Statistical wind speed forecasting models for small sample datasets: Problems, Improvements, and prospects
CN113962495A (en) Wind power plant power prediction method and device and electronic equipment
Jia et al. Daily reference evapotranspiration prediction for irrigation scheduling decisions based on the hybrid PSO-LSTM model
Sabri et al. Accurate photovoltaic power prediction models based on deep convolutional neural networks and gated recurrent units
Khatun et al. Two novel error-updating model frameworks for short-to-medium range streamflow forecasting using bias-corrected rainfall inputs: Development and comparative assessment
Smith et al. Testing probabilistic adaptive real‐time flood forecasting models
CN112596126A (en) Wind speed prediction model training method, prediction method and device and electronic equipment
Escalante-Sandoval et al. Regional monthly runoff forecast in southern Canada using ANN, K-means, and L-moments techniques
Corani et al. An application of pruning in the design of neural networks for real time flood forecasting
Özbay Karakuş Impact of climatic factors on the prediction of hydroelectric power generation: a deep CNN-SVR approach
CN112732777A (en) Position prediction method, apparatus, device and medium based on time series
Allen Advanced statistical post-processing of ensemble weather forecasts

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20220121