A data-driven-based large wind farm wake rapid calculation method and system

Abstract

The application discloses a large wind farm wake rapid calculation method and system based on data driving, and creatively studies wind farm wake modeling work through a technical framework of "group division-offline modeling-online calculation". According to historical wind direction and wind speed information, the wake propagation path is mined, and the mean square loss error between the wind speed obtained by model calculation and the actual wind speed is reduced as the target. The wolf swarm algorithm is used to optimize and solve the attenuation parameters of the analytical wake model. The corrected field-level wake model is obtained through offline training. Then, combined with the wind speed and direction prediction results, the unit operation state and the influence of the unit wake, the deployment and application of the online rapid calculation of the wake are realized. The efficiency and precision of real-time wake modeling calculation are greatly improved, and high-reliable data support is provided for subsequent online analysis tasks such as wind turbine load online calculation, output performance evaluation and optimization control strategy of the wind farm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind farm wake calculation, in particular to a large wind farm wake fast calculation method and system based on data driving. BACKGROUND

[0002] Energy is an important material basis for promoting human survival and development. Due to the limitation of natural resources, the reserves of traditional fossil energy are decreasing year by year, and a large amount of environmental ecological problems are generated in the process of energy conversion. In order to reduce the dependence of social production on traditional fossil energy, a large amount of renewable energy has been developed and applied. Among them, wind energy is considered as one of the most potential and technically and economically renewable energy, which has been focused by governments of various countries, and is regarded as an important means to effectively alleviate energy shortage and cope with environmental problems. China has vast territory and long coastline, and the reserves of wind energy resources are rich. The superior natural conditions create the possibility for large-scale development and utilization of wind energy resources. China is currently the world's largest wind power country, and wind power accounts for the first proportion in non-water new energy in China.

[0003] Wake effect refers to the fact that a wind turbine obtains energy from the wind while forming a wake area with reduced wind speed downstream. If there is a wind turbine in the wake area downstream, the input wind speed of the downstream wind turbine will be lower than that of the upstream wind turbine. The mechanical structure of wind turbines and the position arrangement of wind turbines often bring great influence of wake effect to wind farms. Especially for offshore wind farms with relatively simple operating environment, the wind turbines are densely arranged, the wakes of each wind turbine interfere with each other, and the wake effect is seriously superimposed in the wind farm, which reduces the power generation efficiency of the whole wind farm, causes the power generation capacity to be damaged, and ultimately affects the economic benefits of the wind farm. In the operation of the wind farm, theoretically, the wake optimization management can be carried out according to the real-time operation condition through the wind farm control system to reduce the power generation loss caused by the wake effect. However, the existing wind farm wake calculation device cannot accurately calculate the wake effect in the wind farm in real time, and the complexity of the wake effect leads to a long time consumption of the existing wind farm wake calculation, which is not conducive to providing reliable reference basis for subsequent online analysis of wind farm operation.

[0004] In view of the problem of power loss caused by the wake effect of wind farm, scholars have proposed solutions to accurately model the wake. The current mainstream methods can be summarized into two categories, namely analytical wake model method and numerical simulation method. Due to low cost, fast calculation speed and high efficiency, analytical wake model is still widely used in the prediction of wake distribution in today's engineering practice. Among them, the most pioneering model is Jensen wake model, which considers the top hat distribution of wake velocity decay. However, the assumption of "top hat" wake velocity decay is not suitable for practical application, so Bastankhah et al. established a Gaussian wake model by assuming self-similar Gaussian profile to correct the defects of wake and improve the accuracy of wake prediction. Later, Larsen et al. proposed Larsen wake model based on self-similarity theory by assuming that the wake is a result of mean flow disturbance, and Frandsen et al. proposed Frandsen wake model based on the momentum conservation law of the flow inside and around the wind turbine rotor. These classic analytical wake models rely heavily on empirical constants, cannot well consider the influence of turbulence intensity on the wake and the coupling relationship between wind turbines, and cannot be changed according to the environmental wind speed conditions of specific wind farms, resulting in a large error between the calculation accuracy and the actual situation, which cannot meet the engineering demand of online calculation of wake in wind farm operation. While CFD numerical simulation such as using large eddy simulation (LES) or Reynolds-averaged Navier-Stokes equation (RANS) can significantly improve the accuracy of wake prediction, but the calculation cost is too expensive and the calculation amount is too large, resulting in very low calculation efficiency, which is currently more prevalent in advanced theoretical research and still cannot be widely applied in general large-scale wind farms. SUMMARY

