NL2036572A - Nacelle wind speed correction method and system based on mechanism model and neural network - Google Patents
Nacelle wind speed correction method and system based on mechanism model and neural network Download PDFInfo
- Publication number
- NL2036572A NL2036572A NL2036572A NL2036572A NL2036572A NL 2036572 A NL2036572 A NL 2036572A NL 2036572 A NL2036572 A NL 2036572A NL 2036572 A NL2036572 A NL 2036572A NL 2036572 A NL2036572 A NL 2036572A
- Authority
- NL
- Netherlands
- Prior art keywords
- wind speed
- data
- frequency
- low
- scada
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 55
- 238000012937 correction Methods 0.000 title claims abstract description 40
- 230000007246 mechanism Effects 0.000 title claims abstract description 31
- 230000006870 function Effects 0.000 claims description 36
- 238000012549 training Methods 0.000 claims description 22
- 238000004364 calculation method Methods 0.000 claims description 16
- 230000004913 activation Effects 0.000 claims description 15
- 238000001514 detection method Methods 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 8
- 238000000354 decomposition reaction Methods 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000005516 engineering process Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 230000009191 jumping Effects 0.000 claims 1
- 230000001537 neural effect Effects 0.000 abstract 1
- 238000010408 sweeping Methods 0.000 description 13
- 238000010586 diagram Methods 0.000 description 8
- 238000011156 evaluation Methods 0.000 description 5
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 235000007119 Ananas comosus Nutrition 0.000 description 1
- 244000099147 Ananas comosus Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000013519 translation Methods 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
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- 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
- G06N3/08—Learning methods
-
- 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
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/06—Power analysis or power optimisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Geometry (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computer Hardware Design (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Wind Motors (AREA)
Abstract
The present invention discloses a nacelle wind speed correction method and system based on a mechanism model and a neural network. A wind speed data and a power data measured in a SCADA system are selected, and a SCADA wind speed is diVided into wind 5 speed datasets under different working conditions according to a rated wind speed. A theoretical wind speed is calculated through empirical formulas; a high-frequency residual and a low-frequency residual of a real wind speed and a theoretical wind speed are calculated by means of a wavelet transform, a relationship between the SCADA wind speed and the high-frequency residual and the low-frequency residual is established by the neural 10 network, a SCADA wind speed data from actual operating wind farms is input into the trained neural network to obtain a corresponding high-frequency residual and a corresponding low-frequency residual.
Description
NACELLE WIND SPEED CORRECTION METHOD AND SYSTEM BASED ON
MECHANISM MODEL AND NEURAL NETWORK
[0001] The present application relates to the field of data processing and data transmission, and in particular to a nacelle wind speed correction method and system based on a mechanism model and a neural network.
[0002] With the development of a wind power technology, it is necessary to obtain accurate wind speed data when performing economic benefit evaluation, power generation evaluation, fleet layout, and operation control in wind fields. Due to difficulty in directly measuring a real wind speed, a nacelle wind speed recorded by a Supervisory Control and
Data Acquisition System (SCADA) is usually used as an important parameter for analysis in evaluation calculations of wind turbines. However, due to factors such as airflow distortion caused by blade rotation and a nacelle, the wind speed values measured by the
SCADA often cannot directly reflect real wind speed of impellers and must be corrected before use. Such correction of the wind speed is called a nacelle transfer function.
[0003] At present, methods for correcting a SCADA wind speed mainly include a theoretical calculation correction method and a function fitting method. Here, a theoretical calculation correction method is mainly based on an aerodynamics theory, using multiple parameters of wind turbine units and operational data recorded by the SCADA to achieve correction of the nacelle wind speed through theoretical calculation. However, due to a large number of parameters and complex internal mechanisms, such method lacks calculation accuracy and cannot cover different models. The function fitting method corrects the nacelle wind speed by fitting a direct functional relationship between the wind speed of an anemometer tower and the nacelle wind speed; the method is simple and easy to use, but greatly relies on the data of the anemometer tower. A configuration site of the anemometer tower is harsh in demands and expensive. Generally, only one typical wind turbine unit is selected to be provided with the anemometer tower for characteristic evaluation, which cannot meet needs of all wind turbines in the entire field. Therefore, such method has high application costs and limited application scenarios.
[0004] In order to overcome the shortcomings of the prior art, the present invention provides a nacelle wind speed correction method and system that utilize neural networks, theoretical characteristics, model evaluation and correction techniques, etc., and consider wind power working conditions. Specifically, the present invention provides a nacelle wind speed correction method and system based on a mechanism model and a neural network. A real wind speed is measured by installing a Light Detection and Ranging (LiDAR) at impellers, and then a physical mechanism is combined with a data model. The method and the system provided by the present invention have the characteristics of high calculation speed, high accuracy, and good generalization, and can be widely used for wind speed correction of different models of wind turbines.
