CN116579479A - Wind farm power ultra-short-term prediction method, system, computer and storage medium - Google Patents
Wind farm power ultra-short-term prediction method, system, computer and storage medium Download PDFInfo
- Publication number
- CN116579479A CN116579479A CN202310541109.5A CN202310541109A CN116579479A CN 116579479 A CN116579479 A CN 116579479A CN 202310541109 A CN202310541109 A CN 202310541109A CN 116579479 A CN116579479 A CN 116579479A
- Authority
- CN
- China
- Prior art keywords
- wind
- wind turbine
- edge
- power plant
- inflow
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000003860 storage Methods 0.000 title claims description 7
- 238000013528 artificial neural network Methods 0.000 claims abstract description 49
- 238000009826 distribution Methods 0.000 claims abstract description 23
- 238000013139 quantization Methods 0.000 claims abstract description 6
- 239000010410 layer Substances 0.000 claims description 80
- 230000006870 function Effects 0.000 claims description 69
- 238000011144 upstream manufacturing Methods 0.000 claims description 59
- 230000002776 aggregation Effects 0.000 claims description 36
- 238000004220 aggregation Methods 0.000 claims description 36
- 239000011159 matrix material Substances 0.000 claims description 24
- 238000004364 calculation method Methods 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 9
- 210000004027 cell Anatomy 0.000 claims description 8
- 238000012935 Averaging Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000005315 distribution function Methods 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 5
- 230000004931 aggregating effect Effects 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 239000002356 single layer Substances 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- 238000010276 construction Methods 0.000 claims 1
- 238000010248 power generation Methods 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Primary Health Care (AREA)
- Quality & Reliability (AREA)
- Power Engineering (AREA)
- Wind Motors (AREA)
Abstract
The application discloses an ultra-short-term prediction method for wind power plant power, which comprises the steps of constructing a double LSTM model for predicting wind speed and wind direction of wind power plant inflow, inputting wind speed and wind direction of wind power plant inflow at historical moment into the double LSTM prediction model, and predicting wind speed and wind direction of wind power plant inflow at future moment; constructing a time-varying directed graph generator based on a complex arrangement space quantization method, a 2D-Jensen model and a wake deflection model, inputting the predicted wind power plant inflow wind speed and wind direction at the future moment, the set yaw angles of all wind turbines and the position coordinates of all wind turbines into the time-varying directed graph generator, and generating a time-varying directed graph at the future moment; and (3) constructing a neural network power prediction model of the edge feature weight distribution graph, inputting a time-varying directed graph at a future moment into the neural network power prediction model of the graph, and performing ultra-short-term prediction on the power of each wind turbine in the wind power plant. The method can realize ultra-short-term power prediction of the wind power plant under the influence of time-varying inflow wind conditions and complex space arrangement.
Description
Technical Field
The application belongs to the field of wind power plant power prediction, and particularly relates to a wind power plant power ultra-short-term prediction method, a wind power plant power ultra-short-term prediction system, a wind power plant power ultra-short-term prediction computer and a wind power plant power ultra-short-term prediction storage medium based on an edge characteristic weight distribution graph neural network of an inflow wind condition prediction and time-varying directed graph generator.
Background
Wind energy is one of new energy sources with development prospect due to the advantages of wide distribution, no pollution and the like. With the rapid development of the wind power generation industry, a large-scale clustered wind power plant and a large-size high-power output unit form a main stream wind power generation mode. The existing wind power plant power prediction methods are based on single-row arrangement or regular arrangement, and the multi-wake effect which is difficult to effectively quantify under the influence of irregular complex arrangement and dynamic inflow wind conditions is not considered. The power ultra-short-term prediction under the influence of the wake flow of the wind power plant with complex arrangement is realized, the power loss condition of the wind power plant can be analyzed, and the power generation efficiency of the wind power plant can be effectively improved by combining the power prediction with wake flow optimization control, so that the wind power plant power prediction based on the inflow wind condition prediction and the edge feature weight distribution graph of the time-varying directed graph generator has important significance.
Disclosure of Invention
The application aims to provide a wind power plant power ultra-short-term prediction method, a system, a computer and a storage medium.
The technical solution for realizing the purpose of the application is as follows: a wind power plant power ultra-short-term prediction method is based on an inflow wind condition prediction and edge feature weight distribution graph neural network of a time-varying directed graph generator, and comprises the following specific steps:
step 1, collecting historical moment data of all wind turbines in a wind power plant, and calculating the inflow wind speed and wind direction of the wind power plant;
step 2, constructing a double LSTM model for predicting wind speed and wind direction of wind power plant inflow, inputting wind speed and wind direction of wind power plant inflow at a historical moment into the double LSTM prediction model, and predicting wind speed and wind direction of wind power plant inflow at a future moment;
step 3, constructing a time-varying directed graph generator based on a complex arrangement space quantization method, a 2D-Jensen model and a wake deflection model, inputting the predicted wind power plant inflow wind speed and direction, the set yaw angle of each wind turbine and the position coordinates of each wind turbine into the time-varying directed graph generator to generate a time-varying directed graph at the future time;
and 4, constructing a side characteristic weight distribution graph neural network power prediction model, inputting a future time-varying directed graph into the graph neural network power prediction model, and performing ultra-short-term prediction on the power of each wind turbine in the wind power plant.
The wind power plant power ultra-short-term prediction system is used for realizing the ultra-short-term prediction of wind power plant power based on the wind power plant power ultra-short-term prediction method.
A computer device comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein when the processor executes the computer program, the ultra-short-term prediction of wind power plant power is realized based on the ultra-short-term prediction method of wind power plant power.
A computer readable storage medium having stored thereon a computer program which, when executed by a processor, enables ultra-short term prediction of wind farm power based on the wind farm power ultra-short term prediction method.