[0005] The existing wind farm wake calculation method cannot achieve real-time and accurate calculation of the wake effect in the wind farm, and the complexity of the wake effect leads to long time consumption of the existing wind farm wake calculation, especially the cluster deployment of wind turbines, which leads to serious wake effect in the wind farm, not only has a great adverse effect on the working load and output performance of the downstream wind turbine, but also makes the coupling relationship of the wind turbine incoming flow more complex, and the existing wake model is difficult to meet the requirements of efficiency and accuracy in engineering application;In view of the above problems, the present application provides a kind of large wind farm wake fast calculation method and system based on data driving, creatively proposes the technical framework of "group division-offline modeling-online calculation" to carry out the research work of real-time fast modeling of wind farm wake, according to the historical wind direction and wind speed information, the wake propagation path is mined, and the mean square loss error between the wind speed calculated by the model and the actual wind speed is taken as the target, the wolf swarm algorithm is used to optimize and solve the analytical wake model, the wind farm level model is trained and established, then combined with the wind speed and wind direction prediction results, the operating state of the unit and the influence of the unit wake, the large wind farm wake fast calculation technology and system based on data driving are researched, which greatly improves the efficiency and accuracy of real-time wake modeling calculation, and provides reliable data service for subsequent online analysis of wind farm operation.

[0006] The technical scheme adopted by the present application to solve its technical problems is: a large wind farm wake fast calculation method based on data driving, which comprises the following steps:

[0007] (1) obtaining the basic parameter information of wind turbine and the arrangement information of wind turbine in wind farm, obtaining the output data recorded by SCADA system in a period of time;

[0008] (2) using the arrangement information of wind turbine in wind farm and the environmental wind direction information in historical SCADA data obtained in step (1), processing the average incoming flow wind direction in a period of time into positive wind direction and oblique wind direction;

[0009] (3) grouping the wind direction information processed in step (2), setting the first row of wind turbine through which the wind flows in the wind farm as the boundary windward unit, inferring the wake propagation path according to the wind direction, and dividing the units under the same wake propagation path into the same group;

[0010] (4) using the basic parameter information of wind turbine obtained in step (1), based on the principle of mass conservation and momentum conservation, and setting the initial value of each wake attenuation coefficient in Park model, modeling the wake effect of single wind turbine, and based on the wake propagation path information obtained in step (3), based on the linear superposition principle of wake loss, modeling the wake effect of each group, obtaining the whole field level wake model;

[0011] (5) According to the historical wind speed information of each wind turbine extracted from the SCADA data in step (1) and the wind speed information output by the full-field wake model in step (4), the mean square loss error of the model calculated wind speed and the actually measured wind speed is obtained, denoted as MSELoss;

[0012] (6) Guided by the minimization of MSELoss obtained in step (5), the optimal value of each wake attenuation coefficient is solved, and the full-field wake model of the wind farm in this historical time is corrected;

[0013] (7) The corrected wake model obtained in step (6) is deployed online, and according to the wind speed and wind direction predicted by the real-time environmental information, the machine group on the same wake propagation path is divided, the wake speed of each wind turbine in each machine group is calculated quickly, and then reliable data service is provided for the subsequent online analysis of wind farm operation.

[0014] Further, the wind turbine basic parameter information of step (1) specifically includes wind turbine cut-in wind speed, wind turbine cut-out wind speed, blade number, impeller diameter, impeller swept area, wind wheel hub height, wind turbine rated power, wind turbine power curve, wind turbine thrust coefficient, and wind turbine axial induction factor; the wind turbine arrangement information in the wind farm specifically includes the number of wind turbines in the wind farm, the longitude and latitude parameters and the altitude parameters of each wind turbine.

[0015] Further, the forward wind direction and the oblique wind direction in step (2) are specifically: the forward wind direction refers to the average wind direction perpendicular to the row or column of the wind farm, denoted as d1=[positive east E, positive west W, positive south S, positive north N], and the oblique wind direction refers to the average wind direction with a 45° angle with the row or column of the wind farm, denoted as d2=[northeast EN, southeast ES, northwest WN, southwest WS]; if the average input wind direction does not belong to the forward wind direction or the oblique wind direction, it is mapped to the forward wind direction or the oblique wind direction according to the Bayesian criterion with the minimum error rate:

[0016]

[0017] Wherein, m represents the current average input wind direction, d1 is the forward wind direction, d2 is the oblique wind direction, P(d1) is the probability of the forward wind direction in the historical information, P(m|d1) is the probability that the current wind direction is the forward wind direction, P(d2) is the probability of the oblique wind direction in the historical information, P(m|d2) is the probability that the current wind direction is the oblique wind direction, l(m) is the likelihood ratio, and P(d2) / P(d1) is the likelihood ratio threshold.