[0005] In a first aspect, the present invention provides a nacelle wind speed correction method based on a mechanism model and a neural network, where the method includes the following steps:
[0006] selecting a wind speed data and a power data measured in a SCADA system, and dividing the wind speed data and the power data according to a preset rule to obtain a dataset under a variable pitch working condition and a dataset under a non-variable pitch working condition, where
[0007] the dividing based on the preset rule includes: judging whether the wind speed data 1s less than a rated wind speed data; if yes, determining that the wind speed data 1s a wind speed under the non-variable pitch working condition; else, determining that the wind speed data is a wind speed under the variable pitch working condition;
[0008] a resolution of the wind speed data and the power data in the SCADA system is 1s level;
[0009] the dataset under the variable pitch working condition and the dataset under the non- variable pitch working condition are converted into a data with a resolution of 30s level on average as a feature dataset;
[0010] obtaining air density information and sweeping area information, and calculating the air density information, the sweeping area information, the wind speed data, and the power data according to a preset empirical formula to obtain a theoretical wind speed ¥= under corresponding working conditions, where
[0011] the preset empirical formula is:
[0012] m= Poen PSV") Ya where
[0013] Fyn is the power data, ¥# is the wind speed data, # is the air density information, and S is the sweeping area information;
[0014] obtaining a real wind speed +, and then decomposing the theoretical wind speed ¥= and the real wind speed ¥: into corresponding high-frequency data and low-frequency data by means of a wavelet transform, respectively, to obtain corresponding high-frequency residuals ¥= and low-frequency residuals 7, and using the high-frequency residuals = and the low-frequency residuals ¥« as a target dataset, where the wind speed data is a feature dataset;
[0015] where the real wind speed ¥s is measured by LIDAR, and then the real wind speed *: and the theoretical wind speed ¥= are decomposed respectively by means of the wavelet transform to obtain corresponding high-frequency data Fx, ¥m and low-frequency data 2,
Vat, the Ys and ”»« are subject to subtraction to obtain a high-frequency residual ¥e between a real wind speed high-frequency data and a theoretical wind speed high-frequency data, and Y= and Ym are subject to subtraction to obtain a low-frequency residual ¥« between a real wind speed low-frequency data and a theoretical wind speed low-frequency data, specifically, after the real wind speed data is obtained, the real wind speed data is decomposed into a high-frequency data and a low-frequency data by means of the wavelet transform, corresponding time points are recorded, and then, based on the time points of the real wind speed, the theoretical wind speed is decomposed into a high-frequency data and a low-frequency data;
[0016] taking the feature dataset as input and the target dataset as output, performing training using the neural network, determining a time step by an optimization method, and establishing a wind speed-residual prediction model based on the working conditions and the high-frequency data and the low-frequency data;
[0017] inputting a SCADA wind speed data with a certain time step into a trained wind speed-residual prediction model to obtain high-frequency residual prediction value data and low-frequency residual prediction value data corresponding to the time step; and
[0018] performing linear addition on the SCADA wind speed data and the high-frequency residual prediction value data and the low-frequency residual prediction value data to obtain a final real wind speed correction value.
[0019] Alternatively, a real wind speed ”’ in front of a wind turbine hub is measured by the
LiDAR; and
[0020] a nacelle wind speed 2 is measured through a SCADA system.
[0021] Alternatively, the taking the feature dataset as input and the target dataset as output, performing training using the neural network, determining a time step by an optimization method, and establishing a wind speed-residual prediction model based on the working conditions and the high-frequency data and the low-frequency data includes:
[0022] performing training using the neural network, and establishing four wind speed- residual prediction models under two working conditions: a high-frequency residual prediction model using a SCADA wind speed as input and the high-frequency residual as output under the variable pitch working condition; a low-frequency residual prediction model using the SCADA wind speed as input and the low-frequency residual as output under the variable pitch working condition; a high-frequency residual prediction model using a SCADA wind speed as input and the high-frequency residual as output under the non-variable pitch working condition; and a low-frequency residual prediction model using the SCADA wind speed as input and the low-frequency residual as output under the non- variable pitch working condition.
[0023] Alternatively, in a process of establishing a wind speed-residual prediction model, training time steps and prediction time steps are subject to optimized solving by using a gamma detection technology, and parameters of each layer are subject to optimized combined solving by a layer-by-layer random jump method.
[0024] Alternatively, the layer-by-layer random jump method includes the following steps:
[0025] 1) determining parameters, jump ranges, and jump step sizes of random jumps:
[0026] determining the parameters that need to be optimized as the number of units, dropout value, and an activation function of a second layer, namely LSTM, and the number of units and an activation function of a fourth layer, namely a Dense layer; setting the range and jump step size of each parameter; and randomly selecting activation functions as a sigmoid function, a tanh function, and a relu function, where the parameter values are calculated using the following formula: 00271 Pi = Mia + step + rand (| "mm | Te mm where
[0028] ™ is a calculated value for a current round, +2 is a calculated value for a previous round, sx is a minimum range value, a is a maximum range value, “7 is the jump step size, and rand) is a function of randomly selecting integers within a given range, iz2ie NY
[0029] 2) in the first round, randomly initializing the number of units, the dropout value,
mapping dimensions, and the activation function of a selected layer within a given range; then dividing a certain quantity of wind speed and residual datasets into a training set and a testing set for training and testing, and calculating accuracy to obtain an accuracy value of the corresponding parameter; recording parameter combination and results; then adding a
S random positive integer or a random negative integer to the number of units and the dropout value of the selected layer, and then multiplying by the jump step size, to obtain a next set of parameter values, where the parameter values cannot exceed the given range, the activation function is also randomly selected, and the operation is repeated for five rounds; and
[0030] 3) calculating the accuracy of five sets of random parameters, selecting the parameter combination with the highest accuracy, then executing step 2) on the parameters of the fourth layer, similarly, after calculating the accuracy of the five sets of parameters, selecting the parameter combination with the highest accuracy, and finally, obtaining an optimal parameter combination.
