CN116822370B - Ultra-short-term output prediction method for wind power cluster under data driving - Google Patents
Ultra-short-term output prediction method for wind power cluster under data driving Download PDFInfo
- Publication number
- CN116822370B CN116822370B CN202310847303.6A CN202310847303A CN116822370B CN 116822370 B CN116822370 B CN 116822370B CN 202310847303 A CN202310847303 A CN 202310847303A CN 116822370 B CN116822370 B CN 116822370B
- Authority
- CN
- China
- Prior art keywords
- wind
- wind speed
- wake
- data
- wind power
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 230000000694 effects Effects 0.000 claims abstract description 27
- 230000015654 memory Effects 0.000 claims abstract description 17
- 230000009467 reduction Effects 0.000 claims abstract description 6
- 239000013598 vector Substances 0.000 claims description 34
- 230000014509 gene expression Effects 0.000 claims description 19
- 230000004913 activation Effects 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 9
- 230000007547 defect Effects 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 7
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 238000009792 diffusion process Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 6
- 238000003646 Spearman's rank correlation coefficient Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000004088 simulation Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 abstract description 11
- 238000013527 convolutional neural network Methods 0.000 abstract description 6
- 238000004458 analytical method Methods 0.000 description 4
- 238000011144 upstream manufacturing Methods 0.000 description 4
- 230000002238 attenuated effect Effects 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000002860 competitive effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000000053 physical method Methods 0.000 description 2
- 241000341910 Vesta Species 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000013016 damping Methods 0.000 description 1
- 230000006735 deficit Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000001771 impaired effect Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 206010027175 memory impairment Diseases 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/08—Fluids
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Geometry (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Hardware Design (AREA)
- Biomedical Technology (AREA)
- Evolutionary Biology (AREA)
- Pure & Applied Mathematics (AREA)
- Medical Informatics (AREA)
- Fluid Mechanics (AREA)
- Algebra (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a wind power cluster ultra-short-term output prediction method under data driving, which comprises the following steps: taking the space-time distribution characteristics of wind power resources of the wind power clusters into consideration, taking wind speed prediction data as training samples, acquiring a feature sequence after dimension reduction by adopting a convolutional neural network, and realizing time sequence wind speed information prediction with rapidness and accuracy by utilizing a long-short-term memory (LSTM) network model; and (3) constructing a 3D-Frandsen wake model based on the wake effect of the wind power plant and boundary compensation thereof, and realizing decoupling calculation of the power of each wind turbine generator set by using the equivalent wind speed of the wind power plant. According to the method provided by the invention, the high-precision prediction of the ultra-short-term output of the wind power cluster is realized by introducing the LSTM network model driven by wind speed prediction data. In addition, the method also verifies the necessity of calculating wake flow effects among wind turbines and performing boundary compensation when wind power cluster output prediction is performed, and lays a boundary parameter foundation for real-time regulation and control in the wind power clusters in the day.
Description
Technical Field
The invention relates to the technical field of new energy output prediction, in particular to an ultra-short-term output prediction method of a wind power cluster under data driving.
Background
Wind farm power prediction can be classified into ultra-short-term prediction, short-term prediction and medium-long-term prediction according to a prediction time scale, wherein the ultra-short-term prediction and the short-term prediction are mainly used in practical application. Short-term prediction refers to prediction of minutes to days in advance, and has the main function of meeting the running and scheduling requirements of a wind power generation system and a grid-connected power system thereof. The significance of wind power plant power prediction is as follows: 1) The accurate wind power plant power prediction can help to adjust the scheduling plan of the power system, so that adverse effects of wind power uncertainty on a power grid are weakened, and meanwhile, the rotation reserve capacity and the running cost of the power system are reduced; 2) Compared with other controllable power generation modes, the inherent randomness of the wind power is relatively poor in the competitive power of the electric power market, the low-reliability power supply mode of the wind power can directly bring economic penalty, and the accurate and reliable wind power prediction can effectively raise the competitive power of the wind power; 3) The wind power plant can reasonably select to maintain and overhaul equipment when the wind speed is lower than the cut-in wind speed or the output power of the wind power plant is small according to the wind power plant power prediction result, and the capacity coefficient of the wind power plant is improved.
The large-scale development of wind power resources brings higher uncertainty to a power system, and wind power prediction is to predict a power value of a wind power plant in a certain period of time in the future according to wind power plant historical data, fan parameter information and geographic position information. The common wind power prediction method can be divided into: physical methods, statistical methods, artificial intelligence methods, combinatorial methods, and the like. The physical method based on digital weather forecast fully considers factors such as geography, environment and the like of the wind power plant, and can realize prediction without accumulating a large amount of historical data, so that the method is widely applied to power prediction of a newly built wind power plant; the time sequence prediction method has simple model and less calculation amount, but because the input data is single, the influence of other information is difficult to consider, the robustness is poor, and the prediction precision can be rapidly reduced along with the increase of the time scale; the artificial intelligence method has very good performance on the processing of the mapping relation, can deeply explore the interrelationship and the mapping mechanism among all variables of the wind power plant, can effectively capture and display the characteristics of data fluctuation, can process larger-scale data information, can preprocess non-numerical information in a data coding mode and the like, realizes the effective receiving of the non-numerical data, and provides more powerful data support for prediction; the combined prediction method can fully utilize the advantages of multiple methods and avoid the defects of a single method, but the combined prediction method also needs to consider how to select an appropriate method and a combination mode of the multiple methods according to application scenes.