[0018] Further, the step (4) is based on the Park model principle to model the wake effect of a single wind turbine, for the wake effect generated by a single wind turbine, assuming that the wind turbine i is in the effective area of the wake effect formed by the wind turbine j, generating a single machine vertical section wake velocity distribution, expressed as:

[0019]

[0020] wherein V ∞ is the ambient wind speed, V i is the inflow wind speed of the wind turbine i, C T is the thrust coefficient of the wind turbine, D r is the impeller diameter of the wind turbine, x is the distance between the wind turbine i and the wind turbine j, and the predicted farthest distance is 9 times the impeller diameter D r , k is the wake expansion attenuation coefficient, expressed as:

[0021]

[0022] wherein U i (x, D w , a i ) is the size of the wind energy available to the wind turbine i, a i is the axial induction factor of the wind turbine i, a j is the axial induction factor of the wind turbine j, D w is the wake radius of the wind turbine i in the wake area of the wind turbine j, and δ is the wind energy loss factor, obtaining the size of the wind energy available to the downstream wind turbine i expressed as:

[0023] U i (x, D w , a i ) = V ∞ (1-δU j (x, D w , a j ))

[0024] wherein U j (x, D w , a j ) is the size of the wind energy available to the wind turbine j.

[0025] Further, the specific method for calculating the area of the wake superposition area in the step (4) is as follows: the shadow part of the wake superposition = (sector O1AB-triangle O1AB) + (sector O2AB-triangle O2AB), wherein O1 is the rotation center point of the wind turbine i, O2 is the center point of the wake area of the wind turbine j, and A and B are two different intersection points of the wind wheel swept area of the wind turbine i and the wake area of the wind turbine j, respectively;

[0026] Wherein, the two triangles of triangle O1AO2 and triangle O1BO2 form a quadrilateral, which can be obtained by twice the area of triangle O1AO2, and the specific derivation is as follows:

[0027]

[0028] ∠AO1B = 2∠AO1O2

[0029]

[0030]

[0031]

[0032] Wherein, S O1AB is the area of the sector O1AB, r1 is the impeller radius of wind turbine i, r2 is the wake radius of wind turbine j, and d is the distance between O1 and O2; Similarly, the area of sector O2AB is obtained:

[0033]

[0034] S O2AB = ∠AO2O1*r2 2

[0035] Wherein the area of the quadrilateral is represented as:

[0036]

[0037] Therefore, the area of the wake superposition region can be obtained as:

[0038]

[0039] Wherein, represents the area of the wake superposition region of wind turbine i in the wake area of wind turbine j.

[0040] Further, the linear superposition principle of wake loss in step (4) is represented by the formula:

[0041]

[0042] Wherein, V j is the incoming wind speed of wind turbine j, V ij is the incoming wind speed of wind turbine i affected by the wake area of wind turbine j, and N i is the number of wind turbines in the cluster where wind turbine i is located.

[0043] Further, the full-field wake model in step (4) is represented by the formula:

[0044]

[0045] wherein θ w is the angle between the wind wheel rotation plane and the hub height plane of the wind turbine i, is the wind energy of the wind turbine i affected by the wake of the wind turbines in the wind farm.

[0046] Further, the mean square loss error of the model wind speed obtained in step (5) is expressed by the formula:

[0047]

[0048] wherein N is the number of wind turbines in the wind farm, is the historical wind speed of the wind turbine i extracted from the SCADA data, V i is the wind speed of the wind turbine i calculated by the full-field wake model.

[0049] Further, step (6) is specifically: using the wolf swarm algorithm to solve the optimal value of each wake attenuation coefficient, obtaining:

[0050]

[0051] The judgment condition that the MSELoss reaches the optimal solution target or exceeds the maximum iteration number is reached, and finally the corrected wake model is obtained.

[0052] The application discloses a large wind farm wake fast calculation system based on data driving, comprising:

[0053] A wind turbine position parameter information acquisition module acquires the arrangement information of wind turbines in the wind farm, including the number of wind turbines in the wind farm, the longitude and latitude parameters and the altitude parameters of each wind turbine, and converts the longitude and latitude parameters and the altitude parameter information into matrix position information of a three-dimensional coordinate system.

[0054] A boundary windward turbine acquisition module divides the distribution of the wind farm under different wind directions according to the acquired position information.

[0055] A wake propagation path prediction module predicts the wake propagation path of the wind farm based on the minimum error rate Bayesian criterion according to the environmental wind direction information in the historical SCADA data.

[0056] A wake model offline training module models the wake effect of a single wind turbine, models the wake effect of each wind group, obtains a full-field wind farm wake model, and optimally solves the initial value of each wake attenuation coefficient based on the historical wind speed information, and finally obtains a corrected wake model.

[0057] The wake model online application module divides the same wake propagation path of the machine group according to the input predicted wind speed and predicted wind direction information, and realizes the rapid calculation of the wake speed of each wind turbine in each machine group.