[0031] Alternatively, after obtaining the SCADA wind speed data and determining the working conditions based on the SCADA wind speed data, the SCADA wind speed data is input into the corresponding high-frequency residual prediction model and low-frequency residual prediction model to obtain the high-frequency residual ¥= and the low-frequency residual ¥« of the corresponding time steps, finally, the high-frequency residual ¥es and the low-frequency residual ¥« of the corresponding time steps are added to the nacelle wind speed ”’z to obtain the corrected real wind speed Yo. Le.
[0032] Vp =F tb 3
[0033] In a second aspect, the present invention provides a nacelle wind speed correction system based on a mechanism model and a neural network, for realizing the nacelle wind speed correction method based on a mechanism model and a neural network, where the system includes:
[0034] a division module, where the division module is used for selecting a wind speed data and a power data measured in a SCADA system, and dividing the wind speed data and the power data according to a preset rule to obtain a dataset under a variable pitch working condition and a dataset under a non-variable pitch working condition;
[0035] a calculation module, where the calculation module 1s used for obtaining air density information and sweeping area information, and calculating the air density information, the sweeping area information, the wind speed data, and the power data according to a preset empirical formula to obtain a theoretical wind speed ¥» under corresponding working conditions;
[0036] a decomposition module, where the decomposition module is used for obtaining a real wind speed !+, and then decomposing the theoretical wind speed ¥» and the real wind speed ? into corresponding high-frequency data and low-frequency data by means of a wavelet transform, respectively, to obtain corresponding high-frequency residuals ¥e and low-frequency residuals *« | and using the high-frequency residuals Ye and the low- frequency residuals ¥« as a target dataset, where the wind speed data is a feature dataset;
[0037] an establishment module, where the establishment module is used for taking the feature dataset as input and the target dataset as output, performing training using the neural network, determining a time step by an optimization method, and establishing a wind speed-residual prediction model based on the working conditions and the high-frequency data and the low-frequency data,
[0038] an input module, where the input module is used for inputting a SCADA wind speed data with a certain time step into a trained wind speed-residual prediction model to obtain high-frequency residual prediction value data and low-frequency residual prediction value data corresponding to the time step; and
[0039] an output module, where the output module is used for performing linear addition on the SCADA wind speed data and the high-frequency residual prediction value data and the low-frequency residual prediction value data to obtain a final real wind speed correction value.
[0040] The present invention discloses a nacelle wind speed correction method and system based on a mechanism model and a neural network. A wind speed data and a power data measured in a SCADA system are selected, and a SCADA wind speed is divided into wind speed datasets under different working conditions according to a rated wind speed. A theoretical wind speed is calculated through empirical formulas; a high-frequency residual and a low-frequency residual of a real wind speed and a theoretical wind speed are calculated by means of a wavelet transform, a relationship between the SCADA wind speed and the high-frequency residual and the low-frequency residual is established by the neural network, a SCADA wind speed data from actual operating wind farms is input into the trained neural network to obtain a corresponding high-frequency residual and a corresponding low-frequency residual; and the SCADA wind speed data is in one-to-one correspondence with residual data, and linear addition is performed to obtain a real wind speed correction value. The method and the system provided by the present invention have the characteristics of high calculation speed, high accuracy, and good generalization, and can be widely used for wind speed correction of different models of wind turbines.
[0041] FIG. 1 shows a flow diagram of steps of a nacelle wind speed correction method based on a mechanism model and a neural network of the present invention;
[0042] FIG. 2 shows an architecture diagram of the nacelle wind speed correction method based on a mechanism model and a neural network of the present invention;
[0043] FIG. 3 shows a structure diagram of a neural network of the present invention; and
[0044] FIG. 4 shows a module diagram of a nacelle wind speed correction system based on a mechanism model and a neural network of the present invention.
[0045] In order to better understand the above objectives, features, and advantages of the present invention, a further detailed description of the present invention will be provided below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.
[0046] Many specific details are elaborated in the following description to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited by the specific embodiments disclosed below.
[0047] FIGs. 1 and 2 show a flow diagram and an architecture diagram of the steps of a nacelle wind speed correction method based on a mechanism model and a neural network of the present invention. The nacelle wind speed correction method 100 based on a mechanism model and a neural network includes the following steps:
[0048] S102: selecting a wind speed data and a power data measured in a SCADA system, and dividing the wind speed data and the power data according to a preset rule to obtain a dataset under a variable pitch working condition and a dataset under a non-variable pitch working condition, where
[0049] the step is used to establish an original dataset, the wind speed and power data in the
SCADA system are selected, and the wind speed and the power data are divided into a wind speed dataset under a variable pitch working condition and a wind speed dataset under a non-variable pitch working condition based on whether the wind speed reaches the rated wind speed, and specifically:
S [0050] when the wind speed data is divided based on different working conditions, it is necessary to first determine the model of wind turbines, then determine a corresponding rated wind speed value when reaching the rated power, and then use the rated wind speed as a basis for judgment. A 2MW direct-driven wind turbine in a certain wind field is taken as an example, the rated power is determined to be 2000kw and the rated wind speed is 12m/s.
