A method, model and system for predicting wake in wind farms
By dividing the wake region of a wind farm into near-wake and far-wake regions, an adaptive wind farm engineering wake model is constructed, which solves the problem of insufficient accuracy in calculating velocity loss in the wake region and improves the power generation efficiency and economic benefits of the wind farm.
Patent Information
- Application Number
- CN202310236517.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-13
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-03-13
AI Technical Summary
Existing technologies fail to effectively consider the impact of near-wake on far-wake in wind farms, resulting in insufficient accuracy in calculating velocity loss in the wake region, which affects wind farm power generation and operation and maintenance costs.
The wake region of a wind farm is divided into near-wake and far-wake regions. An adaptive wind farm engineering wake model is constructed. Wake features are extracted using machine learning methods, and the wake expansion rate is optimized by combining genetic algorithms. The relationship between incoming flow information and wake features is established, thus constructing an adaptive wind farm engineering wake model.
It improves the accuracy of velocity loss calculation in the wake region, optimizes wind farm layout, and reduces power generation loss and operation and maintenance costs.
Smart Images

Figure CN116227357B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of new energy wind power technology, and in particular relates to a method, model and system for predicting wind farm wake. Background Technology
[0002] To achieve higher power generation, besides increasing the size of wind turbines and continuously improving the single-unit capacity of wind turbines, it is also necessary to expand the scale of wind farms to deploy more wind turbines. However, under normal circumstances, the increase in the number of wind turbines is not proportional to the increase in power generation. Once a certain number of wind turbines is reached, further increasing the number of turbines can actually lead to a decrease in power generation. This is because of the wake effect in wind farms, which causes downstream wind turbines to capture lower wind speeds, increase turbulence intensity, and reduce power generation. In addition, the wake effect of wind turbines also increases the fatigue load on downstream wind turbines and increases the probability of failure, thereby increasing the operation and maintenance costs of wind farms.
[0003] To reduce the impact of wake effects on wind turbines and improve the economic efficiency and safe operation of wind farms, it is necessary to study the wake characteristics and wake field details of wind turbines. Early, widely accepted, and efficient engineering wake models only analyzed the far wake region, neglecting the influence mechanism of near-wake velocity distribution on the far wake. Furthermore, existing engineering wake models only perform well under specific experimental conditions and still have significant errors in actual wind farms. In addition, considering the complex inflow atmospheric boundary layer flow field's influence on meteorological factors such as wind speed, wind direction, and stability, it is necessary to consider the wake velocity distribution of the wind farm area holistically and introduce intelligent algorithms such as machine learning to further improve the computational accuracy of engineering wake models. Summary of the Invention
[0004] The purpose of this invention is to provide a wind farm wake prediction method, model and system. By further considering the influence of near wake development characteristics on far wake, it effectively improves the accuracy of velocity loss calculation in the wake region, better serves the optimization layout work, and solves the problem of large errors in the power calculation of existing technologies applied to actual wind farms.
[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:
[0006] This invention relates to a method for predicting wake in wind farms, the method comprising the following steps:
[0007] S1: Based on the distance between the wake section and the wind turbine, the wake region of the wind turbine is divided into the near wake region and the far wake region;
[0008] S2: Construct wake models for the near-wake and far-wake regions of the wind turbine, respectively;
[0009] S3: Introduce the wake superposition method to construct a full-field wake model;
[0010] S4: The wake expansion rates k and k in the near and far wake regions are... * As a parameter to be determined, SCADA data from the wind farm was selected, and wake features were extracted from the data at each time point to obtain the optimal wake expansion rate dataset.
[0011] S5: Using wind farm inflow data as input and wake parameter data as output, train a machine learning model, establish the relationship between inflow information and wake characteristics, and construct an adaptive wind farm engineering wake model.
[0012] S6: Based on the input flow information, the wake parameters are predicted by the trained machine learning model, and then combined with the analytical wake model to perform wake-related calculations within the wind farm.