Wind power prediction models based on historical power data are difficult to accurately reflect the change rule of actual power of a wind farm, wake effects caused by wind energy captured by upstream wind turbines can have significant influence on output power of downstream turbines, and conventional wind power prediction models usually ignore wake effects among wind turbines.
Disclosure of Invention
In order to solve the problems, the invention provides a method for predicting the ultra-short-term output of a wind power cluster under data driving, which adopts an LSTM network model to predict the wind speed of the incoming flow of a wind power plant, and effectively improves the accuracy of short-term wind speed prediction. In addition, according to the topological structure of the wind power cluster in the incoming wind direction, a 3D-Frandsen wake model considering the wake effect is constructed, further the output power of each fan is decoupled and analyzed, the predicted value of the wind power plant output is obtained after integration, and the obtained predicted result is more in line with the actual running condition of each wind motor in the wind power cluster.
In order to achieve the above object, the present invention is realized by the following technical scheme:
the invention discloses a wind power cluster ultra-short-term output prediction method under data driving, which comprises the following operations:
step 1, acquiring wind speed prediction data, including historical data and digital weather forecast data, and preprocessing the wind speed prediction data;
step 2, adopting CNN to reduce the dimension of the preprocessed wind speed prediction data, obtaining a feature sequence after the dimension reduction, constructing an LSTM network model of wind speed prediction, and inputting the feature sequence into the LSTM network model to predict the incoming wind speed of the wind power plant;
step 3, considering wake effects of the wind power plant and boundary compensation thereof, constructing a 3D-Frandsen wake model based on Gaussian distribution optimization, and simulating wind speed distribution in the wind power plant through the 3D-Frandsen wake model and the incoming wind speed obtained in the step 2;
and 4, carrying out wind power cluster output prediction based on the wind speed distribution simulation result in the step 3 and the wind wheel surface equivalent wind speed theorem.
The invention further improves that: the wind speed prediction data in the step 1 are wind speed attribute data, including wind speed, wind direction, temperature, humidity and barometric pressure information, and the preprocessing of the wind speed attribute data comprises:
step 1.1, detecting continuity of wind speed attribute data, marking defect data, screening abnormal points of the data by using a proximity algorithm, marking, and interpolating the abnormal points and the defect data by using a nearest neighbor interpolation algorithm;
step 1.2, carrying out normalization processing on the wind speed attribute data obtained in the step 1.1:
in the method, in the process of the invention,for the initial wind speed attribute data,/>The normalized initial wind speed attribute data; x is x max And x min Respectively obtaining the maximum value and the minimum value in the initial wind speed attribute data;
and 1.3, screening the normalized initial wind speed attribute data by adopting a Spearman rank correlation coefficient.
The invention further improves that: the data screened in the step 1.3 comprise wind speed, air temperature, wind direction and air pressure information.
The invention further improves that: the expression of the LSTM network model is as follows:
f t =σ g (W f x t +U f h t-1 +b f );
i t =σ g (W i x t +U i h t-1 +b i );
o t =σ g (W o x t +U o h t-1 +b o );
wherein t is the current time, t-1 is the previous time, x t For input vector of LSTM unit, i.e. wind speed attribute data, f t Activation vector i for forgetting gate t Activation vector for input/update gate, o t To output the activation vector of the gate, h t To conceal the state vector, i.e. the output vector of the LSTM cell,h t-1 the state vector is hidden for the previous time instant,inputting an activation vector for a cell, processing information representing a current state, c t For memory cell state vector, c t-1 For the previous time, the state vector of the memory cell, W f 、W i 、W o And W is c Weight matrix of forgetting gate, input/update gate, output gate and memory cell, U f 、U i 、U o And U c Bias matrices of forgetting gate, input/update gate, output gate and memory cell, respectively, b f 、b i 、b o And b c Bias vectors of the forgetting gate, the input/update gate, the output gate and the memory unit respectively, and the symbol o is Hadamard product, namely, the calculation is carried out by multiplying item by item, and sigma is calculated g Is a sigmoid activation function, sigma c (. Cndot.) and sigma h (. Cndot.) is the hyperbolic tangent function of the input as. Cndot. For input z, the hyperbolic tangent function is:
the invention further improves that: the specific steps of the step 3 include:
step 3.1, constructing Frandsen wake expressions considering wind power wake effects and boundary compensation thereof based on the Frandsen wake model:
in the formula, v 0 For incoming wind speed, v w For wake zone wind speed, r 0 Is the radius of the fan disc surface, r w For the wake cross-sectional radius at x downstream of the wind turbine, alpha is the wake diffusion coefficient,to compensate the coefficient, the external turbulence of the control body is expressedThe compensation part duty ratio, kappa is the compensation attenuation coefficient and represents the compensation rate of the turbulence outside the control body to the control body;
and 3.2, correcting the Frandsen wake expression in the step 3.1, setting the radial wake velocity loss rate to be subjected to Gaussian distribution, and establishing a 3D-Frandsen wake model based on Gaussian distribution optimization, wherein the expression is as follows:
wherein r is the radial radius taking the axis of the wind wheel as the center of a circle, and sigma is the standard deviation of the radial wake flow speed loss distribution; step 3.3, simulating a wind speed value of a wind turbine generator in the wind power plant through a 3D-Frandsen wake model and an incoming wind speed, wherein v is set w-sum In order to simulate the wind speed value of each wind turbine generator wake after superposition at a target point, a square sum superposition model is adopted to simulate the wind speed of the superposition part of the wind turbine wake areas, and the expression is as follows:
in the formula, v wj For the wind speed of the j-th wind turbine generator in the wake area of the target point, when the target point is not in the wake influence area, v wj =v 0 J is the number of wind turbines.