[0051] Alternatively, a real wind speed ¥: in front of a wind turbine hub is measured by the
LIDAR;
[0052] a nacelle wind speed ¥« is measured through a SCADA system;
[0053] further, by taking the rated wind speed as a limit, it is judged whether the data of the wind speed Fz of the wind turbine recorded by the SCADA system is less than the rated wind speed data,
[0054] if yes, it is determined that the wind speed data is a wind speed under the non- variable pitch working condition;
[0055] else, it is determined that the wind speed data is a wind speed under the variable pitch working condition.
[0056] Alternatively, a resolution of the wind speed data and the power data in the SCADA system is 1s level; the dataset under the variable pitch working condition and the dataset under the non-variable pitch working condition are converted into a data with a resolution of 30s level on average as a feature dataset;
[0057] S104: obtaining air density information and sweeping area information, and calculating the air density information, the sweeping area information, the wind speed data, and the power data according to a preset empirical formula to obtain a theoretical wind speed ¥» under corresponding working conditions, where
[0058] the step is used to process the dataset, a SCADA wind speed, power, environment and model parameters are input into empirical formulas to obtain a corresponding theoretical wind speed. Y= Specifically, air density information and sweeping area information are obtained, and the air density information, the sweeping area information, the wind speed data, and the power data are calculated according to a preset empirical formula to obtain a corresponding theoretical wind speed ¥=;
[0059] where the empirical formula used for calculating the theoretical wind speed ¥= is as follows: . mn / ; yy 2 po
[0060] Vm = Fen / PSV JFV here + r
[0061] where Fem is generator output power, ¥x is the wind speed recorded by the SCADA,
Pis the air density, and § is the sweeping area;
[0062] the derivation process of the empirical formula is as follows:
[0063] according to the aerodynamics theory, when air mass flows through a region $ at a speed ¥, the power of air motion is expressed as: > — 3 gE
[0064] Pina = PV 5/2 (1)
[0065] in formula (1), # is the air density and & is the sweeping area;
[0066] a power coefficient Co is usually defined as: + op / KE
[0067] C,=8 {p81 2)
[0068] in formula (2), = is the mechanical power of the wind turbine; . C . . . .
[0069] according to * equation (2), the relationship between the mechanical power of the wind turbine and the power coefficient can be written as: # EN = 3 100707 Pz =0.505C, (RI 20) 6)
[0071] where 25 9B/®
[0072] 4 is a tip speed ratio, ® is a blade radius, and ®« is a generator speed;
[0073] considering that it is difficult to directly measure the mechanical power of the wind turbine, actual measured power is generator power in the SCADA system, the above formula (3) can be rewritten as: u yy 3 4 3 (00747 Peen = 051PSC (RI y@,)' = ke, “
[0075] where # is an energy conversion rate.
[0076] According to the aerodynamics theory, the airflow passing through wind turbine blades can be divided into three categories: one category represents a wind speed °° in front of the blades; one category represents a wind speed *2 when passing through the blades; and one category represents a wind speed ¥z behind the blades. An anemometer is usually mounted near the wind turbine blades, so that the SCADA wind speed can be considered as
Va. According to the aerodynamics theory, air kinetic energy absorbed by the wind turbine blades can be expressed as:
[0077] Br = PSV (vv) =0.5p8v,(% —v;) (5)
[0078] ¥2 = Va = ¥% (6)
[0079] formula (6) is substituted into above formula (5) to obtain:
[0080] x= 2 psv jv vg) (7)
[0081] Differences between the mechanical power of the wind turbine and the generator power are ignored, and formula (4) is substituted into formula (7) to obtain:
[0082] ko, = 2pSv, (Vg) (8)
[0083] From formulas (4} and (8), it can be concluded that: (00847 Vi = Fo: [QpSVi)+v, =P, [(QpPSV3) + 4 (9)
[0085] Finally, the empirical formula for the real wind speed and the SCADA wind speed is obtained as follows:
[0086] Vi = En [sv + (10)
[0087] where Fo is generator output power, and ¥+ is the wind speed recorded by the
SCADA.
[0088] S106: obtaining a real wind speed ¥:, and then decomposing the theoretical wind speed Y= and the real wind speed ¥s into corresponding high-frequency data and low- frequency data by means of a wavelet transform, respectively, to obtain corresponding high-frequency residuals ¥e and low-frequency residuals ¥«, and using the high-frequency residuals Yer and the low-frequency residuals Fai as a target dataset, where the wind speed data is a feature dataset;
[0089] in this step, the real wind speed ”+ is measured by LiDAR, and then the real wind speed Pand the theoretical wind speed Vos are decomposed respectively by means of the wavelet transform to obtain corresponding high-frequency data ¥u, sa and low-frequency data ¥u, Va, the’ and Pas are subject to subtraction to obtain a high-frequency residual Ve between the real wind speed high-frequency data and the theoretical wind speed high- frequency data, and ¥s and ”= are subject to subtraction to obtain a low-frequency residual ¥4 between the real wind speed low-frequency data and the theoretical wind speed low- frequency data.