[0013] Furthermore, the wake distribution in the near-wake region exhibits a double Gaussian pattern, while the wake distribution in the far-wake region exhibits a single Gaussian pattern. The boundary between the near and far wake regions can be determined by the following formula:
[0014]
[0015] Where a = 0.58, b = 0.077, and c is a proportionality coefficient, determined based on the actual wind field conditions; TI x,hub The turbulence intensity in the flow direction at the hub height is given by D, where D is the runner diameter and C is the runner diameter. T This is the thrust coefficient;
[0016] Let x represent the distance from the wake section to the wind turbine. When x < x d At that time, the wake in the near-wake region exhibits a double Gaussian distribution.
[0017] Furthermore, the wake in the near-wake region exhibits a double Gaussian distribution, and the Gaussian shape function f(r(y,z),σ(x)) can be expressed as:
[0018]
[0019] Where r0 is the radial distance from the minimum point of the wake wind speed relative to the hub centerline, σ represents the spread function, y represents the radial distance of the spatial point from the hub center in the horizontal direction with the hub center position as the zero point, and z represents the height difference between the spatial point and the ground in the vertical direction with the ground as the zero point. h Indicates the wheel hub height.
[0020] Furthermore, the wake in the far wake region exhibits a single Gaussian distribution, and the Gaussian shape function f(r(y,z),σ(x)) can be expressed as:
[0021]
[0022] Where r represents the distance from any position within the wake cross section to the wake centerline, and σ represents the spread function.
[0023] Furthermore, the feature extraction steps of the wake model are as follows:
[0024] S4.1: Determine the layout of the wind farm and the basic parameters of the wind turbines;
[0025] S4.2: Determine wind speed, wind direction, and output power data for each unit, and determine the wake expansion rates k and k in the near and far wake regions. * Set as a variable parameter;
[0026] S4.3: Establish the relationship between the wake expansion rate and the calculated power generation based on the dataset in step 2;
[0027] S4.4: Compare the actual output power of the wind farm with the power calculated by the wake model to assess whether the wake expansion rate meets the requirements;
[0028] S4.5: Optimize the above problem using a genetic algorithm to obtain the optimal wake expansion rate;
[0029] S4.6: Perform the above steps for each recording time of the SCADA data to establish a wake feature set.
[0030] Furthermore, the basic parameters of the wind turbine include rotor diameter, hub height, thrust coefficient curve, and power curve.
[0031] Furthermore, the determination of the wake expansion rate parameter is as follows: the power recorded by retrieving SCADA data is compared with the power calculated by the wake model, and the root mean square error (RMSE) is used for evaluation.
[0032] The present invention has the following beneficial effects:
[0033] 1. This invention further considers the impact of near-wake development characteristics on far-wake, and introduces machine learning methods to extract the incoming flow characteristics that affect the wake distribution, establish the relationship between incoming flow information and wake characteristics, and construct an adaptive wind farm engineering wake model, which effectively improves the accuracy of velocity loss calculation in the wake region and better serves the optimization layout work.
[0034] 2. This invention uses a genetic algorithm to determine the wake expansion rate, solving the problem that existing technologies, due to numerous simplification assumptions in model construction, cannot fully consider complex inflow wind conditions when determining the wake expansion rate. This results in engineering wake models performing well on experimental data, but exhibiting significant errors in power calculations for actual wind farms.
[0035] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description
[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 This is a flowchart of an intelligent wake prediction method based on the characteristics of wake evolution in wind farms;
[0038] Figure 2 This is a schematic diagram of an intelligent wake prediction method that simulates the wake evolution characteristics of wind farms;
[0039] Figure 3 This is a schematic diagram showing the changes in wake wind speed distribution in the horizontal direction;
[0040] Figure 4 This is a schematic diagram showing the vertical distribution of wake wind speed. Detailed Implementation
[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0042] In the description of this invention, it should be understood that the terms "opening", "upper", "lower", "thickness", "top", "middle", "length", "inner", "around", etc., which indicate orientation or positional relationship, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting this invention.