The invention further improves that: in the step 4, based on the wind wheel surface equivalent wind speed theorem, a fan power curve is optimized, and a fan power output expression taking the wind wheel surface equivalent wind speed into consideration is provided:
P=2ρS 0 v rs 3 α(1-α) 2 ;
in the formula, v rs Is the equivalent wind speed of the wind wheel surface, v i For the wind speed value at the different wind speed distribution areas i of the wind wheel surface, S i V is i The area of the placeDomain z i The area of the wind wheel surface at the position S 0 The wind turbine is characterized in that the wind turbine is a wind turbine surface area, namely a fan swept area, n is the wind turbine surface wind speed distribution area number, ρ is the air density, and α is the wake diffusion coefficient.
The beneficial effects of the invention are as follows: 1) The LSTM network model comprehensively considering the historical data and the digital weather forecast data has better precision in the short-term wind speed prediction of the wind power plant; 2) The equivalent wind speed and the corresponding power based on the 3D-Frandsen wake model are adopted to better accord with the actual running condition of each wind motor in the wind power generation cluster.
Drawings
FIG. 1 is a wind power cluster yield prediction technique roadmap of the present invention;
FIG. 2 is a layout diagram of wind power cluster locations according to an embodiment of the invention;
FIG. 3a is a schematic view of attenuation of wind speed in a cross section at an axial distance of 1100m down from incoming wind according to an embodiment of the present invention;
FIG. 3b is a schematic view of attenuation of wind speed in a cross section at an axial distance of 1500m down from incoming wind according to an embodiment of the present invention;
FIG. 3c is a schematic view of the attenuation of wind speed in cross section at an axial distance of 2000m down from the incoming wind according to an embodiment of the present invention;
FIG. 3d is a schematic view of attenuation of wind speed in a cross section at an axial distance of 3000m down from incoming wind according to an embodiment of the present invention;
FIG. 3e is a schematic view of the attenuation of wind speed in cross section at 4000m axial distance down from incoming wind according to an embodiment of the present invention;
FIG. 3f is a schematic view of attenuation of wind speed in a cross section at an axial distance of 6000m from incoming wind direction according to an embodiment of the present invention;
FIG. 4 is a graph showing the comparison of the predicted result and the actual value of different models according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. It will be apparent that the described embodiments are some, but not all, embodiments of the invention. All other embodiments, which are obtained by a person skilled in the art based on the described embodiments of the invention, fall within the scope of protection of the invention.
As shown in FIG. 1, the method for predicting the ultra-short-term output of the wind power cluster under the driving of data comprises the following operations:
step 1, acquiring wind speed prediction data, including historical data and digital weather forecast data, and preprocessing the wind speed prediction data;
the wind speed prediction data is wind speed attribute data, including wind speed, wind direction, temperature, humidity and air pressure information, and the preprocessing operation includes:
detecting continuity of wind speed attribute data and marking defect data, screening and marking abnormal points of the data by using a proximity algorithm, and interpolating the abnormal points and the defect data by using a nearest neighbor interpolation algorithm, so that data defect or abnormality can be avoided;
and carrying out normalization processing on the obtained wind speed attribute data, wherein the normalization processing is as follows:
in the method, in the process of the invention,for the initial wind speed attribute data,/>The normalized initial wind speed attribute data; x is x max And x min Respectively obtaining the maximum value and the minimum value in the initial wind speed attribute data;
in order to reduce the estimation error of the contribution value caused by the difference of different data orders, the information redundancy existing in the correlation between the initial attributes is considered, the Spearman rank correlation coefficient is introduced to screen the input data, and the screened data comprise wind speed, air temperature, wind direction and air pressure information. For a capacity of n 1 Will n 1 For the original dataConversion to grade dataAnd thus its correlation coefficient ρ s The method comprises the following steps:
in the method, in the process of the invention,for the difference between the two variable levels observed, i.e. +.>i 1 Are time ordered sequence numbers.