[0090] Specifically, after the real wind speed data is obtained, the real wind speed data is decomposed into a high-frequency data and a low-frequency data by means of the wavelet transform, corresponding time points are recorded, then, based on the time points of the real wind speed, the theoretical wind speed is decomposed into a high-frequency data and a low-frequency data, and a wavelet transform formula is shown in formula (11): da VN
WoO =a] y=)
[0091] a (11)
[0092] In the formula (11), ¥) represents a mother wavelet function, #¢€R represents a contraction-expansion factor and a translation factor, ¥ must satisfy two conditions:
Jews, [od st,
[0093] first: >= Tm 0< Cf “| wlan) Joos $9
[0094] second: °
[0095] where war) [0 od is the Fourier transform of VÒ. At this point, continuous wavelet transform of a signal FO is calculated by formula (12);
CET Go SO Had
[0096] it €} {ore JV t } (12)
[0097] where Wisla) is a wavelet coefficient, Val is a conjugate of Val)
[0098] To describe a wavelet transform of a time sequence x), 1=12.., N the continuous
Cg . . i Ai, u ee wavelet transform is discretized. Given @ = 2%; c=k£2 in formula (11), at this time, the formula (11) is changed into:
[0099] A =2"2 yk) (13)
[0100] The formula for calculating a discrete wavelet coefficient of the time sequence RC is:
WI, =WT(2,k2) => xO], 0)
[0101] = (14)
[0102] Therefore, an inverse transformation formula for reconstructing an original signal through the wavelet coefficient is: x{#) + > WT, ¥ (0)
[0103] fron (15)
[0104] S108: taking the feature dataset as input and the target dataset as output, performing training using the neural network, determining a time step by an optimization method, and establishing a wind speed-residual prediction model based on the working conditions and the high-frequency data and the low-frequency data;
[0105] FIG. 3 shows the structure diagram of a neural network of the present invention,
The neural network used is a self-attention mechanism based long short-term memory recurrent neural network (Self-Attention-LSTM). The neural network has a five-layer structure, a first layer is an input layer which inputs the feature dataset; a second layer is the
LSTM layer which calculates the data of the first layer in sequence in batches through
S LSTM neurons to obtain a corresponding hidden state as input of the next layer; a third layer is a self-attention layer which maps and recombines the hidden state input by the second layer using a self-attention mechanism, and dimensions of Q, K, and V after mapping are 10; a fourth layer is a Dense layer (Dense); and a fifth layer is an output layer which outputs values obtained by final calculation of the model. The weight and bias parameters of the neural network are a relationship model between the wind speed-residual, and a residual value can be obtained by inputting the wind speed into the neural network.
[0106] Alternatively, in one embodiment, the taking the feature dataset as input and the target dataset as output, performing training using the neural network, determining a time step by an optimization method, and establishing a wind speed-residual prediction model based on the working conditions and the high-frequency data and the low-frequency data includes:
[0107] performing training using the neural network, and establishing four wind speed- residual prediction models under two working conditions: a high-frequency residual prediction model using a SCADA wind speed as input and the high-frequency residual as output under the variable pitch working condition; a low-frequency residual prediction model using the SCADA wind speed as input and the low-frequency residual as output under the variable pitch working condition; a high-frequency residual prediction model using a SCADA wind speed as input and the high-frequency residual as output under the non-variable pitch working condition; and a low-frequency residual prediction model using the SCADA wind speed as input and the low-frequency residual as output under the non- variable pitch working condition.
[0108] in a process of establishing a wind speed-residual prediction model, training time steps and prediction time steps are subject to optimized solving by using a gamma detection technology, and parameters of each layer are subject to optimized combined solving by a layer-by-layer random jump method.
[0109] Specifically, for different wind turbines, input and output time steps are determined using a gamma detection algorithm, and the parameters of each layer are determined using a layer-by-layer random jump method, but the overall structure of the neural network remains unchanged.
[0110] Gamma detection directly estimates a minimum mean square error present in a nonlinear model through observation data, and the parameters of input data can be determined in this way.
[0111] Assuming that the form of a sample data is 6,3) = Caan} 12, N} % represents the input in the {4 row, each %: may contain multiple features, ix represents n feature, and ¥: represents the ’ output. Assuming that a vector contains useful factors that affect the output, the relationship between the vector and the factors can be expressed as:
[0112] Y= f{x}+¢ (16)
[0113] J is a smooth unknown regression function, and § represents a noise random variable. Because any constant deviation can be contained in the unknown function, expectation is £&)=9 and variance is Vag) < to
[0114] If two points * and 3 UISLJEN) are very close in distance in input space, their corresponding outputs +: and Ys should also be very close in distance in the output space. If not, we believe that this difference is caused by noise $. 40) function calculates the average value of squared values of proximity distances of each point in the input space, as shown in formula (17):
Sk) = LS gf
[0115] NE (17)
[0116] where *l&]is the &u proximity point of %, i.e, *:l&lis the uw point closest to *,
Pd¥lis the ky, proximity point of :, i.e, YA&lis the ky point closest to , Il represents an Euclidean distance, and in the output space, a corresponding (3 function is shown in the following formula: 1 & os 2
FB) = lk] 3,
[0117] ING (18)
[0118] Gamma detection calculates gamma statistic F by drawing a linear regression line ata point, using the following formula:
[0119] 7 =A8+T (19)
[0120] In formula (19), a vertical intercept (when & is zero) describes a value of the gamma statistic represented by I". If F is very small, the regression function J exists, an output value ¥ highly depends on an input variable x, and close correlativity exists between input and output. If I" is large, it indicates that the input is independent of the output.