[0058] The upstream big data and wind farm energy management platform provides wind turbine basic parameter information, wind turbine arrangement information in the wind farm, and output data recorded by the SCADA system; and stores relevant chart data obtained after the calculation results are visualized by the wake model online application module;

[0059] The upstream wind speed prediction module performs real-time short-term prediction on the environmental wind speed information.

[0060] The downstream wind farm optimization scheduling platform performs subsequent online analysis tasks, including wind turbine load online calculation, output performance evaluation, and online solution of optimization control strategy.

[0061] The input ends of the wake model offline training module and the wake propagation path prediction module of the system are connected with the upstream big data and wind farm energy management platform; the output ends of the wake model offline training module and the wind speed prediction module are connected with the input end of the wake model online application module; the output end of the wind farm wind turbine position parameter information acquisition module is connected with the input end of the boundary windward machine group acquisition module; the output end of the boundary windward machine group acquisition module is connected with the input end of the wake propagation prediction module; the output end of the wake propagation path prediction module is connected with the input end of the wake model offline training module; and the wake model online application module is connected with the downstream wind farm optimization scheduling platform.

[0062] The beneficial effects of the present application are: the present application creatively proposes a "machine group division-offline modeling-online calculation" technical framework to perform real-time and rapid modeling of the wind farm wake, mines the wake propagation path according to historical wind direction and wind speed information, uses the wolf swarm algorithm to optimize and solve the decay parameters of the analytical wake model, trains and obtains a corrected field-level wake model, and then combines the wind speed and wind direction prediction results, the operating state of the machine group, and the influence of the machine group wake to realize the deployment and application of online rapid wake calculation, greatly improves the efficiency and accuracy of real-time wake modeling calculation, provides high-reliable data for subsequent online analysis tasks such as wind turbine load online calculation, output performance evaluation, and optimization control strategy of the wind farm, and further improves the power generation capacity of the whole wind farm. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 It is an implementation schematic diagram of the wake rapid calculation system of the present application.

[0064] Figure 2 Flow chart for model parameter correction using wolf pack algorithm of the present application;

[0065] Figure 3 Principle diagram for fleet division of the present application;

[0066] Figure 4 Principle diagram for Park model used in the present application;

[0067] Figure 5 Schematic diagram of wake superposition of the present application;

[0068] Figure 6 Schematic diagram of wind farm full-field wake velocity distribution at hub height plane in the embodiment;

[0069] Figure 7 Schematic diagram of three-dimensional visualization of wind farm full-field wake velocity distribution in the embodiment. DETAILED DESCRIPTION

[0070] The present application will be further described in detail below in combination with the drawings and specific embodiments.

[0071] As shown in the drawings, Figure 1 the data-driven large-scale wind farm wake rapid calculation method proposed in the embodiment of the present application includes the following steps:

[0072] (1) Obtain basic parameter information of wind turbines, including wind turbine cut-in wind speed, wind turbine cut-out wind speed, number of blades, impeller diameter, impeller swept area, wind wheel hub height, wind turbine rated power, wind turbine power curve, wind turbine thrust coefficient, wind turbine axial induction factor; wind turbine arrangement information in the wind farm, including the number of wind turbines in the wind farm, the longitude and latitude parameters and the altitude parameters of each wind turbine, and the output data recorded by the SCADA system in a period of time;

[0073] (2) Use the wind turbine arrangement information in the wind farm and the environmental wind direction information in the historical SCADA data obtained in step (1) to process the average inflow wind direction in a period of time as a forward wind direction and an oblique wind direction; wherein the forward wind direction refers to the average wind direction perpendicular to the row or column of the wind farm, denoted as d1=[East E, West W, South S, North N], the oblique wind direction refers to the average wind direction having a 45° angle with the row or column of the wind farm; denoted as d2=[North-East EN, South-East ES, North-West WN, South-West WS], if the average input wind direction does not belong to the forward wind direction or the oblique wind direction, it is mapped to the forward wind direction or the oblique wind direction according to the Bayesian rule with the minimum error rate:

[0074]

[0075] Wherein, m represents the current average input wind direction, d1 is the positive wind direction, d2 is the oblique wind direction, P(d1) is the probability of the positive wind direction in the historical information, P(m|d1) is the probability of the current wind direction being the positive wind direction, P(d2) is the probability of the oblique wind direction in the historical information, P(m|d2) is the probability of the current wind direction being the oblique wind direction, l(m) is the likelihood ratio, P(d2) / P(d1) is the likelihood ratio threshold.