Step 2, adopting CNN to reduce the dimension of the screened wind speed prediction data, obtaining a feature sequence after the dimension reduction, constructing an LSTM network model of wind speed prediction, and inputting the feature sequence into the LSTM network model to predict the incoming wind speed of the wind power plant;
according to the embodiment, CNN is adopted to reduce the dimension of the wind speed prediction data after screening, and the basic features in the data after dimension reduction are extracted through the modes of local connection and weight sharing, so that a feature sequence is constructed. In order to avoid the problem of gradient disappearance during processing of long-term time series of wind speed data, an LSTM network model of wind speed prediction is provided:
f t =σ g (W f x t +U f h t-1 +b f );
i t =σ g (W i x t +U i h t-1 +b i );
o t =σ g (W o x t +U o h t-1 +b o );
wherein t is the current time, t-1 is the previous time, x t For input vector of LSTM unit, i.e. wind speed attribute data, f t Activation vector i for forgetting gate t Activation vector for input/update gate, o t To output the activation vector of the gate, h t For hiding state vector, i.e. output vector of LSTM unit, h t-1 The state vector is hidden for the previous time instant,inputting an activation vector for a cell, processing information representing a current state, c t For memory cell state vector, c t-1 For the previous time, the state vector of the memory cell, W f 、W i 、W o And W is c Weight matrix of forgetting gate, input/update gate, output gate and memory cell, U f 、U i 、U o And U c Bias matrices of forgetting gate, input/update gate, output gate and memory cell, respectively, b f 、b i 、b o And b c Bias vectors of the forgetting gate, the input/update gate, the output gate and the memory unit respectively, and the symbol o is Hadamard product, namely, the calculation is carried out by multiplying item by item, and sigma is calculated g Is a sigmoid activation function, sigma c (. Cndot.) and sigma h (. Cndot.) is the hyperbolic tangent function of the input as. Cndot. For input z, the hyperbolic tangent function is:
three gates in the LSTM network model solve the problem of gradient disappearance together: 1) The input gate records the input wind speed value into the cell state; 2) The forgetting door selectively forgets redundant or secondary memory, and determines the forgetting proportion of the memory unit at the current moment to the information at the previous moment; 3) The output gate selectively takes the encoded hidden state vector and the cell state as inputs to the LSTM cell at the next time.
Step 3, considering wake effects of the wind power plant and boundary compensation thereof, constructing a 3D-Frandsen wake model based on Gaussian distribution optimization, and simulating wind speed distribution in the wind power plant through the 3D-Frandsen wake model and the incoming wind speed obtained in the step 2;
based on Frandsen wake model, taking the gas flow interacted at the boundary of the control body into consideration, providing Frandsen wake expression with boundary compensation function, further deriving a radial wake speed loss calculation formula and a 3D-Frandsen wake model based on Gaussian distribution optimization based on the radial wake speed loss approximately obeying Gaussian distribution, analyzing the influence of upstream wind power on the input wind speed of downstream wind power, simulating the wind speed of the overlapping part of the wind power generator wake area by adopting a square sum superposition model, and realizing accurate calculation of the wind speed of each point in the wind power plant.
The Frandsen wake expression taking the wind power wake effect and boundary compensation into account is constructed as follows:
in the formula, v 0 Is the incoming wind speed; v w Wind speed is wake zone; r is (r) 0 The radius of the disk surface of the fan; r is (r) w The cross-sectional radius of wake flow at the downstream x of the wind turbine; alpha is the wake diffusion coefficient;representing the ratio of the turbulence outside the control body to the compensation part of the control body for the compensation coefficient; kappa is the compensation attenuation coefficient and represents the rate at which turbulence outside the control body compensates for the control body.
Considering that the speed of the cross section of the actual inter-fan wake area is not uniform, the radial wake speed loss rate approximately follows a Gaussian distribution, and the Frandsen wake expression is required to be corrected, therefore, a 3D-Frandsen wake model based on Gaussian distribution optimization is established by setting the radial wake speed loss rate to follow the Gaussian distribution:
wherein r is a radiation radius taking the axis of the wind wheel as the center of a circle; σ is the standard deviation of the radial wake velocity deficit distribution.
When the input wind speed of a certain downstream wind turbine is calculated, the influence of all upstream wind turbine on the wind turbine is considered, namely wake superposition effect exists, and v is set w-sum In order to simulate the wind speed value of each wind turbine generator wake after superposition at a target point, a square sum superposition model is adopted to simulate the wind speed of the superposition part of the wind turbine wake areas, and the mathematical expression is as follows:
in the formula, v wj For the wind speed of the j-th wind turbine generator in the wake area of the target point, when the target point is not in the wake influence area, v wj =v 0 J is the number of wind turbines.
And 4, carrying out wind power cluster output prediction based on the wind speed distribution simulation result in the step 3 and the wind wheel surface equivalent wind speed theorem.
The method comprises the steps of considering the temporal and spatial non-uniformity of wind speed distribution in a wind power plant, establishing a fan space distribution diagram taking a natural wind direction as a reference coordinate axis, describing wind speed values of each point in the wind power plant according to a 3D-Frandsen wake model, realizing the accurate description of wind speed space-time distribution characteristics of the wind power plant, optimizing a fan power curve based on a wind wheel surface equivalent wind speed theorem, and providing a fan power output expression taking the wind wheel surface equivalent wind speed into consideration.