[0121] Since the value of the gamma statistic is affected by the size of the sample value, the result is normalized by considering the statistic ¥ratie The statistic ¥ratio is defined as:
Fratio= wd
[0122] a(y) (20)
[0123] In formula (20), #0)" represents a variance of the output ¥. A value ratio close to 0 indicates that the output ¥ has a high degree of predictability, while a value Fratie close to 1 indicates that the output ¥ is random and independent of the input variable x.
[0124] Relevant explanation on the selection of embedding dimension for gamma detection: T is a value of the gamma statistic corresponding to the situation when M is taken for the embedding dimension;
[0125] the smaller the gamma statistic, the more reasonable the corresponding embedding dimension, i.e., the closer the value of the statistic ¥rafiv is to 0, the more reasonable the corresponding embedding dimension is; and therefore, gamma detection can be performed respectively based on progressive increase of the embedding dimension, the statistics ratie are used for making judgments, and the appropriate embedding dimension is selected.
[0126] In the process of determining the parameters of each layer using a layer-by-layer random jump method, the layer-by-layer random jump method includes the following steps: 1) determining parameters, jump ranges, and jump step sizes of random jumps:
[0127] determining the parameters that need to be optimized as the number of units, dropout value, and an activation function of a second layer, namely LSTM, and the number of units and an activation function of a fourth layer, namely a Dense layer; setting the range and jump step size of each parameter; and randomly selecting activation functions as a sigmoid function, a tanh function, and a relu function, where the parameter values are calculated using the following formula: (0128) Me = Mh + 1p * rand (| Pe |e [mm (21)
[0129] where # is a calculated value for a current round, #1 is a calculated value for a previous round, Peis is a minimum range value, aa is a maximum range value, ¥ is the jump step size, and ¥ and() is a function of randomly selecting integers within a given range, iz2ie NT
[0130] 2) in the first round, randomly initializing the number of units, the dropout value, mapping dimensions, and the activation function of a selected layer within a given range; then dividing a certain quantity of wind speed and residual datasets into a training set and a testing set for training and testing, and calculating accuracy to obtain an accuracy value of the corresponding parameter; recording parameter combination and results; then adding a random positive integer or a random negative integer to the number of units and the dropout value of the selected layer, and then multiplying by the jump step size, to obtain a next set of parameter values, where the parameter values cannot exceed the given range, the
S activation function is also randomly selected, and the operation is repeated for five rounds; and
[0131] 3) calculating the accuracy of five sets of random parameters, selecting the parameter combination with the highest accuracy, then executing step 2) on the parameters of the fourth layer, similarly, after calculating the accuracy of the five sets of parameters, selecting the parameter combination with the highest accuracy, and finally, obtaining an optimal parameter combination.
[0132] S110: inputting a SCADA wind speed data with a certain time step into a trained wind speed-residual prediction model to obtain high-frequency residual prediction value data and low-frequency residual prediction value data corresponding to the time step; and
[0133] the SCADA second level wind speed is extracted once every 10 to 20 times the resolution multiplied by the time step according to the input time step and resolution during training. The SCADA wind speed is divided into a wind speed under a variable pitch working condition and a wind speed under a non-variable pitch working condition based on whether it reaches the rated wind speed, then, an average value is calculated based on a set resolution, the wind speed under the variable pitch working condition after resolution reduction is input into the trained high-frequency residual prediction model and the trained low-frequency residual prediction model under the variable pitch working condition to obtain the high-frequency residual prediction value and the low-frequency residual prediction value at the wind speed under the variable pitch working condition; and the wind speed under the non-variable pitch working condition after resolution reduction is input into the trained high-frequency residual prediction model and the trained low-frequency residual prediction model under the non-variable pitch working condition to obtain the high- frequency residual prediction value and the low-frequency residual prediction value at the wind speed under the non-variable pitch working condition.
[0134] S112: performing linear addition on the SCADA wind speed data and the high- frequency residual prediction value data and the low-frequency residual prediction value data to obtain a final real wind speed correction value.
[0135] After obtaining the high-frequency residual prediction value Ya and the low- frequency residual prediction value ¥s at the wind speed under the variable pitch working condition, as well as the high-frequency residual prediction value and the low-frequency residual prediction value at the wind speed under the non-variable pitch working condition, the SCADA wind speed under the variable pitch working condition, the high-frequency residual prediction value at the wind speed under the variable pitch working condition, and
S the low-frequency residual prediction value at the wind speed under the variable pitch working condition are linearly added to obtain a final real variable pitch wind speed correction value, and the SCADA wind speed under the non-variable pitch working condition, the high-frequency residual prediction value at the wind speed under the non- variable pitch working condition, and the low-frequency residual prediction value at the wind speed under the non-variable pitch working condition are linearly added to obtain a final real non-variable pitch wind speed correction value.
[0136] That is to say, a corrected real wind speed Ve is obtained by adding the high- frequency residual ”: and the low-frequency residual ¥s of the corresponding time steps to the nacelle wind speed 4, i.e.