[0076] (3) The wind direction information obtained by processing in step (2) is divided into groups, the first row of wind turbines through which the wind flows is set as the boundary wind turbine group, the wake propagation path is inferred according to the wind direction, and the groups under the same wake propagation path are divided into the same group; Figure 3

[0077] (4) The basic parameter information of the wind turbine group obtained in step (1) is used, the principles of mass conservation and momentum conservation are used, and the initial values of the wake attenuation coefficients in the model are set, such as Figure 4 The wake effect of a single wind turbine is modeled based on the Park model principle, and for the wake effect generated by a single wind turbine, it is assumed that the wind turbine i is in the effective area of the wake effect formed by the wind turbine j, the wake velocity distribution of the single machine vertical section is generated, which is expressed as:

[0078]

[0079] Wherein, V ∞ is the ambient wind speed, V i is the inflow wind speed of the wind turbine i, C T is the thrust coefficient of the wind turbine, D r is the impeller diameter of the wind turbine, x is the distance between the wind turbine i and the wind turbine j, and the predicted farthest distance is 9 times the impeller diameter D r , k is the wake expansion attenuation coefficient, which is expressed as:

[0080]

[0081] Wherein, U i (x,D w ,a i ) is the size of the wind energy available to the wind turbine i, a i is the axial induction factor of the wind turbine i, a j is the axial induction factor of the wind turbine j, D w is the wake radius of the wind turbine i in the wake area of the wind turbine j, and δ is the wind energy loss factor. The size of the wind energy available to the downstream wind turbine i is expressed as:

[0082] U i (x,D w ,a i ​) = V ∞ (1 - δU j (x, D w , a j ))

[0083] wherein, U j (x, D w , a j ) is the wind energy available to wind turbine j; thereafter, the area of the wake superposition region is calculated, such as Figure 5 characterized in that the specific method of calculating the area of the wake superposition region in step (4) is as follows, the shadow part of the wake superposition = (sector O1AB-triangle O1AB) + (sector O2AB-triangle O2AB), wherein O1 is the center point of rotation of wind turbine i, O2 is the center point of the wake region of wind turbine j, A and B are two different intersection points of the swept area of the wind wheel of wind turbine i and the wake region of wind turbine j, respectively;

[0084] wherein, the two triangles O1AO2 and O1BO2 form a quadrilateral, which can be obtained by twice the area of triangle O1AO2, and the specific derivation is as follows:

[0085]

[0086] ∠AO1B = 2∠AO1O2

[0087]

[0088]

[0089]

[0090] wherein, S O1AB is the area of sector O1AB, r1 is the impeller radius of wind turbine i, r2 is the wake radius of the wake region of wind turbine j, and d is the distance between O1 and O2; similarly, the area of sector O2AB is obtained:

[0091]

[0092] S O2AB = ∠AO2O1*r2 2

[0093] wherein the area of the quadrilateral is represented as:

[0094]

[0095] Therefore, the area of the wake superposition region is obtained as:

[0096]

[0097] wherein, represents the area of the wake superposition region of wind turbine i in the wake zone of wind turbine j;

[0098] Further based on the wake propagation path information obtained in step (3), based on the linear superposition principle of wake loss:

[0099]

[0100] wherein, V j is the inflow wind speed of wind turbine j, V ij is the inflow wind speed of wind turbine i affected by the wake zone of wind turbine j, N i is the number of wind turbines in the wind turbine group where wind turbine i is located;

[0101] Modeling the wake effect of each wind turbine group, the full-field wake model is expressed by the formula:

[0102]

[0103] wherein, θ w is the angle between the wind wheel rotation plane of wind turbine i and the hub height plane of the wind turbine, is the wind energy of wind turbine i affected by the wake of the wind turbines in the wind turbine group.

[0104] (5) According to the historical wind speed information of each wind turbine extracted from the SCADA data in step (1) and the wind speed information output by the model in step (4), the mean square loss error sum of the model calculated wind speed and the actual measured wind speed is obtained:

[0105]

[0106] Let the mean square loss error sum be MSELoss.

[0107] (6) Guided by the minimization of MSELoss obtained in step (5), such as Figure 2 , the wolf swarm algorithm is used to solve the optimal value of each wake attenuation parameter variable, and the following is obtained:

[0108]

[0109] Through the optimal solving goal of MSELoss or the judgment condition of exceeding the maximum iteration number, the wake model after the correction of multiple parameter variables is finally obtained.

[0110] (7) Deploy the modified wake model obtained in step (6) online. Based on the predicted wind speed and predicted wind direction output by the wind speed prediction module, divide the turbine groups with the same wake propagation path, realize the rapid calculation of the wake speed of each wind turbine in each turbine group, and transmit the results directly to the wind power optimization scheduling platform to provide reliable data services for the online analysis of subsequent wind farm operation. At the same time, visualize the results and transmit them back to the big data and wind farm energy management platform in the form of charts, etc.