Compared with the prior art, the application has the remarkable advantages that: 1) The time-varying directed graph generator can characterize the complex wake influence relation in the complex-arrangement wind power plant through a complex-arrangement space quantization method and a physical wake model. 2) The wake loss factors are used as edge characteristics, so that the influence degree of wake among all wind turbines in the wind power plant can be effectively quantified, a directed graph is constructed through the edge characteristics, and the power prediction performance of the graph neural network can be effectively improved. 3) The wind power station wind power generation system has the advantages that accurate ultra-short-term prediction of wind speed and wind direction of wind power station inflow is realized through an inflow wind condition prediction model, and ultra-short-term power prediction of the wind power station under the influence of complex wake flow can be realized based on the inflow wind condition at the predicted future moment by combining a time-varying directed graph generator and an edge characteristic weight distribution graph neural network.
Drawings
FIG. 1 is a flow chart of a method for ultra-short term prediction of wind farm power according to the present application.
FIG. 2 is a flow chart of a time-varying directed graph generator of the present application.
FIG. 3 is a schematic diagram of the positions of wind turbines in a real wind farm according to an embodiment.
Fig. 4 is a schematic diagram of the network layer structure of the improved graph of the present application.
FIG. 5 is a schematic diagram of the neural network layer update and aggregation process of the present application.
Detailed Description
The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.
The application discloses a wind power plant power ultra-short-term prediction method, which is based on an inflow wind condition prediction and edge feature weight distribution graph neural network of a time-varying directed graph generator, and realizes wind power plant power ultra-short-term prediction, as shown in fig. 1, and comprises the following specific steps:
step 1, collecting historical data of all wind turbines in a wind power plant in current to past N times, reserving key characteristic data (wind speed, wind direction, yaw angle and the like), and calculating inflow wind speed and wind direction of the wind power plant;
SCADA data or laser radar data of each wind turbine in the wind power plant at the past N moments are collected, the wind speed and wind direction characteristics of each wind turbine are reserved, and the wind speed and wind direction of each wind turbine are averaged to obtain the inflow wind speed and wind direction of the wind power plant, wherein the specific formula is as follows:
in θ in For wind power plant inflow wind direction v in For wind power generation, the wind speed of the wind power generation flow is n is the total number of wind turbines in the wind power plant, and theta i V is the wind direction of the ith wind turbine i The wind speed of the ith wind turbine is obtained.
Step 2, constructing a double LSTM model for predicting wind speed and wind direction of wind power plant inflow, respectively inputting wind power plant inflow wind speeds and wind directions at past N times into the double LSTM prediction model, and predicting wind speed and wind direction of wind power plant inflow at future times;
the method comprises the steps of constructing a double LSTM model for predicting the wind speed and the wind direction of the wind power plant inflow, respectively predicting the wind speed and the wind direction of the wind power plant inflow by using two LSTM models with the same structure, and constructing an LSTM neural network by the following steps:
I t =σ(W xi X t +W hi H t-1 +B i )
F t =σ(W xf X t +W hf H t-1 +B f )
O t =σ(W xo X t +W ho H t-1 +B o )
C t =F t ⊙C t-1 +I t ⊙tanh(W xc X t +W hc H t-1 +B c )
H t =O t ⊙tanh(C t )
wherein I is t 、F t 、O t Respectively denoted as a matrix of update gates, forget gates and output gates at time t, C t 、H t The cell state and hidden layer state matrix, X, are respectively expressed as time t t Input matrix W of long-short-period memory unit at t moment xi 、W xf 、W xo 、W xc Weights of update gate, forget gate, output gate and cell state in input layer neural network, W hi 、W hf 、W ho 、W hc Weights of update gate, forget gate and output gate and cell state in hidden layer neurons, respectively, B i 、B f 、B o 、B c Biases of update gate, forget gate, output gate and cell state, respectively, sigma is sigmoid activation function;
the training process of a double LSTM model (called as double LSTM model for short) for predicting the inflow wind speed and the wind direction of a wind power plant requires that the wind speed, the wind direction, the yaw angle and the power historical data of each wind turbine of the wind power plant are collected for a long time, and the data are complemented when the data are missing, so that continuous time sequence data are obtained; averaging all wind turbine wind speeds and wind directions under each moment of the continuous time sequence to obtain a wind power plant historical inflow wind speed and wind direction time sequence; the complete time sequence of the inflow wind speed and the wind direction is standardized, and the standardized formula is as follows:
wherein x' is a normalized time sequence, x is an original time sequence, x max Is the maximum value in the time series; taking the standardized inflow wind speed and direction time sequence as double LSTM model training set data; setting corresponding iteration step number, batch processing number and other super parameters by using MSE, MAE or other indexes as loss functions, and optimizing double LSTM model parameters by an optimizer; when the loss function in the model training process is not reduced, the optimization parameters are completed, and the creation of the double LSTM model is completed.