The fan power output P expression based on the wind wheel surface equivalent wind speed is as follows:
P=2ρS 0 v rs 3 α(1-α) 2
in the formula, v rs Is the equivalent wind speed of the wind wheel surface, v i For the wind speed at different wind speed distribution areas i of the wind wheel surface, S i V is i Region z where i The area of the wind wheel surface at the position S 0 The wind turbine is characterized in that the wind turbine is a wind turbine surface area, namely a fan swept area, n is the wind turbine surface wind speed distribution area number, ρ is the air density, and α is the wake diffusion coefficient.
This example is described in further detail below in connection with specific embodiments.
In the embodiment, the accuracy and the effectiveness of wind power plant wind speed prediction of the LSTM network model are verified by using the historical data of a certain actual wind power plant. The wind farm wind turbine monitoring and data acquisition (SCADA) system samples every 10 minutes, with the sampled data as shown in Table 1:
table 1 wind farm sample data
In the embodiment, the diameter of the fan impeller is 60m, the longitudinal spacing between fans is 7D, and the transverse spacing is 4D. In order to explore the performance of different models in the wind farm power quantization link, the embodiment sets the following 2 different calculation examples for comparison analysis:
calculation example 1: only considering the fan power performance analysis under the condition that two wind motors are vertically arranged.
Calculation example 2: experimental study with reference to Vesta a vertically staggered wind farm arrangement was set up, as shown in fig. 2, with D being the fan wheel diameter.
The input wind speeds and power outputs for the different wind farm models for the 2 calculations are shown in Table 2:
TABLE 2 comparison of wind Power under different calculation examples
Note that: v represents the average input wind speed of the wind farm fans; p represents the average output power of the wind farm fans.
According to the results shown in table 2, the analysis is as follows:
1) Comparing the situation without considering the wake effect with the other three situations based on different wake models, the power output of the wind power plant can be reduced to a certain extent due to the reduction of wind energy captured by the fan caused by the consideration of the loss caused by the wake effect.
2) Comparing the situation based on Park wake model with the situation based on 3D-Frandsen wake model, it can be known that the average wind speed and the average output power of the wind farm are attenuated, but the compensation effect of the wake control body external gas on the wake is not considered under the situation of Park wake model, the wake influence of the downstream fan is overestimated, and therefore the attenuation effect is stronger.
3) Comparing the situation based on the Frandsen wake model with the situation considering the 3D-Frandsen wake model, the difference between the average value of the input wind speed of the wind turbine and the average value of the output wind power is smaller, but the 3D-Frandsen wake model considers the uneven distribution of the wind speed of the wake area of the wind turbine, and has higher precision and performance when analyzing a more complex wind farm structure.
4) As can be seen from the results of comparative examples 1 and 2 in consideration of various conditions of wake effects, when the wind farm scale is increased and the number of fans is increased, the wake effect inside the wind farm will have a greater effect on the power output of the wind farm.
Further analysis of example 2 using the 3D-Frandsen wake model can result in a schematic representation of the attenuation of wind velocity in cross sections at different axial distances down the incoming wind as shown in fig. 3 a-3 f. As can be seen from fig. 3 a-3 f, after natural wind passes through the fan, there is a phenomenon of wind speed decay directly behind the fan wheel. As can be seen by comparing fig. 3a and 3b, there is a superposition of the damping effect, i.e. the back wind exhaust will again attenuate the wind speed of the attenuated natural wind. As can be seen from the results of fig. 3d, 3e and 3f, the effect of this attenuation is continuously impaired by the boundary compensation effect as the distance increases until the wind speed returns to the incoming wind speed at infinity.
To further analyze the effect of upstream fan wake on downstream fans, a wind speed decay factor ω is defined v Power attenuation factor omega P The method comprises the following steps:
in the formula, v in-w And P w Inputting wind speed and power v for wind turbines respectively taking wake effects into account 0 And P 0 Wind turbine incoming flow wind speed and power which do not consider wake influence are respectively.
The equivalent wind speed of each exhaust fan is obtained by a wind wheel surface equivalent wind speed method, and further the wind speed attenuation and the power attenuation of each exhaust fan can be obtained as shown in table 3:
TABLE 3 attenuation coefficient under the conditions of EXAMPLE 2
The results in table 3 show that when natural wind passes through the front exhaust fan, a certain attenuation exists, the running state of the rear exhaust fan is further affected, and the wake flow has a superposition phenomenon, namely, the incoming flow passes through a plurality of fans, the wind speed is attenuated for a plurality of times, the input wind speed of the rear exhaust fan is greatly reduced, and the power output of the rear exhaust fan is reduced by about 30%.
In order to evaluate the effectiveness of the prediction method more objectively and accurately, the embodiment selects 4 error evaluation indexes, namely root mean square error I RMSE Square absolute error I MAE Normalized root mean square error I NRMSE And normalized average absolute errorDifference I NMAE The expressions are respectively:
wherein x is i Andrespectively true value and measured value, x ni And->The normalized true value and the measured value are respectively, and N is the number of samples.