[0111] like Figure 1 As shown, the present invention proposes a data-driven rapid wake calculation system for large-scale wind farms, implemented using the above-mentioned calculation method, comprising:

[0112] The wind farm wind turbine location parameter information acquisition module acquires the wind turbine layout information within the wind farm, including the number of wind turbines in the wind farm, the latitude and longitude parameters and altitude parameters of each wind turbine, and converts the latitude and longitude parameters and altitude parameters into matrix location information in a three-dimensional coordinate system.

[0113] The boundary windward turbine acquisition module, based on the acquired location information, classifies the distribution of turbine clusters under different wind directions within the wind farm;

[0114] The wake propagation path prediction module predicts the wake propagation path of the wind farm based on the environmental wind direction information in historical SCADA data and the Bayesian criterion with minimum error rate.

[0115] The wake model offline training module models the wake effect of a single wind turbine and the wake effect of various turbine groups to obtain a field-level wake model. Based on historical wind speed information, the initial values ​​of each wake attenuation coefficient are optimized to obtain the corrected wake model.

[0116] The wake model online application module, based on the input predicted wind speed and predicted wind direction information, divides the wind turbine groups with the same wake propagation path, and realizes the rapid calculation of the wake velocity of each wind turbine in each group.

[0117] The upstream big data and wind farm energy management platform provides basic parameter information of wind turbines, wind turbine layout information within the wind farm, and output data recorded by the SCADA system; it also stores relevant chart data obtained after the wake model online application module visualizes the calculation results.

[0118] The upstream wind speed prediction module provides real-time short-term predictions of environmental wind speed information.

[0119] The downstream wind farm optimization scheduling platform performs subsequent online analysis tasks, including online calculation of wind turbine loads, output performance evaluation, and online solution of optimization control strategies.

[0120] The input end of the wake model offline training module and the wake propagation path prediction module of the system is connected with the upstream big data and the wind farm energy management platform; the output end of the wake model offline training module and the wind speed prediction module is connected with the input end of the wake model online application module; the output end of the wind farm wind turbine position parameter information acquisition module is connected with the input end of the boundary windward unit acquisition module; the output end of the boundary windward unit acquisition module is connected with the input end of the wake propagation prediction module; the output end of the wake propagation path prediction module is connected with the input end of the wake model offline training module; and the wake model online application module is connected with the downstream wind farm optimization scheduling platform.

[0121] The application will be further explained and described below with specific embodiments:

[0122] In this embodiment, the effectiveness of the method provided by the application is verified based on MATLAB simulation software. The wind speed and wind direction data used come from a certain offshore wind farm in China, the simulation time is 600s, and the given basic parameter setting value range is shown in the following table:

[0123] Basic parameters Numerical range Air density 1.225 kg / m 3 ]] Number of wind turbines in the wind farm 72 Cut-in wind speed 3 m / s Cut-out wind speed 25 m / s Number of blades 3 Impeller diameter 150m Impeller swept area 17671m 2 ]] Rotor hub height 95m Rated power 6 MW

[0124] According to the "group division-offline modeling-online calculation" technical framework proposed by the application, the research work of real-time and rapid modeling of wind farm wake is carried out. The wake propagation path is mined according to the historical wind direction and wind speed information, and the mean square loss error between the wind speed calculated by the model and the actual wind speed is reduced as the target. The wolf swarm algorithm is used to optimize and solve the analytical wake model. After 100 iterations, the mean square loss error MSELoss is reduced from the initial value 0.218 to 0.123, and the decay parameter variables k and δ are optimized from the initial values 0.08 and 0.12 to 0.142 and 0.197 respectively, so as to obtain the corrected wind farm field model. Then, the wind speed and wind direction prediction results are combined to perform online rapid calculation of the wake, and the visualization of the calculation results are shown in Figure 6 and Figure 7 , wherein Figure 6 is a schematic diagram of the speed distribution of the whole field wake of the wind farm at the hub height plane, Figure 7 is a three-dimensional visualization schematic diagram of the whole field wake speed distribution of the wind farm.

[0125] The wake calculation results can provide high-reliable data support for subsequent online analysis tasks such as wind farm wind turbine load online calculation, output performance evaluation and optimization control strategy.

[0126] The above examples are used to explain the present application, but not to limit the present application, any modification and change made to the present application within the spirit and protection scope of the claims, fall into the protection scope of the present application.