Step 3, inputting data such as wind speed and wind direction of wind farm inflow at future time predicted by the model, set yaw angle of each wind turbine, position coordinates of each wind turbine, wind turbine parameters (wind wheel radius, rated wind speed, thrust coefficient and the like) and the like into a time-varying directed graph generator based on a complex arrangement space quantization method, a 2D-Jensen model and a wake deflection model, and generating a time-varying directed graph at the future time;
after the wind power plant inflow wind direction is obtained, coordinate transformation is carried out according to the relative spatial position of wind turbines by taking the main inflow wind direction as a reference, a fixed coordinate system taking the positive east as an X axis and the positive north as a Y axis is converted into a new coordinate system taking the direction of the inflow wind direction as the X axis and the direction perpendicular to the inflow wind direction as the Y axis, and the position coordinates of each wind turbine are converted into new coordinates, wherein the calculation formula is as follows:
wherein x is i ′,y i ' is the new coordinate, x after conversion of the ith wind turbine i ,y i Converting the coordinate, theta for the ith wind turbine in For the wind farm main inflow wind direction (theta) in The wind power station is absolute wind direction, namely the included angle between the wind power station main inflow wind direction and the north direction);
after each wind machine in the wind power plant converts a coordinate system according to the inflow wind direction, an X-axis coordinate sequence X' = (X) of all n wind machines in the wind power plant is obtained 1 ′,x′ 2 ,…,x n-1 ,x n ') ordering the elements in the X-axis coordinate sequence X' from small to large to obtain an ordered coordinate sequence X '' min→max According to X 'at this time' min→max Determining the upstream and downstream sequence of the wind turbine. The wind turbine with the minimum X-axis coordinate after conversion is the wind turbine at the most upstream of the wind power plant, the wind turbine with the maximum X-axis coordinate after conversion is the wind turbine at the most downstream of the wind power plant, and the upstream and downstream sequential expressions of the wind turbines are as follows:
(WT 1 ,WT 2 ,…WT n-1 ,WT n ,) min→max
wherein WT represents each wind turbine, min- & gt max represents sorting from small to large according to X-axis coordinates;
as shown in fig. 2, judging whether wake effects exist between each pair of upstream and downstream wind turbines based on the upstream and downstream sequences of the wind turbines through a wake deflection model and a 2D-Jensen model, and calculating the wake effect degree; if wake flow influence exists, a directional edge is established between the upstream wind turbine and the downstream wind turbine, the upstream wind turbine is a transmitting node, the downstream wind turbine is a receiving node, and the characteristic of the edge is that wake flow loss factors of the upstream wind turbine to the downstream wind turbine are set; if the wind power generation system does not exist, no edge is established between the upstream wind power machine and the downstream wind power machine; after traversing all wind turbines in the wind power plant, obtaining a side feature set e of the whole wind power plant sr And an adjacency matrix a; the adjacency matrix A represents whether wake flow influence exists between each upstream wind turbine and each downstream wind turbine, and the edge feature set e sr Characterizing the degree of wake influence between each upstream and downstream wind turbines;
the specific method for judging whether wake influence exists between each upstream wind turbine and each downstream wind turbine according to the wake deflection model and the 2D-Jensen model and calculating the wake influence degree comprises the following steps:
calculating the relative position and azimuth angle of the two wind turbines according to the new coordinates converted by the two wind turbines, defining an upstream wind turbine and a downstream wind turbine according to the upstream and downstream sequences of the wind turbines, and defining an included angle theta between the connecting line of the upstream and downstream wind turbines and the central axis of the tail flow of the upstream wind turbine L Theta by wake deflection model L Solving, wherein the calculation formula is as follows:
θ L =0.3C T ·β 1 +θ FWT -θ in
in θ in For the incoming wind direction angle, θ FWT Is the azimuth angle beta of the upstream wind turbine relative to the downstream wind turbine 1 Yaw angle of upstream wind turbine, C T Is the thrust coefficient of the wind turbine;
defining the vertical distance L between the section of the downstream wind turbine and the upstream wind turbine l The calculation formula is as follows:
L l =Ldcos(θ L )
wherein Ld is the cabin connecting line distance of the upstream and downstream wind turbines;
simultaneously defining the transverse distance between the downstream wind turbine and the upstream wind turbine tail flow center as L r The calculation formula is as follows:
L r =Ldsin(θ L )
ld is the cabin connecting line distance of the upstream and downstream wind turbines;
the wake radius calculation formula of the section position of the downstream wind turbine is as follows:
R l =kL l +r 0
where k is the wake decay coefficient, r 0 The radius of the wind wheel of the wind turbine is;
from the calculated wake radius R l Judging whether the downstream wind turbine is influenced by wake flow of the upstream wind turbine, wherein the specific method comprises the following steps:
1) If L r -r 0 ′>R l The downstream wind turbine is not affected by the wake flow of the upstream wind turbine;
2) If L r -r 0 ′≤R l The downstream wind turbine is subjected toWake effects of upstream wind turbines;
wherein r is 0 ' is the projection length of the radius of the wind wheel of the downstream wind turbine in the direction vertical to the center of the tail flow of the upstream wind turbine, and the calculation formula is as follows:
r 0 ′=r 0 cos(0.3C T ·β 1 +β 2 )
wherein beta is 2 Is the yaw error angle of the downstream wind turbine.
Therefore, when the downstream wind turbine is affected by the wake of the upstream wind turbine, calculating the wake loss factor delta u of the downstream wind turbine by using a 2D-Jensen wake model, and taking the wake loss factor delta u as the edge characteristic between the nodes of the upstream wind turbine and the downstream wind turbine, wherein the calculation formula is as follows:
wherein u is w (L l ,L r ) For equivalent wind speed of downstream wind turbine under the influence of upstream wind turbine wake flow, x is downstream wake flow longitudinal distance, u 0 C for upstream wind turbine inflow wind speed T Is the thrust coefficient of the wind turbine.