The errors of a hybrid network model (indicated by CL 1) which simultaneously considers the historical data and the digital weather forecast data, a hybrid network model (indicated by CL 2) which only considers the historical data and other typical prediction models are compared and analyzed, the wind speed of 3h in the future is predicted by taking 10min as a step length, and the result of the defined error index is shown in the table 4:
TABLE 4 prediction results of various methods
Note that: ARIMA and RNN represent a differentially integrated moving average autoregressive model and a recurrent neural network model, respectively.
The data in table 4 illustrates that the accuracy of predictions can be improved to some extent by introducing digital weather forecast data. Meanwhile, the CNN-LSTM network has better performance compared with other traditional prediction models. Because of forgetfulness existing in the structure of the RNN network, the prediction effect is not very good after the distance is prolonged, and various errors are the largest in several prediction models.
In order to more intuitively show the performance of the combined model in wind power plant power prediction, fig. 4 shows a comparison chart of the prediction results and actual values of different models, and the model is compared with a CNN-LSTM model for directly predicting wind power plant power. As can be seen from FIG. 4, the prediction results are approximately similar to the actual values, but the prediction efficiency of the model of the invention is obviously better than that of the CNN-LSTM model for directly predicting the power of the wind power plant.
In summary, the wind power plant incoming flow wind speed is predicted by adopting a Convolutional Neural Network (CNN) -Long and Short Term Memory (LSTM) network fusion model, a 3D-Frandsen wake model considering wake effect is constructed according to the topological structure of a wind power cluster in the incoming flow wind direction, and further the output power of each fan is decoupled and analyzed, and the predicted value of the wind power plant output is obtained after integration. According to the invention, the LSTM network model comprehensively considering the historical data and the digital weather forecast data has better precision in the short-term wind speed forecast of the wind power plant, and the equivalent wind speed and the corresponding power based on the 3D-Frandsen wake model are adopted to better accord with the actual running condition of each wind motor in the wind power plant group.
It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
While the foregoing is directed to embodiments of the present invention, other and further details of the invention may be had by the present invention, it should be understood that the foregoing description is merely illustrative of the present invention and that no limitations are intended to the scope of the invention, except insofar as modifications, equivalents, improvements or modifications are within the spirit and principles of the invention.
Claims (2)
1. The ultra-short-term output prediction method of the wind power cluster under data driving is characterized by comprising the following steps of: the method comprises the following operations:
step 1, acquiring wind speed prediction data, including historical data and digital weather forecast data, and preprocessing the wind speed prediction data;
step 2, adopting CNN to reduce the dimension of the preprocessed wind speed prediction data, obtaining a feature sequence after the dimension reduction, constructing an LSTM network model of wind speed prediction, and inputting the feature sequence into the LSTM network model to predict the incoming wind speed of the wind power plant;
step 3, considering wake effects of the wind power plant and boundary compensation thereof, constructing a 3D-Frandsen wake model based on Gaussian distribution optimization, and simulating wind speed distribution in the wind power plant through the 3D-Frandsen wake model and the incoming wind speed obtained in the step 2;
step 4, wind power cluster output prediction is carried out based on the wind speed distribution simulation result in the step 3 and the wind wheel surface equivalent wind speed theorem;
the wind speed prediction data in the step 1 are wind speed attribute data, including wind speed, wind direction, temperature, humidity and barometric pressure information, and the preprocessing of the wind speed attribute data comprises:
step 1.1, detecting continuity of wind speed attribute data, marking defect data, screening abnormal points of the data by using a proximity algorithm, marking, and interpolating the abnormal points and the defect data by using a nearest neighbor interpolation algorithm;
step 1.2, carrying out normalization processing on the wind speed attribute data obtained in the step 1.1:
in the method, in the process of the invention,for the initial wind speed attribute data,/>For normalized initial wind speed attribute data, x max And x min I is the maximum value and the minimum value in the initial wind speed attribute data respectively 1 Sequence numbers ordered by time;
step 1.3, screening the normalized initial wind speed attribute data by adopting a Spearman rank correlation coefficient, and specifically comprising the following steps:
the method comprises the steps of (1) reducing a contribution value estimation error caused by different data order differences, considering information redundancy existing in correlation among initial attributes, introducing a Spearman rank correlation coefficient to screen input data, wherein the screened data comprise wind speed, air temperature, wind direction and air pressure information;