[0043] Please see Figure 1-4 As shown, this invention is an intelligent wake prediction method that reflects the wake evolution characteristics of wind farms. The specific steps are as follows:
[0044] First, an engineering wake model reflecting the evolution characteristics of wind farm wakes is proposed. Based on the distance between the wake section and the wind turbine, the wind turbine wake region is divided into near-wake and far-wake regions. The wake wind speed distribution spatially evolves from a double Gaussian model to a single Gaussian model as the downstream distance increases. The relevant wake model is constructed below:
[0045] If we ignore the viscous and pressure terms in the momentum equation, and apply the mass and momentum conservation equations, we can obtain the following wake equation:
[0046] ρ∫U W (U ∞ -U W )dA=T (1)
[0047] ρ represents atmospheric density, U W U represents the wake velocity. ∞ Let A represent the velocity of the free flow, A represent the wake cross-sectional area at any downstream distance, and T represent the wind turbine thrust, which can be determined by the following formula:
[0048]
[0049] C T A represents the thrust coefficient, and A0 represents the effective area of the wind turbine runner surface subjected to thrust T.
[0050] Self-similarity in the wake describes the normalized velocity loss as:
[0051]
[0052] In the formula, C(σ(x)) represents the amplitude function of the maximum normalized velocity deficit, and f(r(y,z),σ(x)) represents the Gaussian shape function of the wake distribution. The wind turbine wake is divided into a near-wake region and a far-wake region. The wake distribution in the near-wake region exhibits an asymmetric double-Gaussian pattern, while the wake distribution in the far-wake region exhibits a symmetrical single-Gaussian pattern. The boundary between the near and far wake regions can be determined by the following formula:
[0053]
[0054] Where, x d This represents the boundary between the far wake and the near wake, where a = 0.58, b = 0.077, and c is a proportionality coefficient determined based on the actual wind field conditions. TI x,hub The turbulence intensity in the flow direction at the hub height.
[0055] Let x represent the distance from the wake section to the wind turbine. When x < x d At this time, the wake in the near-wake region exhibits a double Gaussian distribution, and the Gaussian shape function f(r(y,z),σ(x)) can be expressed as:
[0056]
[0057] Where r0 is the radial distance from the minimum point of the wake wind speed relative to the hub centerline, and σ represents the spread function, it can be expressed as:
[0058]
[0059] Where ε1 is the initial wake radius, which can be determined based on the runner diameter D and the thrust coefficient C. T The initial wake radius ε1 is calculated using the following formula:
[0060]
[0061]
[0062] Substituting equations (3) and (5) into equation (1), we get:
[0063]
[0064] The following expression can be obtained by integration:
[0065] T=πρU ∞ 2 C(σ(x))(MC(σ(x))N) (10)
[0066] Where M and N are error functions, the specific calculation formula is as follows:
[0067]
[0068]
[0069] In the formula
[0070] Substituting equation (2) into equation (7), we can obtain the amplitude function C(σ(x)) as follows:
[0071]
[0072] when When the equation is solved, the modulus of the complex root is chosen to approximate the solution.
[0073]
[0074] When x > x d At this time, the wake in the far wake region exhibits a single Gaussian distribution, and the Gaussian shape function f(r(y,z),σ(x)) can be expressed as:
[0075]
[0076] The self-similarity normalized velocity deficit in a single Gaussian wake can be expressed as:
[0077]
[0078] The velocity of the far wake can then be expressed as:
[0079]
[0080] Substituting equations (2) and (16) into equation (1) and integrating, we get:
[0081]
[0082] By solving equation (17), we can obtain
[0083]
[0084] in,
[0085] Substituting equation (18) into equation (15) yields
[0086] (20)
[0087] Where z h This represents the wheel hub height.
[0088] Since we are calculating the velocity field of the entire wind farm, we need to consider the superposition of wakes, so we introduce a wake superposition method:
[0089]
[0090] Where Δu is the velocity loss caused by n upstream wind turbines, Δu i The speed loss is caused by the i-th upstream wind turbine.