Finally summarizing the wake influence and the calculated wake loss factors between the upstream and downstream wind turbines into a wind turbine boundary feature set e sr An adjacent matrix A; the adjacency matrix is specifically defined as an n multiplied by n matrix, n represents the number of nodes existing in the directed graph, if the node i wind machine has wake influence on the node j wind machine, the element in the ith row and the jth column in the adjacency matrix is 1, and if the wake influence does not exist, the element is 0, and the formula is as follows:
setting the inflow wind direction of the wind power plant at the future moment as a global characteristic, and setting the inflow wind speed of the wind power plant at the future moment as each wind turbine node characteristic; finally, the global feature g in Edge feature e sr Node characteristics n i Summarizing with adjacency matrix A to future timeDirected graph g= (e) sr ,n i ,g in ,A)。
Step 4, constructing a side characteristic weight distribution graph neural network power prediction model, inputting a future time-varying directed graph into the graph neural network power prediction model, and performing ultra-short-term prediction on the power of each wind turbine in the wind power plant;
as shown in fig. 4, the number of layers of the edge feature weight distribution graph neural network model is generally selected to be 2-4, wherein the first layer is a graph neural network layer (weight layer) adopting directional graph edge feature weight distribution, and the other layers are general graph neural network layers (general layers) not adopting edge weight distribution; as shown in FIG. 5, each layer of the neural network layer has three update functions and three aggregate functions, which are edge update functions Φ e (. Cndot.) node update function Φ n (. Cndot.) Global update function Φ g (. Cndot.) Adjacent edge aggregation function phi e→n (. Cndot.) edge aggregation function phi e→g (. Cndot. Convergence function phi n→g (. Cndot.); the weight layer is to embed edge feature weight distribution function rho on the basis of the general layer e (·);
In the single-layer graph neural network layer, the input graph is G in =(e sr ,n i ,g in A) with all edge features e sr Node characteristics n i And global feature g in The update sequence is as follows:
first, edge feature updates are performed, each directed edge update requiring a neighboring sending node feature n s And receiving node characteristics n r And the edge itself feature e sr Inputting all features into the edge update function Φ e In (-), the edge update function is defined by the neural network MLP e (. Cndot.) generation, updated edge feature e sr ' is:
e sr ′=Φ e (e sr ,n s ,n r )=MLP e (e sr ,n s ,n r )
edge weight w sr The definition is as follows:
wherein x is the longitudinal distance of downstream wake flow, the weight layer adopts an edge characteristic weight distribution function to update edge characteristics e after edge sr ' multiplying the final edge feature e by the weight sr ”:
e sr ”=ρ e (e sr )×e sr ′=w sr ×e sr ′
The general layer does not adopt an edge characteristic weight distribution function, and the final edge characteristic is the updated edge characteristic:
e sr ″=e sr ′
after the updating of all edge characteristics in the graph neural network layer is completed, updating the node characteristics, wherein each node characteristic n i The update procedure requires all adjacent edge features e si Self node characteristics n i And global feature g in First, the adjacent edge aggregation function phi is used e→n (. Cndot.) aggregating s adjacent edge features that have connections to the node; the weight layer adopts a summation function as an adjacent edge aggregation function, the general layer adopts an averaging function as an adjacent edge aggregation function, and the characteristics of adjacent edges after aggregation are defined as follows:
weight layer:
general layer:
the polymerized adjacent edge feature E si Self node characteristics n i And global feature g in Input to node update function Φ n In (-), the node update function is defined by the neural network MLP n (. Cndot.) generation, updated node characteristics n i ' is:
n i ′=Φ n (E si ,n i ,g in )=MLP n (E si ,n i ,g in )
after all node characteristics are updated, global characteristic updating is carried out, and all edge characteristics, all node characteristics and global characteristics of a wind farm are required for global characteristic updating; first using the full edge aggregation function phi e→g (.) aggregating edge features of all m directed edges in the directed graph; the weight layer adopts a summation function as an all-edge aggregation function, the general layer adopts an averaging function as an all-edge aggregation function, and all-edge characteristics after aggregation are defined as follows:
weight layer:
general layer:
after the aggregation of all edge features is completed, all node aggregation functions phi are used n→g (-) all p node features in the directed graph; the weight layer adopts a summation function as an all-node aggregation function, the general layer adopts an averaging function as an all-node aggregation function, and all node characteristics after aggregation are defined as follows:
weight layer:
general layer:
after the aggregation of all the edge features and the node features is completed, the aggregated edge features E, the aggregated node features N and the global features g are aggregated in Inputting global update function Φ g In (-), the global feature update function is defined by the neural network MLP g (. Cndot.) the updated global feature g' is generated as follows:
g′=Φ g (E,N,g in )=MLP g (E,N,g in )
after the global features are updated, the calculation of a single-layer graph neural network layer is completed, the adjacent matrix A is not changed in the whole updating and aggregation process, and the graph neural network layer outputs edge features e sr "node characteristics n i 'and global feature g', edge feature e to be output sr "node characteristics n i ' and global feature gv and adjacency matrix A are taken as input features and input to the graph neural network layer of the next layer;
edge characteristics, node characteristics, edge characteristics and adjacent matrixes in the directed graph are calculated through all graph neural network layers to obtain an output directed graph G out =(e sr out ,n i out ,g out A), inputting node characteristics in the output directed graph into the full-connection layer MLP p (. Cndot.) finally calculating the characteristics of each node in the directed graph, wherein the node characteristics are the power characteristics power of each wind turbine i :
power i =MLP p (n i out )
Collecting historical data of wind speeds, wind directions, yaw angles and output power of all wind turbines in a longer-time wind power plant, cleaning the historical data, eliminating non-positive data of power or wind speeds, and guaranteeing the quality of a training data set; generating a time-varying directed graph from the cleaned historical data set through a time-varying directed graph generator, dynamically adjusting the inflow wind condition and the yaw angle of the global, node, edge characteristics and the adjacent matrix of the time-varying directed graph along with the continuous change, and taking the generated time-varying directed graph as a training set of the graph neural network model;
inputting a time-varying directed graph training data set, training and optimizing the graph neural network model by using an optimization algorithm according to a loss function, wherein the loss function adopts an MSE index, and specifically comprises the following steps:
in the power i k Inputting a power predicted value of an ith fan node in the directed graph at the moment k, and p i k And inputting an actual power value of the ith wind turbine of the directed graph at the moment k, wherein q is the total number of input data, and n is the total number of wind turbines in the wind field.
The application further provides a wind power plant power ultra-short-term prediction system based on the side characteristic weight distribution graph of the inflow wind condition prediction and time-varying directed graph generator, and the wind power plant power ultra-short-term prediction method based on the side characteristic weight distribution graph of the inflow wind condition prediction and time-varying directed graph generator is utilized to realize wind power plant power ultra-short-term prediction.
The computer equipment comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein when the processor executes the computer program, the wind power field power ultra-short-term prediction method based on the inflow wind condition prediction and the edge feature weight distribution graph neural network of the time-varying directed graph generator is utilized to realize the wind power field power ultra-short-term prediction.
A computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements wind farm power ultra-short-term prediction using the method for distributing graph neural network wind farm power ultra-short-term prediction based on the edge feature weights of the inflow wind condition prediction and time-varying directed graph generator.
Examples
In order to verify the effectiveness of the inventive solution, the following experiments were performed using SCADA data of a real wind farm.