[0137] eat Vata (22)
[0138] According to the present embodiment, the present invention discloses the nacelle wind speed correction method based on a mechanism model and a neural network. The wind speed data and the power data measured in a SCADA system are selected, and a
SCADA wind speed is divided into wind speed datasets under different working conditions according to a rated wind speed. A theoretical wind speed is calculated through empirical formulas; a high-frequency residual and a low-frequency residual of a real wind speed and a theoretical wind speed are calculated by means of a wavelet transform, a relationship between the SCADA wind speed and the high-frequency residual and the low-frequency residual is established by the neural network, a SCADA wind speed data from actual operating wind farms is input into the trained neural network to obtain a corresponding high-frequency residual and a corresponding low-frequency residual; and the SCADA wind speed data is in one-to-one correspondence with residual data, and linear addition is performed to obtain a real wind speed correction value. The method and the system provided by the present invention have the characteristics of high calculation speed, high accuracy, and good generalization, and can be widely used for wind speed correction of different models of wind turbines.
[0139] FIG. 4 shows a module diagram of a nacelle wind speed correction system based on a mechanism model and a neural network of the present invention. The nacelle wind speed correction system 200 based on a mechanism model and a neural network, for realizing the nacelle wind speed correction method based on a mechanism model and a neural network, includes:
[0140] a division module 210, where the division module is used for selecting a wind speed
S data and a power data measured in a SCADA system, and dividing the wind speed data and the power data according to a preset rule to obtain a dataset under a variable pitch working condition and a dataset under a non-variable pitch working condition;
[0141] a calculation module 220, where the calculation module is used for obtaining air density information and sweeping area information, and calculating the air density information, the sweeping area information, the wind speed data, and the power data according to a preset empirical formula to obtain a theoretical wind speed Y= under corresponding working conditions;
[0142] a decomposition module 230, where the decomposition module is used for obtaining a real wind speed ”s, and then decomposing the theoretical wind speed ¥= and the real wind speed ¥s into corresponding high-frequency data and low-frequency data by means of a wavelet transform, respectively, to obtain corresponding high-frequency residuals ¥a and low-frequency residuals ¥«, and using the high-frequency residuals ¥e and the low- frequency residuals ¥« as a target dataset, where the wind speed data is a feature dataset;
[0143] an establishment module 240, where the establishment module is used for taking the feature dataset as input and the target dataset as output, performing training using the neural network, determining a time step by an optimization method, and establishing a wind speed-residual prediction model based on the working conditions and the high-frequency data and the low-frequency data;
[0144] an input module 250, where the input module is used for inputting a SCADA wind speed data with a certain time step into a trained wind speed-residual prediction model to obtain high-frequency residual prediction value data and low-frequency residual prediction value data corresponding to the time step; and
[0145] an output module 260, where the output module is used for performing linear addition on the SCADA wind speed data and the high-frequency residual prediction value data and the low-frequency residual prediction value data to obtain a final real wind speed correction value.
[0146] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working process of an electronic device of the present invention can refer to the corresponding processes in the aforementioned embodiments, as well as the same parts and beneficial effects as the aforementioned embodiments, which will not be repeated here.
[0147] According to the present embodiment, the present invention discloses the nacelle 5S wind speed correction system based on a mechanism model and a neural network. The wind speed data and the power data measured in a SCADA system are selected, and SCADA wind speed is divided into wind speed datasets under different working conditions according to rated wind speed. Theoretical wind speed is calculated through empirical formulas; a high-frequency residual and a low-frequency residual of a real wind speed and a theoretical wind speed are calculated by means of a wavelet transform, a relationship between the SCADA wind speed and the high-frequency residual and the low-frequency residual is established by the neural network, a SCADA wind speed data from actual operating wind farms is input into the trained neural network to obtain a corresponding high-frequency residual and a corresponding low-frequency residual; and the SCADA wind speed data is in one-to-one correspondence with residual data, and linear addition is performed to obtain a real wind speed correction value. The method and the system provided by the present invention have the characteristics of high calculation speed, high accuracy, and good generalization, and can be widely used for wind speed correction of different models of wind turbines.
[0148] In the several embodiments provided in the present application, it should be understood that the disclosed method and system can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the modules is only a logical functional division. In actual implementation, there may be other division methods, for example: multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, coupling, or direct coupling, or communication connection between the various components displayed or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical, or in other forms.
[0149] The modules mentioned above as separate components can be, or may not be physically separated, and the components displayed as modules can be, or may not be physical units; the modules or the components can be located in one place or distributed across multiple network units; and some or all of the units can be selected according to actual needs to achieve the purpose of the present embodiment.
[0150] In addition, in each embodiment of the present invention, all functional units can be integrated into one processing unit, or each unit can be used as a separate unit, or two or more units can be integrated into one unit; and the integrated units mentioned above can be implemented in the form of hardware, or can be implemented in the form of hardware and software functional units.
S [0151] Those of ordinary skill in the art can understand that all or part of the steps to implement the above method embodiments can be completed through hardware related to program instructions. The aforementioned program can be stored in computer readable storage media. When executed, the program executes the steps including the above method embodiments. The aforementioned storage media include: mobile storage devices, Read-
Only Memory (ROM), Random Access Memory (RAM), disks or CDs, and other media that can store program codes.