for a capacity of n 1 Will n 1 For the original dataConversion into rating data->And thus its correlation coefficient ρ s The method comprises the following steps:
in the method, in the process of the invention,for the difference between the two variable levels observed, i.e. +.>i 1 Sequence numbers ordered by time;
the specific steps of the step 3 include:
step 3.1, constructing Frandsen wake expressions considering wind power wake effects and boundary compensation thereof based on the Frandsen wake model:
in the formula, v 0 For incoming wind speed, v w For wake zone wind speed, r 0 Is the radius of the fan disc surface, r w For the wake cross-sectional radius at x downstream of the wind turbine, alpha is the wake diffusion coefficient,for the compensation coefficient, represent the ratio of the turbulence outside the control body to the compensation part of the control body, and κ is the compensation attenuation coefficient and represents the compensation rate of the turbulence outside the control body to the control body;
and 3.2, correcting the Frandsen wake expression in the step 3.1, setting the radial wake velocity loss rate to be subjected to Gaussian distribution, and establishing a 3D-Frandsen wake model based on Gaussian distribution optimization, wherein the expression is as follows:
wherein r is the radial radius taking the axis of the wind wheel as the center of a circle, and sigma is the standard deviation of the radial wake flow speed loss distribution;
step 3.3, simulating a wind speed value of a wind turbine generator in the wind power plant through a 3D-Frandsen wake model and an incoming wind speed, wherein v is set w-sum In order to simulate the wind speed value of each wind turbine generator wake after superposition at a target point, a square sum superposition model is adopted to simulate the wind speed of the superposition part of the wind turbine wake areas, and the expression is as follows:
in the formula, v wj For the wind speed of the j-th wind turbine generator in the wake area of the target point, when the target point is not in the wake influence area, v wj =v 0 J is the number of wind turbines;
in the step 4, based on the wind wheel surface equivalent wind speed theorem, a fan power curve is optimized, and a fan power output expression taking the wind wheel surface equivalent wind speed into consideration is provided:
P=2ρS 0 v rs 3 α(1-α) 2 ;
in the formula, v rs Is the equivalent wind speed of the wind wheel surface, v i For the wind speed value at the different wind speed distribution areas i of the wind wheel surface, S i V is i Region z where i The area of the wind wheel surface at the position S 0 The wind turbine is characterized in that the wind turbine is a wind turbine surface area, namely a fan swept area, n is the wind turbine surface wind speed distribution area number, ρ is the air density, and α is the wake diffusion coefficient.
2. The method for predicting ultra-short-term output of a wind power cluster under data driving according to claim 1, wherein the method comprises the following steps of: the expression of the LSTM network model is as follows:
f t =σ g (W f x t +U f h t-1 +b f );
i t =σ g (W i x t +U i h t-1 +b i );
o t =σ g (W o x t +U o h t-1 +b o );
wherein t is the current time, t-1 is the previous time, x t For input vector of LSTM unit, i.e. wind speed attribute data, f t Activation vector i for forgetting gate t Activation vector for input/update gate, o t To output the activation vector of the gate, h t For hiding state vector, i.e. output vector of LSTM unit, h t-1 The state vector is hidden for the previous time instant,inputting an activation vector for a cell, processing information representing a current state, c t For memory cell state vector, c t-1 For the previous time, the state vector of the memory cell, W f 、W i 、W o And W is c Weight matrix of forgetting gate, input/update gate, output gate and memory cell, U f 、U i 、U o And U c Bias matrices of forgetting gate, input/update gate, output gate and memory cell, respectively, b f 、b i 、b o And b c Bias vectors, symbol +.>For Hadamard product, i.e. calculated by multiplying by item, sigma g Is a sigmoid activation function, sigma c (. Cndot.) and sigma h (. Cndot.) is the hyperbolic tangent function of the input as. Cndot. For input z, the hyperbolic tangent function is:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310847303.6A CN116822370B (en) | 2023-07-11 | 2023-07-11 | Ultra-short-term output prediction method for wind power cluster under data driving |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310847303.6A CN116822370B (en) | 2023-07-11 | 2023-07-11 | Ultra-short-term output prediction method for wind power cluster under data driving |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116822370A CN116822370A (en) | 2023-09-29 |
CN116822370B true CN116822370B (en) | 2024-02-20 |
Family
ID=88124004
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310847303.6A Active CN116822370B (en) | 2023-07-11 | 2023-07-11 | Ultra-short-term output prediction method for wind power cluster under data driving |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116822370B (en) |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102570453A (en) * | 2012-01-06 | 2012-07-11 | 甘肃省电力公司风电技术中心 | Short-term wind power prediction method and system based on multiple numerical weather prediction sources |
CN105958545A (en) * | 2015-12-25 | 2016-09-21 | 国网甘肃省电力公司电力科学研究院 | Wind power plant active emergency control method in power grid emergency state based on power prediction |
CN108629461A (en) * | 2018-05-14 | 2018-10-09 | 华北电力大学 | A kind of Near Wake Flow prediction model based on simplified momentum theorem |
CN111475909A (en) * | 2019-10-28 | 2020-07-31 | 国网江西省电力有限公司电力科学研究院 | Wind turbine generator output correlation mapping modeling method based on long-term and short-term memory network |