Characteristic data of wind directions, wind speeds, yaw angles and output power of all wind turbines in a certain real wind farm in China are collected, and the wind farm is provided with 33 wind turbines, and the position relationship is shown in figure 3. All wind power machines in the wind power plant are the same in model number, the radius of an impeller of the wind power machine is 38.5m, the rated power of the wind power machine is 1500kW, the thrust coefficient of the wind power machine is 0.7, and the wake flow attenuation coefficient is 0.075.
And calculating the inflow wind speed and inflow wind direction sequence of the wind farm according to the wind speed and wind direction data of each wind turbine, and carrying out standardized operation on the inflow wind speed and wind direction sequence. Two LSTM models are constructed and trained, the two models respectively predict the inflow wind speed and the inflow wind direction of the wind power plant, the past 50 time points are input to the current inflow wind speed and wind direction univariate data sequence, and the predicted wind power plant inflow wind speed and wind direction univariate at the future time are output. The training process uses MAE as the loss function, the optimizer selects Adam algorithm, iterates 2000 steps, and batch training number selection 72. Two LSTM predictive models are trained using historical inflow wind condition time series data.
Inputting the wind speed and the wind direction of wind power plant inflow at the future moment and the set yaw angle of each wind turbine into a time-varying directed graph generator to generate a time-varying directed graph taking the wind power plant inflow at the future moment as a global characteristic, taking a wake loss factor among wind turbines as an edge characteristic and taking the wind power plant inflow wind speed as a node characteristic.
The graph neural network power prediction model adopts three layers of graph neural network layers, wherein the first layer is a network layer adopting edge characteristic weight distribution, and the second layer is a general network layer not adopting edge characteristic weight distribution. The activation function is a ReLU function, the optimizer selects Adam algorithm, and the learning rate is set to be 3e -6 Iteration 20000 steps.
And after model training is finished, carrying out power prediction accuracy test by using the test set data graph neural network model, wherein the final wind power plant total power prediction accuracy rate is 87%, the average total power prediction absolute error of 33 wind turbines in the wind power plant is 1900kw, and the correlation coefficient of the true value and the predicted value is 0.949. The result shows that the method is accurate and effective, and can realize accurate ultra-short-term prediction of the power output of the accurate wind power plant.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples merely represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the present application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of the application should be assessed as that of the appended claims.
Claims (9)
1. The wind power plant power ultra-short-term prediction method is characterized by comprising the following specific steps of:
step 1, collecting historical moment data of all wind turbines in a wind power plant, and calculating the inflow wind speed and wind direction of the wind power plant;
step 2, constructing a double LSTM model for predicting wind speed and wind direction of wind power plant inflow, inputting wind speed and wind direction of wind power plant inflow at a historical moment into the double LSTM prediction model, and predicting wind speed and wind direction of wind power plant inflow at a future moment;
step 3, constructing a time-varying directed graph generator based on a complex arrangement space quantization method, a 2D-Jensen model and a wake deflection model, inputting the predicted wind power plant inflow wind speed and direction, the set yaw angle of each wind turbine and the position coordinates of each wind turbine into the time-varying directed graph generator to generate a time-varying directed graph at the future time;
and 4, constructing a side characteristic weight distribution graph neural network power prediction model, inputting a future time-varying directed graph into the graph neural network power prediction model, and performing ultra-short-term prediction on the power of each wind turbine in the wind power plant.
2. The ultra-short-term prediction method for wind farm power according to claim 1, wherein step 1, collecting historical moment data of each wind turbine in the wind farm, and calculating wind speed and wind direction of wind farm inflow, the specific method is as follows:
SCADA data or laser radar data of each wind turbine in the wind power plant at the past N moments are collected, the wind speed and wind direction characteristics of each wind turbine are reserved, and the wind speed and wind direction of each wind turbine are averaged to obtain the inflow wind speed and wind direction of the wind power plant, wherein the specific formula is as follows:
in θ in For wind power plant inflow wind direction v in For wind power plant inflow wind speed, n is total number of wind turbines in wind power plant, theta i V is the wind direction of the ith wind turbine i The wind speed of the ith wind turbine is obtained.
3. The method for ultra-short-term prediction of wind farm power according to claim 1, wherein step 2, a double LSTM model for predicting wind farm inflow wind speed and direction is constructed, wind farm inflow wind speed and direction at a historical moment are input into the double LSTM prediction model, wind farm inflow wind speed and direction at a future moment are predicted, and wherein:
the double LSTM model for predicting the wind speed and the wind direction of the wind power plant adopts two LSTM models with the same structure, which respectively predict the wind speed and the wind direction of the wind power plant, and the network structure of the LSTM model is as follows:
I t =σ(W xi X t +W hi H t-1 +B i )
F t =σ(W xf X t +W hf H t-1 +B f )
O t =σ(W xo X t +W ho H t-1 +B o )
C t =F t ⊙C t-1 +I t z⊙tanh(W xc X t +W hc H t-1 +B c )
H t =O t ⊙tanh(C t )
wherein I is t 、F t 、O t Respectively denoted as a matrix of update gates, forget gates and output gates at time t, C t 、H t The cell state and hidden layer state matrix, X, are respectively expressed as time t t Input matrix W of long-short-period memory unit at t moment xi 、W xf 、W xo 、W xc Weights of update gate, forget gate, output gate and cell state in input layer neural network, W hi 、W hf 、W ho 、W hc Weights of update gate, forget gate and output gate and cell state in hidden layer neurons, respectively, B i 、B f 、B o 、B c Biases of update gate, forget gate, output gate and cell state, respectively, sigma is sigmoid activation function;
after the time sequence of the inflow wind speed and the wind direction is standardized, the time sequence of the inflow wind speed and the wind direction is used as training set data of a double LSTM prediction model of the inflow wind condition, MSE, MAE or other indexes are used as loss functions, corresponding iteration step numbers and batch processing numbers are set, parameters of the double LSTM prediction model are optimized through an optimizer, when the loss functions in the model training process are not reduced any more, the optimization parameters are completed, and the construction of the double LSTM model for predicting the inflow wind speed and the wind direction of the wind power plant is completed.