[0091] After constructing the full-field wake model, the wake expansion rates k and k in the near and far wake regions are determined. * As parameters to be determined, SCADA data from the wind farm was used. First, the SCADA data was filtered to identify wind farm inflows with wake effects, as well as the yaw angle and output power data of each turbine. Based on this data, wake features were extracted from the data at each time point to obtain the optimal wake parameter dataset.
[0092] The feature extraction steps for the wake model are as follows:
[0093] (1) Prepare the layout of the wind farm and the basic parameters of the wind turbine, including rotor diameter, hub height, thrust coefficient curve and power curve, etc.
[0094] (2) Prepare wind speed, wind direction and output power data for each unit;
[0095] (3) Establish the relationship between the wake expansion rate and the calculated power generation by combining the dataset in step 2;
[0096] (4) Establish an optimization model, retrieve the power recorded by SCADA data and compare it with the power calculated by the wake model, and use the root mean square error (RMSE) for evaluation.
[0097] (5) The above problem is optimized using a genetic algorithm to obtain the optimal wake expansion rate;
[0098] (6) Perform the above steps for each recording time of the SCADA data to establish a wake feature set.
[0099] Furthermore, the layout of the wind farm includes parameters such as the number, spacing, and arrangement of wind turbines.
[0100] Using wind farm inflow data as input and wake parameter data as output, a machine learning model is trained to establish the relationship between inflow information and wake characteristics, thus constructing an adaptive wind farm engineering wake model. For test data outside the training set, the current inflow information is input, and the wake parameters are predicted by the trained machine learning model. These parameters are then combined with the established analytical wake model to perform wake-related calculations within the wind farm.
[0101] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0102] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims
1. A wind farm wake prediction method, characterized in that, The method includes the following steps: S1: Based on the distance between the wake section and the wind turbine, the wake region of the wind turbine is divided into the near wake region and the far wake region; S2: Construct wake models for the near-wake and far-wake regions of the wind turbine, respectively; S3: Introducing a wake superposition method to construct a full-field wake model; S4: wake expansion rate of near wake region and wake expansion rate of far wake region As a pending parameter, the SCADA data of the wind farm is selected, the wake characteristics of each time point are extracted, and the best wake expansion rate data set is obtained. S5: Using wind farm inflow data as input and wake parameter data as output, train a machine learning model, establish the relationship between inflow information and wake characteristics, and construct an adaptive wind farm engineering wake model. S6: Based on the incoming flow information, the wake parameters are predicted by the trained machine learning model, and then combined with the established analytical wake model to perform wake-related calculations within the wind farm. The wake distribution in the near-wake region exhibits a double Gaussian pattern, while the wake distribution in the far-wake region exhibits a single Gaussian pattern. The boundary between the near and far wake regions can be determined by the following formula: (1) wherein, represents the demarcation point between the near wake region and the far wake region, represents the distance of the wake cross section from the wind turbine; , , is a proportionality factor, determined according to the actual application of the wind farm; is the stream-wise turbulence intensity at hub height, is the rotor diameter, is the thrust coefficient; when At that time, the wake in the near-wake region exhibits a double Gaussian distribution; The feature extraction steps for the wake model are as follows: S4.1: Determine the layout of the wind farm and the basic parameters of the wind turbines; S4.2: Determine wind speed, wind direction, and output power data for each unit, and the wake expansion rate in the near-wake region. and wake expansion rate in the far wake region Set as a variable parameter; S4.3: Based on the wind speed, wind direction, and output power data of each unit determined in step S4.2, establish the wake expansion rate in the near-wake region. and wake expansion rate in the far wake region The relationship between the power calculated by the wake model and the power calculation. S4.4: Compare the actual output power of the wind farm with the power calculated by the wake model to assess whether the wake expansion rate meets the requirements; S4.5: With the optimization objective of minimizing the error between the actual output power of the wind farm and the power calculated by the wake model, a genetic algorithm is used to optimize the wake expansion rate in the near-wake region. and wake expansion rate in the far wake region Optimize to obtain the optimal result. and ; S4.6: For each recording time of SCADA data, execute steps S4.1 to S4.5 sequentially to obtain the optimal wake expansion rate in the near-wake region corresponding to that recording time. and optimal wake expansion rate in the far wake region And the optimal value corresponding to each recording time. and This forms the wake feature set.
2. The wind farm wake prediction method according to claim 1, characterized in that, The wake in the near-wake region exhibits a double Gaussian distribution, and its Gaussian shape function can be expressed as: (2) in, and Let these represent the positive and negative branch exponent terms in the double Gaussian shape function, respectively, as follows: , (3) in, Indicates the distance from any position within the wake cross-section to the wake centerline. This represents the radial distance in the horizontal direction from the center of the wheel hub, with the center of the wheel hub as the zero point. This represents the vertical height difference between a point in space and the ground, with the ground as the zero point. Indicates the distance between the wake section and the wind turbine. Represents an extension function. Indicates wheel hub height, This represents the radial distance between the point of minimum wake velocity and the centerline of the wheel hub.
3. The wind farm wake prediction method according to claim 1, characterized in that, The wake in the far wake region exhibits a single Gaussian distribution, with a Gaussian shape function. It can be represented as: (4) in, This represents the distance from any point within the wake cross-section to the wake centerline. This represents the radial distance in the horizontal direction from the center of the wheel hub, with the center of the wheel hub as the zero point. This represents the vertical height difference between a point in space and the ground, with the ground as the zero point. This indicates the distance between the wake section and the wind turbine. This indicates an extension function.
4. The wind farm wake prediction method according to claim 1, characterized in that, The basic parameters of the wind turbine include rotor diameter, hub height, thrust coefficient curve, and power curve.
5. The wind farm wake prediction method according to claim 1, characterized in that, The wake expansion rate parameter is determined as follows: the power recorded by SCADA data is compared with the power calculated by the wake model, and the root mean square error (RMSE) is used for evaluation.
6. A wind farm wake prediction system, characterized in that, The system for executing the wind farm wake prediction method of claim 1 includes the following modules: Wake segmentation module: used to divide the wake region of a wind turbine into a near-wake region and a far-wake region based on the distance of the wake section from the wind turbine. Wake Model Construction Module: Used to construct wake models for the near-wake and far-wake regions of a wind turbine, respectively; Full-field wake model construction module: used to introduce wake superposition methods to construct a full-field wake model; Wake Expansion Rate Calculation Module: Used to calculate the wake expansion rate in the near-wake region. and wake expansion rate in the far wake region As undetermined parameters, for each recording moment of SCADA data, based on the wind farm layout, basic wind turbine parameters, wind speed, wind direction, and actual output power of each turbine at that recording moment, the output power of each turbine is calculated using the aforementioned full-field wake model. The optimization objective is to minimize the error between the actual output power and the calculated output power. A genetic algorithm is then used to... and Optimize to obtain the optimal value corresponding to this recorded time. and And the optimal value corresponding to each recording time. and Composing a wake feature set; Machine learning module: Used to train machine learning models with wind farm inflow data as input and wake parameter data as output, establish the relationship between inflow information and wake characteristics, and construct an adaptive wind farm engineering wake model; The calculation output module: Based on the input flow information, it predicts the wake parameters through a trained machine learning model, combines them with the established analytical wake model, and performs wake-related calculations within the wind farm.
7. A wind farm wake prediction device, characterized in that, The device is used to perform a wind farm wake prediction method as described in any one of claims 1-5.
8. A storage medium, characterized in that, The storage medium stores a wind farm wake prediction method as described in any one of claims 1-5.
Citation Information
Patent Citations
Wind turbine generator wake flow regional modeling and rapid analysis method
CN112949217A
Establishment method of three-dimensional full wake flow model of horizontal axis wind turbine
CN114707437A