[0152] Alternatively, if the integrated units of the present invention are implemented in the form of software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, essentially or in other words, the technical solution of the embodiments of the present invention or the part contributing to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions to enable a computing device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage media include various media that can store program codes, such as mobile storage devices, ROMs, RAMs, disks, or CDs.
[0153] Although some specific embodiments of the present invention have been explained in detail through examples, those skilled in the art should understand that the above examples are only for illustrative purposes but not to limit the scope of the present invention. Those skilled in the art should also understand that multiple modifications can be made to the embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is limited by the appended claims.
Claims (7)
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311047504.4A CN116757101B (en) | 2023-08-21 | 2023-08-21 | Cabin wind speed correction method and system based on mechanism model and neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
NL2036572A true NL2036572A (en) | 2024-03-25 |
Family
ID=87959518
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
NL2036572A NL2036572A (en) | 2023-08-21 | 2023-12-18 | Nacelle wind speed correction method and system based on mechanism model and neural network |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN116757101B (en) |
NL (1) | NL2036572A (en) |
Family Cites Families (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102562469B (en) * | 2011-12-27 | 2014-01-22 | 华北电力大学 | Short-term wind driven generator output power predicting method based on correction algorithm |
EP3186998A4 (en) * | 2014-08-29 | 2018-09-26 | Zunum Aero, Inc. | System and methods for implementing regional air transit network using hybrid-electric aircraft |
CN110414045B (en) * | 2019-06-18 | 2023-08-11 | 东华大学 | Short-term wind speed prediction method based on VMD-GRU |
CN110261394B (en) * | 2019-06-24 | 2022-09-16 | 内蒙古工业大学 | Online real-time diagnosis system and method for damage of fan blade |
CN110566406B (en) * | 2019-10-16 | 2020-08-04 | 上海海事大学 | Wind turbine generator set real-time variable pitch robust control system and method based on reinforcement learning |
CN113537582B (en) * | 2021-07-02 | 2022-05-24 | 东北电力大学 | Photovoltaic power ultra-short-term prediction method based on short-wave radiation correction |
CN114091377B (en) * | 2022-01-21 | 2022-05-13 | 南京理工大学 | Method for calculating wake flow wind speed of dynamic double-Gaussian wind turbine based on spatial variation |
CN115374710A (en) * | 2022-09-19 | 2022-11-22 | 兰州理工大学 | Wind speed prediction method based on improved gull optimization algorithm and multi-core extreme learning machine |
-
2023
- 2023-08-21 CN CN202311047504.4A patent/CN116757101B/en active Active
- 2023-12-18 NL NL2036572A patent/NL2036572A/en unknown
Also Published As
Publication number | Publication date |
---|---|
CN116757101B (en) | 2023-11-07 |
CN116757101A (en) | 2023-09-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108334672B (en) | Maximum wind energy capturing method of variable-speed wind generating set based on effective wind speed estimation | |
US10337498B2 (en) | Method and device for detecting equivalent load of wind turbine | |
Lee et al. | A kernel plus method for quantifying wind turbine performance upgrades | |
Yan et al. | A general method to estimate wind farm power using artificial neural networks | |
Richmond et al. | Stochastic assessment of aerodynamics within offshore wind farms based on machine-learning | |
CN107045574B (en) | SVR-based effective wind speed estimation method for low wind speed section of wind generating set | |
Dimitrov et al. | From SCADA to lifetime assessment and performance optimization: how to use models and machine learning to extract useful insights from limited data | |
Kaplan et al. | A novel method based on Weibull distribution for short-term wind speed prediction | |
CN116624343B (en) | Wind turbine generator tower abnormal vibration monitoring and health degree evaluation method and system | |
CN115062668A (en) | Harmonic parameter detection method and system based on RAdam optimization width learning | |
Noppe et al. | Modeling of quasi-static thrust load of wind turbines based on 1 s SCADA data | |
Conti et al. | Probabilistic estimation of the Dynamic Wake Meandering model parameters using SpinnerLidar-derived wake characteristics | |
Ji et al. | Structural performance degradation identification of offshore wind turbines based on variational mode decomposition with a Grey Wolf Optimizer algorithm | |
US20220243347A1 (en) | Determination method and determination apparatus for conversion efficiency of hydrogen production by wind-solar hybrid electrolysis of water | |
Branlard et al. | A digital twin solution for floating offshore wind turbines validated using a full-scale prototype | |
Zhou et al. | Importance of platform mounting orientation of Y-shaped semi-submersible floating wind turbines: A case study by using surrogate models | |
CN112052604B (en) | Method, system, equipment and readable medium for predicting equivalent fatigue load of fan | |
NL2036572A (en) | Nacelle wind speed correction method and system based on mechanism model and neural network | |
Wang et al. | A wind speed prediction method based on improved empirical mode decomposition and support vector machine | |
Karami et al. | Probabilistic neural network to quantify uncertainty of wind power estimation | |
KR20210126207A (en) | Method for building fatigue load models for wind turbine | |
Tao et al. | A combined approach for very short term wind power probability forecast | |
CN114298300A (en) | Uncertainty prediction method, uncertainty prediction device, electronic device, and storage medium | |
Conti et al. | Calibration and validation of the Dynamic Wake Meandering model Part I: Bayesian estimation of model parameters using SpinnerLidar-derived wake characteristics | |
CN111950763B (en) | Method for predicting output power of distributed wind power station |