4. The ultra-short-term wind farm power prediction method according to claim 1, wherein step 3, a time-varying directed graph generator based on a complex arrangement space quantization method, a 2D-Jensen model and a wake deflection model is constructed, the wind turbine set yaw angle, the wind turbine position coordinates and the wind farm inflow wind speed and direction at the future time predicted by the double LSTM prediction model are input into the time-varying directed graph generator, and a time-varying directed graph at the future time is generated, wherein the time-varying directed graph generator comprises the following specific steps:
taking the main inflow wind direction as a reference, carrying out coordinate transformation according to the relative spatial position of the wind turbines, converting a fixed coordinate system taking the positive east as an X axis and the positive north as a Y axis into a dynamic time-varying coordinate system taking the direction of the inflow wind direction as an X axis and the direction perpendicular to the inflow wind direction as a Y axis, and converting the position coordinates of each wind turbine into new coordinates, wherein the calculation formula is as follows:
wherein x is i ′,y i ' is the new coordinate, x after conversion of the ith wind turbine i ,y i Converting the coordinate, theta for the ith wind turbine in Taking the main inflow wind direction of the wind power plant as an absolute wind direction, and taking an included angle between the main inflow wind direction of the wind power plant and the north direction;
under the new coordinate system, acquiring X ' = (X ' of X-axis coordinate sequences of all n wind turbines of the wind power plant ' 1 ,x′ 2 ,…,x′ n-1 ,x′ n ) Ordering the elements in the X-axis coordinate sequence X ' from small to large to obtain an ordered coordinate sequence X ' ' min→max According to X' min→max Determining the upstream and downstream sequences of wind turbines, wherein the wind turbine with the smallest X-axis coordinate is the most upstream wind turbine of a wind power plant, the wind turbine with the largest X-axis coordinate is the most downstream wind turbine of the wind power plant, and the upstream and downstream sequences of the wind turbines are expressed as follows:
(WT 1 ,WT 2 ,…WT n-1 ,WT n ,) min→max
wherein, WT i Representing an ith fan, wherein min-max represents sorting from small to large according to X-axis coordinates;
judging whether wake effects exist between each pair of upstream and downstream wind turbines based on the upstream and downstream wind turbines in sequence through a wake deflection model and a 2D-Jensen model, and calculating the wake effect degree, if the wake effects exist, establishing a directional edge between the upstream and downstream wind turbines, wherein the upstream wind turbine is a transmitting node, the downstream wind turbine is a receiving node, and setting the edge as wake loss factors of the upstream wind turbine to the downstream wind turbine; if not, no edge is established between the upstream wind turbine and the downstream wind turbine, and the specific method comprises the following steps:
calculating the relative position and azimuth angle of the two wind turbines according to the new coordinates converted by the two wind turbines, defining an upstream wind turbine and a downstream wind turbine according to the upstream and downstream sequences of the wind turbines, and defining an included angle theta between the connecting line of the upstream and downstream wind turbines and the central axis of the tail flow of the upstream wind turbine L Theta by wake deflection model L Solving;
θ L =0.3C T ·β 1 +θ FWT -θ in
in θ in To get intoAngle of wind flow direction, theta FWT Is the azimuth angle beta of the upstream wind turbine relative to the downstream wind turbine 1 Yaw angle of upstream wind turbine, C T Is the thrust coefficient of the wind turbine;
defining the vertical distance L between the section of the downstream wind turbine and the upstream wind turbine l ;
L l =Ldcos(θ L )
Wherein Ld is the cabin connecting line distance of the upstream and downstream wind turbines;
simultaneously defining the transverse distance between the downstream wind turbine and the upstream wind turbine tail flow center as L r ;
L r =Ldsin(θ L )
The wake radius calculation formula of the section position of the downstream wind turbine is as follows:
R l =kL l +r 0
where k is the wake decay coefficient, r 0 The radius of the wind wheel of the wind turbine is;
from the calculated wake radius R l Judging whether the downstream wind turbine is influenced by wake flow of the upstream wind turbine, if L r -r 0 ′>R l The downstream wind turbine is not affected by the wake flow of the upstream wind turbine; if L r -r 0 ′≤R l The downstream wind turbine is affected by the wake of the upstream wind turbine, where r 0 ' is the projection length of the radius of the wind wheel of the downstream wind turbine in the direction vertical to the center of the tail flow of the upstream wind turbine, and the calculation formula is as follows:
r 0 ′=r 0 cos(0.3C T ·β 1 +β 2 )
wherein beta is 2 Yaw error angle for downstream wind turbines;
when the downstream wind turbine is affected by the wake of the upstream wind turbine, calculating a wake loss factor delta u of the downstream wind turbine by using a 2D-Jensen wake model, and taking the wake loss factor delta u as an edge characteristic between nodes of the upstream wind turbine and the downstream wind turbine, wherein a calculation formula is as follows:
wherein u is w (L l ,L r ) For equivalent wind speed of downstream wind turbine under the influence of upstream wind turbine wake flow, x is downstream wake flow longitudinal distance, u 0 C for upstream wind turbine inflow wind speed T Is the thrust coefficient of the wind turbine;
after traversing all wind turbines in the wind power plant, obtaining a side feature set e of the whole wind power plant sr And an adjacent matrix A, wherein the adjacent matrix A represents whether wake flow influence exists between each upstream wind turbine and each downstream wind turbine, and the edge feature set e sr Representing the degree of wake influence between each upstream wind turbine and each downstream wind turbine, wherein an adjacent matrix is specifically defined as an n multiplied by n matrix, n represents the number of nodes existing in the directed graph, if the wake influence is caused to the node j wind turbine by the node i wind turbine, the element in the ith row and the jth column in the adjacent matrix is 1, and if the wake influence is not present, the element is 0, and the formula is as follows:
finally, setting the wind power plant inflow wind direction at the future moment as a global feature g in Setting the inflow wind speed of the wind farm at the future moment as the node characteristic n of each wind turbine i Global feature g in Node characteristics n i Edge feature e sr And adjacency matrix a are summarized as a directed graph g= (e) at future time instant sr ,n i ,g in ,A)。