Claims

1. A data-driven method for rapid wake calculation in large-scale wind farms, characterized in that, Includes the following steps: (1) Obtain basic parameter information of wind turbine units and wind turbine unit layout information in wind farm, and obtain the output data recorded by the SCADA system of each unit within a historical period. (2) Using the wind turbine layout information in the wind farm and the environmental wind direction information in the historical SCADA data obtained in step (1), the average inflow wind direction over a historical period is processed into a forward wind direction and an oblique wind direction. (3) The wind direction information obtained in step (2) is divided into groups. The first row of wind turbines through which the wind flows is set as the boundary windward units. The wake propagation path is inferred based on the wind direction, and the units under the same wake propagation path are divided into the same group. (4) Using the basic parameter information of the wind turbine obtained in step (1), based on the principle of mass conservation and momentum conservation, and setting the initial values ​​of each wake attenuation coefficient in the Park model, the wake effect of a single wind turbine is modeled. Then, based on the wake propagation path information obtained in step (3), and based on the principle of linear superposition of wake loss, the wake effect of each group of turbines is modeled to obtain the full field-level wake model. (5) Based on the historical wind speed information of each wind turbine extracted from the SCADA data in step (1) and the wind speed information output by the field-level wake model in step (4), the mean square loss error between the wind speed calculated by the model and the wind speed actually measured is obtained and denoted as MSELoss. (6) Guided by minimizing MSELoss obtained in step (5), the optimal values ​​of each wake attenuation coefficient are solved to correct the wake model of the entire wind farm during this historical period. (7) Deploy the modified wake model obtained in step (6) online. Based on the wind speed and wind direction predicted by real-time environmental information, divide the wind turbine groups with the same wake propagation path, realize the rapid calculation of the wake velocity of each wind turbine in each group, and thus provide reliable data services for the online analysis of subsequent wind farm operation.

2. The data-driven rapid wake calculation method for large-scale wind farms according to claim 1, characterized in that, The basic parameters of the wind turbine mentioned in step (1) include the wind turbine cut-in wind speed, wind turbine cut-out wind speed, number of blades, impeller diameter, impeller swept area, impeller hub height, wind turbine rated power, wind turbine power curve, wind turbine thrust coefficient, and wind turbine axial induction factor; the wind turbine layout information in the wind farm includes the number of wind turbines in the wind farm, the latitude and longitude parameters and altitude parameters of each wind turbine.

3. The data-driven rapid calculation method for wake of large-scale wind farms according to claim 1, characterized in that, The forward and oblique wind directions mentioned in step (2) refer to the average wind direction being perpendicular to the row or column of the wind farm, denoted as d1 = [East E, West W, South S, North N]. The oblique wind direction refers to the average wind direction having a 45° angle with the row or column of the wind farm, denoted as d2 = [Northeast EN, Southeast ES, Northwest WN, Southwest WS]. If the average input wind direction is neither a forward nor an oblique wind direction, it is mapped to a forward or oblique wind direction according to the Bayesian criterion of minimum error rate. Where m represents the current average input wind direction, d1 is the forward wind direction, d2 is the diagonal wind direction, P(d1) is the probability of a forward wind direction in historical information, P(m|d1) is the probability that the current wind direction is a forward wind direction, P(d2) is the probability of a diagonal wind direction in historical information, P(m|d2) is the probability that the current wind direction is a diagonal wind direction, l(m) is the likelihood ratio, and P(d2) / P(d1) is the likelihood ratio threshold.

4. The data-driven rapid calculation method for wake of large-scale wind farms according to claim 1, characterized in that, Step (4) models the wake effect of a single wind turbine based on the Park model principle. For the wake effect generated by a single wind turbine, assuming that wind turbine i is within the effective region of the wake effect formed by wind turbine j, the wake velocity distribution of the vertical profile of a single turbine is generated, as follows: Among them, V ∞ It is the ambient wind speed, V i C is the inflow wind speed of wind turbine i. T It is the thrust coefficient of the wind turbine, D r Let be the rotor diameter of the wind turbine, and x be the distance between wind turbine i and wind turbine j, assuming the predicted maximum distance is 9 times the rotor diameter D. r k is the wake expansion attenuation coefficient, expressed as: Among them, U i (x,D w ,a i ) represents the amount of wind energy available to wind turbine i, a i It is the axial induction factor of wind turbine i, a j It is the axial induction factor of wind turbine j, D w Let δ be the wake radius of wind turbine i within the wake region of wind turbine j, and δ be the wind energy loss factor. The amount of wind energy available to downstream wind turbine i can be expressed as: U i (x,D w ,and i )=V ∞ (1-δU j (x,D w ,and j )) Among them, U j (x,D w ,a j ) represents the amount of wind energy that the wind turbine j can utilize.

5. The data-driven rapid calculation method for wake of large-scale wind farms according to claim 1, characterized in that, The specific method for calculating the area of ​​the wake superposition region in step (4) is as follows: the shaded part of the wake superposition = (sector O1AB - triangle O1AB) + (sector O2AB - triangle O2AB), where O1 is the rotation center point of wind turbine i, O2 is the center point of the wake region of wind turbine j, and A and B are two different intersection points of the wind turbine i's rotor sweeping region and the wake region of wind turbine j. The two triangles O1AO2 and O1BO2 form a quadrilateral, which can be obtained by doubling the area of ​​triangle O1AO2. The specific derivation is as follows: ∠AO1B=2∠AO1O2 Among them, S O1AB Let r1 be the area of ​​sector O1AB, r2 be the radius of the impeller of wind turbine i, r2 be the radius of the wake region of wind turbine j, and d be the distance between O1 and O2; similarly, the area of ​​sector O2AB can be obtained: S O2AB =∠AO2O1*r2 2 The area of ​​the quadrilateral is expressed as: Therefore, the area of ​​the wake superposition region can be obtained as: in, This represents the area of ​​the superimposed wake region of wind turbine i within the wake region of wind turbine j.

6. The data-driven rapid calculation method for wake of large-scale wind farms according to claim 4, characterized in that, The principle of linear superposition of wake loss in step (4) can be expressed by the following formula: Among them, V j The magnitude of the inflow wind speed of wind turbine j, V ij N is the magnitude of the inflow wind speed of wind turbine i affected by the wake region of wind turbine j. i It is the number of wind turbines in the turbine group where wind turbine i is located.

7. The data-driven rapid calculation method for wake of large-scale wind farms according to claim 6, characterized in that, The full-field wake model in step (4) is expressed by the formula: Where, θ w It is the angle between the rotor plane of wind turbine i and the height plane of the wind turbine hub. Let i be the amount of wind energy affected by the wake of the wind turbines in the same group.

8. The data-driven rapid calculation method for wake of large-scale wind farms according to claim 1, characterized in that, The mean squared loss error of the model wind speed obtained in step (5) is expressed by the following formula: Where N is the number of wind turbines in the wind farm. The historical wind speed of wind turbine i, V, is extracted from SCADA data. i The wind speed of wind turbine i is calculated using the full-field wake model.

9. The data-driven rapid calculation method for wake of large-scale wind farms according to claim 8, characterized in that, In step (6), the wolf pack algorithm is used to solve for the optimal value of each wake attenuation coefficient, and the following results are obtained: By using MSELoss to achieve the optimal solution objective or exceed the maximum number of iterations, the corrected wake model is finally obtained.

10. A data-driven rapid wake calculation system for large-scale wind farms, implementing the method as described in any one of claims 1-9, characterized in that, include: The wind farm wind turbine location parameter information acquisition module acquires the wind turbine layout information within the wind farm, including the number of wind turbines in the wind farm, the latitude and longitude parameters and altitude parameters of each wind turbine, and converts the latitude and longitude parameters and altitude parameters into matrix location information in a three-dimensional coordinate system. The boundary windward turbine acquisition module, based on the acquired location information, classifies the distribution of turbine clusters under different wind directions within the wind farm; The wake propagation path prediction module predicts the wake propagation path of the wind farm based on the environmental wind direction information in historical SCADA data and the Bayesian criterion with minimum error rate. The wake model offline training module models the wake effect of a single wind turbine and the wake effect of various turbine groups to obtain a field-level wake model for the entire field. Based on historical wind speed information, the initial values ​​of each wake attenuation coefficient are optimized, and the corrected wake model is finally obtained. The wake model online application module, based on the input predicted wind speed and predicted wind direction information, divides the wind turbine groups with the same wake propagation path, and realizes the rapid calculation of the wake velocity of each wind turbine in each group. The upstream big data and wind farm energy management platform provides basic parameter information of wind turbines, wind turbine layout information within the wind farm, and output data recorded by the SCADA system; it also stores relevant chart data obtained after the wake model online application module visualizes the calculation results. The upstream wind speed prediction module provides real-time short-term predictions of environmental wind speed information. The downstream wind farm optimization scheduling platform performs subsequent online analysis tasks, including online calculation of wind turbine loads, output performance evaluation, and online solution of optimization control strategies. The input terminals of the system's wake model offline training module and wake propagation path prediction module are simultaneously connected to the upstream big data and wind farm energy management platform; the output terminals of the wake model offline training module and wind speed prediction module are simultaneously connected to the input terminal of the wake model online application module; the output terminal of the wind farm wind turbine location parameter information acquisition module is connected to the input terminal of the boundary wind turbine acquisition module; and the output terminal of the boundary wind turbine acquisition module is connected to the input terminal of the wake propagation prediction module. The output of the wake propagation path prediction module is connected to the input of the wake model offline training module; The wake model online application module is connected to the downstream wind farm optimization and scheduling platform.

Images