CN112949945A (en) * | 2021-04-15 | 2021-06-11 | 河海大学 | Wind power ultra-short-term prediction method for improving bidirectional long-short term memory network |
CN113516271A (en) * | 2021-04-05 | 2021-10-19 | 东北电力大学 | Wind power cluster power day-ahead prediction method based on space-time neural network |
CN114398843A (en) * | 2022-01-18 | 2022-04-26 | 中国大唐集团科学技术研究院有限公司中南电力试验研究院 | Three-dimensional wake flow wind speed distribution calculation method suitable for various terrains |
CN115713029A (en) * | 2022-11-01 | 2023-02-24 | 河海大学 | Wind power plant stochastic model prediction optimization control method considering delay |
WO2023093774A1 (en) * | 2021-11-26 | 2023-06-01 | 中国华能集团清洁能源技术研究院有限公司 | Deep learning-based wind power cluster power prediction method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20230037193A1 (en) * | 2021-07-26 | 2023-02-02 | Dalian University Of Technology | Wind power output interval prediction method |
-
2023
- 2023-07-11 CN CN202310847303.6A patent/CN116822370B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102570453A (en) * | 2012-01-06 | 2012-07-11 | 甘肃省电力公司风电技术中心 | Short-term wind power prediction method and system based on multiple numerical weather prediction sources |
CN105958545A (en) * | 2015-12-25 | 2016-09-21 | 国网甘肃省电力公司电力科学研究院 | Wind power plant active emergency control method in power grid emergency state based on power prediction |
CN108629461A (en) * | 2018-05-14 | 2018-10-09 | 华北电力大学 | A kind of Near Wake Flow prediction model based on simplified momentum theorem |
CN111475909A (en) * | 2019-10-28 | 2020-07-31 | 国网江西省电力有限公司电力科学研究院 | Wind turbine generator output correlation mapping modeling method based on long-term and short-term memory network |
CN113516271A (en) * | 2021-04-05 | 2021-10-19 | 东北电力大学 | Wind power cluster power day-ahead prediction method based on space-time neural network |
CN112949945A (en) * | 2021-04-15 | 2021-06-11 | 河海大学 | Wind power ultra-short-term prediction method for improving bidirectional long-short term memory network |
WO2023093774A1 (en) * | 2021-11-26 | 2023-06-01 | 中国华能集团清洁能源技术研究院有限公司 | Deep learning-based wind power cluster power prediction method |
CN114398843A (en) * | 2022-01-18 | 2022-04-26 | 中国大唐集团科学技术研究院有限公司中南电力试验研究院 | Three-dimensional wake flow wind speed distribution calculation method suitable for various terrains |
CN115713029A (en) * | 2022-11-01 | 2023-02-24 | 河海大学 | Wind power plant stochastic model prediction optimization control method considering delay |
Non-Patent Citations (3)
Title |
---|
Short-Term Wind Power Prediction Based on DBSCAN Clustering and Support Vector Machine Regression;Siqi Wang.etc;IEEE;20201231;全文 * |
基于CNN- LSTM的短期风电功率预测;赵建利;白格平;李英俊;鲁耀;;自动化仪表;20200520(05);全文 * |
基于长短期记忆网络与注意力机制的短期光伏发电预测;李清;高春燕;胡长骁;蔡文姗;;电气自动化;20200930(05);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN116822370A (en) | 2023-09-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Cheng et al. | Augmented convolutional network for wind power prediction: A new recurrent architecture design with spatial-temporal image inputs | |
CN113657662B (en) | Downscaling wind power prediction method based on data fusion | |
Li et al. | Short-term prediction of the power of a new wind turbine based on IAO-LSTM | |
CN111900713A (en) | Multi-scene power transmission network planning method considering load and wind power randomness under network source coordination | |
CN111612244A (en) | QRA-LSTM-based method for predicting nonparametric probability of photovoltaic power before day | |
Wang et al. | Wind speed forecasting based on hybrid model with model selection and wind energy conversion | |
Zhang et al. | Achieving wind power and photovoltaic power prediction: An intelligent prediction system based on a deep learning approach | |
CN113449847B (en) | Offshore wind power rolling prediction method considering second-level time series wind speed change | |
CN110909994A (en) | Small hydropower station power generation amount prediction method based on big data drive | |
CN109615142A (en) | A kind of wind farm wind velocity combination forecasting method based on wavelet analysis | |
CN113610285A (en) | Power prediction method for distributed wind power | |
CN113435595A (en) | Two-stage optimization method for extreme learning machine network parameters based on natural evolution strategy | |
CN116822370B (en) | Ultra-short-term output prediction method for wind power cluster under data driving | |
CN109766611B (en) | Wind farm power simplified prediction method considering terrain gradient | |
CN116128211A (en) | Wind-light-water combined short-term optimization scheduling method based on wind-light uncertainty prediction scene | |
CN115296298A (en) | Wind power plant power prediction method | |
Xu et al. | NWP feature selection and GCN-based ultra-short-term wind farm cluster power forecasting method | |
CN114741952A (en) | Short-term load prediction method based on long-term and short-term memory network | |
CN114139788A (en) | Method and device for predicting power of offshore wind turbine generator | |
Zhu et al. | Wind power probabilistic forecasting based on combined decomposition and deep learning quantile regression | |
Liu et al. | Ultra-short-term wind power forecasting based on stacking model | |
Liu et al. | Available power estimation of wind farms based on deep spatio-temporal neural networks | |
CN113313306B (en) | Elastic neural network load prediction method based on improved wolf optimization algorithm | |
Zhan et al. | Probability Density Prediction of Wind Power Based on BiLSTM-DCC | |
Gui et al. | A study of a photovoltaic power prediction method combining BiLSTM and GWO |
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 |