5. The method for predicting the power of each wind turbine in the wind farm in an ultra-short term according to claim 4, wherein in step 4, an edge feature weight distribution graph neural network power prediction model is constructed, a future time-varying directed graph is input into the graph neural network power prediction model, and the ultra-short term prediction of the power of each wind turbine in the wind farm is performed, wherein the edge feature weight distribution graph neural network power prediction model specifically comprises:
selecting 2-4 layers by using the neural network power prediction model of the edge feature weight distribution graph, wherein the layer is the first layerOne layer adopts a graph neural network layer with directional graph edge characteristic weight distribution, which is a weight layer, and the other layers do not adopt the graph neural network layer with edge weight distribution, which is a general layer; each graph neural network layer has three updating functions and three aggregation functions, which are respectively edge updating functions phi e (. Cndot.) node update function Φ n (. Cndot.) Global update function Φ g (. Cndot.) Adjacent edge aggregation function phi e→n (. Cndot.) edge aggregation function phi e→g (. Cndot. Convergence function phi n→g (. Cndot.) the weight layer is an embedded edge feature weight distribution function ρ on a generic layer basis e (·);
In the single-layer graph neural network layer, the input graph is G in =(e sr ,n i ,g in A) with all edge features e sr Node characteristics n i And global feature g in The update sequence is as follows:
first, edge feature updates are performed, each directed edge update requiring a neighboring sending node feature n s And receiving node characteristics n r And the edge itself feature e sr Inputting all features into the edge update function Φ e In (-), the edge update function is defined by the neural network MLP e (. Cndot.) generation, updated edge feature e sr ' is:
e sr ′=Φ e (e sr ,n s ,n r )=MLP e (e sr ,n s ,n r )
edge weight w sr The definition is as follows:
wherein x is the longitudinal distance of downstream wake flow, the weight layer adopts an edge characteristic weight distribution function to update edge characteristics e after edge sr ' multiplying the final edge feature e by the weight sr ”:
e sr ”=ρ e (e sr )×e sr ′=w sr ×e sr ′
The general layer does not adopt an edge characteristic weight distribution function, and the final edge characteristic is the updated edge characteristic:
e sr ”=e sr ′
after the updating of all edge characteristics in the graph neural network layer is completed, updating the node characteristics, wherein each node characteristic n i The update procedure requires all adjacent edge features e si Self node characteristics n i And global feature g in First, the adjacent edge aggregation function phi is used e→n (. Cndot.) aggregating s adjacent edge features that have connections to the node; the weight layer adopts a summation function as an adjacent edge aggregation function, the general layer adopts an averaging function as an adjacent edge aggregation function, and the characteristics of adjacent edges after aggregation are defined as follows:
weight layer:
general layer:
the polymerized adjacent edge feature E si Self node characteristics n i And global feature g in Input to node update function Φ n In (-), the node update function is defined by the neural network MLP n (. Cndot.) generation, updated node characteristics n i ' is:
n i ′=Φ n (E si ,n i ,g in )=MLP n (E si ,n i ,g in )
after all node characteristics are updated, global characteristic updating is carried out, and all edge characteristics, all node characteristics and global characteristics of a wind farm are required for global characteristic updating; first using the full edge aggregation function phi e→g (.) aggregating edge features of all m directed edges in the directed graph; the weight layer adopts a summation function as an all-edge aggregation function, and the general layer adopts an averaging functionThe number is used as a total edge aggregation function, and the total edge characteristics after aggregation are defined as follows:
weight layer:
general layer:
after the aggregation of all edge features is completed, all node aggregation functions phi are used n→g (-) all p node features in the directed graph; the weight layer adopts a summation function as an all-node aggregation function, the general layer adopts an averaging function as an all-node aggregation function, and all node characteristics after aggregation are defined as follows:
weight layer:
general layer:
after the aggregation of all the edge features and the node features is completed, the aggregated edge features E, the aggregated node features N and the global features g are aggregated in Inputting global update function Φ g In (-), the global feature update function is defined by the neural network MLP g (. Cndot.) the updated global feature g' is generated as follows:
g′=Φ g (E,N,g in )=MLP g (E,N,g in )
after the global features are updated, the calculation of a single-layer graph neural network layer is completed, the adjacent matrix A is not changed in the whole updating and aggregation process, and the graph neural network layer outputs edge features e sr ″、Node characteristics n i 'and global feature g', edge feature e to be output sr "node characteristics n i 'and global features g' and adjacency matrix A are taken as input features and input to the graph neural network layer of the next layer;
edge characteristics, node characteristics, edge characteristics and adjacent matrixes in the directed graph are calculated through all graph neural network layers to obtain an output directed graph G out =(e sr out ,n i out ,g out A), inputting node characteristics in the output directed graph into the full-connection layer MLP p (. Cndot.) finally calculating the characteristics of each node in the directed graph, wherein the node characteristics are the power characteristics power of each wind turbine i :
power i =MLP p (n i out )。
6. The method for predicting the ultra-short period of the power of the wind farm according to claim 1, wherein in step 4, the graph neural network model is trained and optimized by adopting an MSE index, and the loss function is specifically:
in the power i k Inputting a power predicted value of an ith fan node in the directed graph at the moment k, and p i k And inputting an actual power value of the ith wind turbine of the directed graph for the moment k, wherein q is the total number of the moment of inputting data, and n is the total number of the wind turbines in the wind field.
7. A wind farm power ultra-short term prediction system, characterized in that the wind farm power ultra-short term prediction is implemented based on the wind farm power ultra-short term prediction method according to any one of claims 1-6.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing an ultra-short term prediction of wind farm power based on the wind farm power ultra-short term prediction method of any of claims 1-6 when the computer program is executed.
9. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, enables ultra-short term prediction of wind farm power based on the wind farm power ultra-short term prediction method of any of claims 1-6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310541109.5A CN116579479B (en) | 2023-05-15 | 2023-05-15 | Wind farm power ultra-short-term prediction method, system, computer and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310541109.5A CN116579479B (en) | 2023-05-15 | 2023-05-15 | Wind farm power ultra-short-term prediction method, system, computer and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116579479A true CN116579479A (en) | 2023-08-11 |
CN116579479B CN116579479B (en) | 2024-04-09 |
Family
ID=87537196
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310541109.5A Active CN116579479B (en) | 2023-05-15 | 2023-05-15 | Wind farm power ultra-short-term prediction method, system, computer and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116579479B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114419878A (en) * | 2021-12-22 | 2022-04-29 | 银江技术股份有限公司 | Method, electronic device and storage medium for urban road network global traffic state prediction |
CN114511158A (en) * | 2022-04-19 | 2022-05-17 | 南京理工大学 | Wind turbine power prediction method based on wake deflection effect and 2DJensen model |
CN114757427A (en) * | 2022-04-22 | 2022-07-15 | 大连理工大学 | Autoregressive corrected LSTM intelligent wind power plant ultra-short term power prediction method |
CN114818483A (en) * | 2022-04-14 | 2022-07-29 | 东南大学溧阳研究院 | Electromechanical disturbance positioning and propagation prediction method based on graph neural network |
US20230018575A1 (en) * | 2021-07-15 | 2023-01-19 | Siemens Aktiengesellschaft | Distribution grid topology identification encoding known toplogial information |
-
2023
- 2023-05-15 CN CN202310541109.5A patent/CN116579479B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20230018575A1 (en) * | 2021-07-15 | 2023-01-19 | Siemens Aktiengesellschaft | Distribution grid topology identification encoding known toplogial information |
CN114419878A (en) * | 2021-12-22 | 2022-04-29 | 银江技术股份有限公司 | Method, electronic device and storage medium for urban road network global traffic state prediction |
CN114818483A (en) * | 2022-04-14 | 2022-07-29 | 东南大学溧阳研究院 | Electromechanical disturbance positioning and propagation prediction method based on graph neural network |
CN114511158A (en) * | 2022-04-19 | 2022-05-17 | 南京理工大学 | Wind turbine power prediction method based on wake deflection effect and 2DJensen model |
CN114757427A (en) * | 2022-04-22 | 2022-07-15 | 大连理工大学 | Autoregressive corrected LSTM intelligent wind power plant ultra-short term power prediction method |
Non-Patent Citations (4)
Title |
---|
MENG ZHANG等: ""Optimal Graph Structure Based Short-Term Solar PV Power Forecasting Method Considering Surrounding Spatio-Temporal Correlations"", 《IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS》, vol. 59, no. 1, XP011932607, DOI: 10.1109/TIA.2022.3213008 * |
XIAOCHONG DONG等: ""Spatio-temporal Convolutional Network Based Power Forecasting of Multiple Wind Farms"", 《JOURNAL OF MODERN POWER SYSTEMS AND CLEAN ENERGY》, vol. 10, no. 2 * |
朱呈勇 等: ""考虑三维旋转和动态失速的风力机气动力建模"", 《中国电机工程学报》, vol. 42, no. 12 * |
苏向敬 等: ""基于 MSTAGNN模型的可解释海上风电场多风机出力预测"", 《电力系统自动化》, vol. 47, no. 9 * |
Also Published As
Publication number | Publication date |
---|---|
CN116579479B (en) | 2024-04-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Lu et al. | Short-term wind power forecasting based on meteorological feature extraction and optimization strategy | |
CN110110434B (en) | Initialization method for probability load flow deep neural network calculation | |
CN112508279B (en) | Regional distributed photovoltaic prediction method and system based on spatial correlation | |
CN110504676A (en) | A kind of State Estimation for Distribution Network based on APSO-BP | |
CN110264012A (en) | Renewable energy power combination prediction technique and system based on empirical mode decomposition | |
CN110443724A (en) | A kind of electric system fast state estimation method based on deep learning | |
CN113988481B (en) | Wind power prediction method based on dynamic matrix prediction control | |
Wang et al. | Ultra-short-term wind power forecasting based on deep belief network | |
CN111798037B (en) | Data-driven optimal power flow calculation method based on stacked extreme learning machine framework | |
Chen et al. | Research on wind power prediction method based on convolutional neural network and genetic algorithm | |
CN104021315A (en) | Method for calculating station service power consumption rate of power station on basis of BP neutral network | |
CN112036651A (en) | Electricity price prediction method based on quantum immune optimization BP neural network algorithm | |
CN116227689A (en) | Photovoltaic panel pollution degree state evaluation and cleaning decision system | |
Huang et al. | Elman neural network considering dynamic time delay estimation for short-term forecasting of offshore wind power | |
CN113128666A (en) | Mo-S-LSTMs model-based time series multi-step prediction method | |
CN116579479B (en) | Wind farm power ultra-short-term prediction method, system, computer and storage medium | |
CN115310209B (en) | Pneumatic shape migration optimization method and related device based on VAE | |
CN114511158B (en) | Wind turbine power prediction method based on wake deflection effect and 2DJensen model | |
CN114221334B (en) | Rapid state estimation method based on graph neural network | |
CN115526376A (en) | Multi-feature fusion generation countermeasure network ultra-short-term wind power prediction method | |
KR102406851B1 (en) | Coordinated optimization method for maximizing the power of wind farm using scalable wake digraph and apparatus performing the same | |
CN115065166A (en) | Low-voltage distribution network state sensing and abnormal alarming method | |
Huang et al. | A long short-term memory based wind power prediction method | |
CN116822370B (en) | Ultra-short-term output prediction method for wind power cluster under data driving | |
Liu et al. | Available power estimation of wind farms based on deep spatio-temporal neural networks |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |