A cfd database optimization method for virtual laser radar wind finding system

Abstract

The application discloses a CFD database optimization method for a virtual laser radar wind measurement system, and belongs to the technical field of wind power generation. ICEM CFD is used to perform geometric modeling and grid division on a target wind farm, so as to obtain a model grid of an unstructured grid, and six boundary layer grids are arranged in the near-ground area of the model grid; an improved actuator disc model is used to build an improved actuator disc of multiple wind turbines on the model grid, so as to obtain a target model grid; a first boundary condition of CFD simulation and a second boundary condition of the improved actuator disc model are set, and are used for CFD simulation on the target model grid until a stable state of a flow field is reached, so as to form a wind farm simulation CFD database of full wind direction and full wind speed containing multiple groups of working condition data. Compared with a traditional CFD database in an existing wind measurement system, the application can make up for the lack of a physical mechanism of an analytical model, correct the time delay effect of a flow field, and reduce the calculation cost and improve the database precision.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology, specifically relating to a CFD database optimization method for a virtual lidar wind measurement system. Background Technology

[0002] Existing wind speed prediction technologies have limitations. The intermittent and unpredictable nature of wind speeds poses challenges to wind power generation. While traditional physical and statistical models have some applications, their high computational demands and difficulty in handling complex terrain or nonlinear relationships result in insufficient prediction accuracy.

[0003] Traditional fixed anemometer towers have a long history of use in wind farms, but they can only provide wind speed and direction data at fixed heights and locations, failing to capture wind speed variations at different heights and under different wind directions. For wind farms with complex terrain, fixed anemometer towers cannot provide comprehensive wind field data and can only conduct observations within a limited range. Although cheaper than lidar equipment, their installation and maintenance costs are still relatively high, and they cannot move or flexibly adjust the observation points.

[0004] To overcome the limitations of fixed wind measurement towers in adapting to complex terrain and their limited measurement range, lidar is widely used in various wind farms for real-time monitoring of incoming wind speed and direction. Floating lidar, as a new generation of wind measurement equipment, can achieve large-scale wind speed measurements, especially suitable for offshore wind farms or terrains where fixed towers are difficult to install. It can accurately measure wind speed and direction over long distances, without being limited by the location of the measurement point. However, its high cost and maintenance expenses limit its widespread installation, particularly at sea or in harsh weather conditions. When waves or wind conditions change significantly, the stability of floating equipment can affect measurement accuracy, and data may be interfered with.

[0005] To address the aforementioned issues, existing technology proposes a virtual lidar wind measurement method and system for wind farms (application publication number CN119294305A). This method combines CFD simulation, lidar data, and machine learning algorithms to achieve preset radar data for the locations of wind turbines in wind farms where lidar is not installed. Traditional virtual lidar wind measurement systems commonly use two models for evaluating the wake effect of wind turbine clusters during CFD simulation: analytical models and turbine entity modeling.

[0006] Analytical models are simplified mathematical expressions derived from fundamental principles of fluid mechanics, describing the velocity and turbulence intensity distributions in the wake. They simplify complex fluid dynamics processes, are computationally efficient, and are easy to implement. However, their accuracy is limited because they neglect key nonlinear terms and complex turbulent structures in the Navier-Stokes equations.

[0007] Solid modeling of wind turbines involves constructing the complete geometry (tower, nacelle, blades) of the wind turbine within the computational domain and setting it as a solid boundary. This method offers advantages such as high accuracy and versatility. However, to represent the complex geometry of wind turbines in detail, solid modeling requires a massive amount of mesh, resulting in extremely high simulation computation costs. Summary of the Invention

[0008] In view of this, the purpose of this invention is to provide a CFD database optimization method for virtual lidar wind measurement systems. The method uses an actuated disk model to evaluate the wake and establish a database. Compared with the traditional CFD simulation database in existing virtual lidar wind measurement methods, this method makes up for the lack of physical mechanisms in analytical models, corrects the time delay effect of the flow field, greatly reduces the computational cost, and improves the accuracy of the database.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A CFD database optimization method for a virtual lidar wind measurement system includes:

[0011] The target wind farm was geometrically modeled and meshed using ICEM CFD to obtain the model mesh. The model mesh is an unstructured mesh, and its near-surface region has six boundary layer meshes.

[0012] Based on the improved actuator disk model, the improved actuator disks of multiple wind turbines are constructed on the model mesh to obtain the target model mesh;

[0013] The first boundary condition for CFD simulation and the second boundary condition for the improved actuator disk model are set to perform CFD simulation on the target model mesh. During the simulation, the full wind direction and full wind speed are simulated by setting the Fluent inlet boundary condition until the flow field reaches a steady state, forming a wind farm simulation CFD database containing multiple sets of operating condition data for the full wind direction and full wind speed. The second boundary condition is set to interior, and the momentum source term is introduced in Fluent to simulate the obstruction and deflection effect of the wind turbine on the incoming wind.

[0014] Furthermore, if the target model mesh is used for all-wind direction simulation, the computational domain for geometric modeling and mesh generation will be set to a square prism.

[0015] Furthermore, the improvement of the actuation disk model is that when simulating multiple wind turbines, the wind speed at each actuation disk is used to correct the momentum source term, and a radial distribution function is introduced to correct the distribution of the momentum source term. This is to prevent the influence of the wake of the upstream unit on the downstream unit when simulating multiple wind turbines at the same time from causing the inflow wind speed of the downstream unit to be undetermined.

[0016] The improved momentum source term is:

[0017]

[0018] In the formula, , and These represent the momentum source terms in the improved x, y, and z directions, respectively. U represents the air density, and U represents the magnitude of the fluid velocity vector, with a value of 1. Where u represents the velocity component in the x-direction, v represents the velocity component in the y-direction, and w represents the velocity component in the z-direction. Indicates the thickness of the actuator disk. , and These represent the drag coefficients of the actuator disk in the x, y, and z directions, respectively. The radial distribution function is represented by r, which is the ratio of the distance between the grid center point and the actuation disk center to the actuation disk radius R0. and These are the model parameters.

[0019] Furthermore, the first boundary conditions for the CFD simulation are set as follows:

[0020] The incoming flow boundary is set as a velocity inlet, the outlet boundary is set as a pressure outlet, the ground surface adopts a no-slip wall condition, and the top and side boundaries are set as symmetrical boundaries.

[0021] When the incoming flow direction is not perpendicular to the inlet face, oblique inflow is achieved by specifying the incoming flow angle, and both outlets are set as pressure outlets.

[0022] Furthermore, during CFD simulations, a Realizable k-ε turbulence model was selected, and the SIMPLE algorithm was used to handle pressure-velocity coupling. Simultaneously, the discretization scheme was set in Fluent software: the gradient term used the least squares element method, the pressure term used a second-order scheme, the momentum term used a second-order upwind scheme, and both the turbulent kinetic energy and turbulent dissipation rate equations used second-order upwind discretization. The maximum number of iterations was set to 1200, and the convergence residual criterion for velocity and pressure was set to [value missing]. .

[0023] Furthermore, during CFD simulations, a separate turbine thrust coefficient is set for each independent operating condition to ensure physical consistency.

[0024] Furthermore, during CFD simulation, a monitoring point is set at the outlet. When the monitored wind speed meets the preset stability conditions, it is determined that the flow field has reached a stable state.

[0025] The beneficial effects of this invention are as follows:

[0026] This invention proposes a CFD database optimization method for virtual lidar wind measurement systems. An improved actuated disk model is used for wake evaluation and database establishment. The actuated disk model is a hybrid method combining analytical models and wind turbine entity modeling, achieving higher accuracy than analytical models while significantly reducing the computational cost of traditional entity modeling. It utilizes CFD to solve fluid equations (RANS or LES), but does not directly analyze the wind turbine geometry details. Instead, it adds source terms to the Fluid simulation to equivalently simulate the forces of the incoming gas. Since conventionally studied wind farms are located in complex mountainous terrain, affected by topographical undulations and the mutual influence of wakes between wind turbines, and considering the requirements of large-scale simulation and computational efficiency, no researchers in the current technical field have applied the actuated disk model to the prediction field of virtual lidar. Therefore, this paper uses an improved actuated disk model for wake effect evaluation, filling the gap in the lack of high-precision databases in CFD simulations of traditional virtual lidar wind measurement systems. It compensates for the lack of physical mechanisms in analytical models, corrects the time delay effect of the flow field, and significantly reduces computational costs while improving database accuracy.

[0027] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0028] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0029] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0030] Figure 1 This is a flowchart of a CFD database optimization method for a virtual lidar wind measurement system according to an embodiment of the present invention;

[0031] Figure 2 This is a schematic diagram of the wind turbine location and lidar measurement points of a virtual lidar wind measurement system according to an embodiment of the present invention;

[0032] Figure 3 This is a wind speed data graph obtained by a virtual lidar wind measurement system in an embodiment of the present invention;

[0033] Figure 4 This is a schematic diagram of the CFD simulation computation domain during CFD database modeling of a virtual lidar wind measurement system in an embodiment of the present invention;

[0034] Figure 5 This is a schematic diagram of a CFD simulation database for a virtual lidar wind measurement system according to an embodiment of the present invention;

[0035] Figure 6 This is a box plot of the short-term prediction error of a virtual lidar wind measurement system output in an embodiment of the present invention. Detailed Implementation

[0036] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0037] like Figure 1 As shown, this invention proposes a CFD database optimization method for a virtual lidar wind measurement system, comprising:

[0038] S101. Use ICEM CFD to perform geometric modeling and mesh generation on the target wind farm to obtain the model mesh; wherein, the model mesh adopts an unstructured mesh, and its near-ground region is set with six boundary layer meshes;

[0039] S102. Based on the improved actuator disk model, construct the improved actuator disks of multiple wind turbines on the model mesh to obtain the target model mesh;

[0040] S103. Set the first boundary condition for the CFD simulation and the second boundary condition for the improved actuator disk model to perform CFD simulation on the target model mesh; during the simulation, the simulation of all wind directions and all wind speeds is achieved by setting the Fluent inlet boundary condition; the second boundary condition is set to interior, and the momentum source term is introduced in Fluent to simulate the obstruction and deflection effect of the fan on the incoming airflow.

[0041] S104. Continue until the flow field reaches a stable state, forming a CFD database for wind farm simulation with all wind directions and all wind speeds containing multiple sets of operating condition data.

[0042] If the target model mesh in step S102 above is used for all-wind direction simulation, the computational domain used for geometric modeling and mesh generation will be set to a square column.

[0043] The improvement of the improved actuator disk model in step S102 above is that when simulating multiple wind turbines, the wind speed at each actuator disk is used to correct the momentum source term, and a radial distribution function is introduced to correct the distribution of the momentum source term. This is to prevent the influence of the wake of the upstream unit on the downstream unit when simulating multiple wind turbines at the same time from causing the inflow wind speed of the downstream unit to be undetermined.

[0044] The improved momentum source term is:

[0045]

[0046] In the formula, , and These represent the momentum source terms in the improved x, y, and z directions, respectively. U represents the air density, and U represents the magnitude of the fluid velocity vector, with a value of 1. Where u represents the velocity component in the x-direction, v represents the velocity component in the y-direction, and w represents the velocity component in the z-direction. Indicates the thickness of the actuator disk. , and These represent the drag coefficients of the actuator disk in the x, y, and z directions, respectively. The radial distribution function is represented by r, which is the ratio of the distance between the grid center point and the actuation disk center to the actuation disk radius R0. and These are the model parameters.

[0047] The first boundary conditions for the CFD simulation in step S103 above are set as follows: the incoming flow boundary is set as a velocity inlet, the outlet boundary is set as a pressure outlet, the ground adopts a no-slip wall condition, and the top and side boundaries are set as symmetrical boundaries; when the incoming flow direction is not perpendicular to the inlet surface, oblique inflow is carried out by specifying the incoming flow angle, and both outlets are set as pressure outlets.

[0048] In the CFD simulation, a Realizable k-ε turbulence model was selected, and the SIMPLE algorithm was used to handle pressure-velocity coupling. Simultaneously, in Fluent software, the discretization scheme was set as follows: the gradient term used the least squares element method, the pressure term used a second-order scheme, the momentum term used a second-order upwind scheme, and both the turbulent kinetic energy and turbulent dissipation rate equations used second-order upwind discretization. The maximum number of iterations was set to 1200, and the convergence residual criterion for velocity and pressure was set to [value missing]. .

[0049] When performing CFD simulations, the corresponding wind turbine thrust coefficient is set separately for each group of independent operating conditions to ensure physical consistency.

[0050] The method for determining the stable state of the flow field in step S104 above is as follows: when performing CFD simulation, a wind speed monitoring point is set at the outlet. When the monitored wind speed meets the preset stability conditions, it is determined that the flow field has reached a stable state.

[0051] In one specific embodiment, taking a mountain wind farm as an example, the main area of ​​the mountain wind farm is about 4.5 km wide in the east-west direction and about 10 km long in the north-south direction.

[0052] 1. Modeling and meshing using ICEM CFD software:

[0053] The model was scaled to a 1:1000 scale, and the computational domain was set to a square column for simulating all wind directions. The final size was 50m (east-west) × 50m (north-south) × 10m (height). The model used an unstructured mesh, and to accurately capture near-surface wind field characteristics, six boundary layer meshes were set in the near-surface region, with a minimum mesh size of 0.001m. The final total number of meshes was 3 million.

[0054] Actuation disk theory: This theory simplifies a real, complex 3D wind turbine into a cylindrical thin disk, which is the actuation disk. A circular thin disk is constructed on this complex terrain, and the maximum grid size of the thin plate is controlled to be below 0.01m.

[0055] 2. Set boundary conditions and parameters for CFD simulation:

[0056] The boundary conditions in the CFD simulation are set as follows: the inflow boundary is set as a velocity inlet, the outlet boundary is set as a pressure outlet, the ground surface uses a no-slip wall condition, and the top and side boundaries are set as symmetrical boundaries. When the inflow direction is not perpendicular to the inlet surface, the inflow angle is specified to consider oblique inflow, and both outlets are set as pressure outlets. The 24 fans are simplified into 24 corresponding improved actuated disk models, that is, cylindrical thin disks with the same ground height, thickness, and diameter as the fan blades are built at the corresponding fan locations, and their boundary conditions are set as interior. The momentum source term is introduced in Fluent to simulate the obstruction and deflection effect of the fans on the inflow.

[0057] In the numerical simulation, this paper preferentially selects the Realizable k-ε turbulence model and employs the SIMPLE algorithm to handle pressure-velocity coupling. Regarding the discretization scheme: the gradient term uses the least squares unit method, the pressure term uses a second-order scheme, the momentum term uses a second-order upwind scheme, and both the turbulent kinetic energy and turbulent dissipation rate equations use second-order upwind discretization. The maximum number of iterations is set to 1200, and the convergence residual criterion for velocity and pressure is set to [value missing]. In addition, monitoring points are set up at the outlet, and when the monitored wind speed tends to stabilize, it is determined that the flow field has reached a stable state.

[0058] As can be seen from the above description of the actuation disk theory, the actuation disk is an imaginary spatial region that can be approximated as being composed of an infinite number of blades. It obstructs and deflects the incoming airflow and converts wind energy into mechanical energy. This assumption is based on the premise that: (1) the airflow passing through the impeller is a uniform long constant flow and the gas is incompressible; (2) the friction between the airflow and the impeller is ignored; and (3) the thrust is evenly distributed to the impeller surface and the static pressure before and after the impeller remains unchanged.

[0059] In CFD numerical simulation, an actuation source term is added to the actuated disk to impede and deflect the incoming airflow. Based on momentum theory, the formula for calculating the conventional momentum source term is:

[0060]

[0061] In the formula, This represents the momentum source term in the n-direction; This represents the axial force in the n direction; R represents the radius of the actuating disk. Indicates the thickness of the actuator disk; Pi; taking the x-direction as an example:

[0062]

[0063] In the formula, Indicates air density; The drag coefficient (also known as the resistance coefficient) of the actuation disk in the x-axis direction is represented by U; U represents the magnitude of the fluid velocity vector, and its value is... Where u represents the component of velocity in the x-direction, v represents the component of velocity in the y-direction, and w represents the component of velocity in the z-direction;

[0064] In the above formula, the momentum source term is calculated from the inflow wind velocity and the thrust coefficient of the wind turbine itself. When simulating the wake field of a single unit, if the wind shearing effect is not considered, the incoming flow can be assumed to be uniform, so the inflow velocity can be directly selected to calculate the momentum source term. Since the momentum source term remains unchanged during the calculation, it is called a fixed source term.

[0065] The first improvement of the improved actuator disk model (or improved actuator disk) proposed in this application is that, when simulating multiple wind turbines, the wake of the upstream unit will affect the downstream unit, and the inflow velocity of the downstream unit cannot be determined. In this case, the momentum source term can be corrected by using the wind velocity at each actuator disk. The fixed momentum source term is transformed into an adaptive momentum source term (because the velocity of the actuator disk changes continuously with the calculation during CFD simulation, and the drag coefficient also changes accordingly, the momentum source term also changes with the flow field, hence it is called an adaptive momentum source term). The improved momentum source term formula is:

[0066]

[0067] in, This represents the drag coefficient (also known as the resistance coefficient) of the actuator disk in the x-axis direction. This represents the axial induction factor.

[0068] The second improvement of the proposed improved actuation disk model (or improved actuation disk) lies in the fact that the momentum source term derived above is uniformly distributed across the entire actuation disk, causing abrupt changes in the volume force at the edge of the actuation disk, resulting in a decoupling of velocity and pressure. To address this issue, a radial distribution function (i.e., the improved ADM) is introduced to correct the distribution of the momentum source term, as shown in the following equation:

[0069]

[0070] In the formula, The radial distribution function is represented by r, which is the ratio of the distance between the grid center point and the actuation disk center to the actuation disk radius R0. and These are model parameters; by adjusting their values, the distribution of momentum source terms in the actuation disk can be made more realistic, thereby improving calculation accuracy. In this application, and The preferred values ​​are -4 / 3 and 2 / 9;

[0071] Similarly, the improved momentum source terms in the y and z directions can be derived as follows:

[0072]

[0073] In the formula, and These represent the lateral drag coefficients of the actuation disk in the y-axis and z-axis directions, respectively.

[0074] The final improved momentum source term obtained is:

[0075]

[0076] 3. Establish a CFD database covering all wind directions and speeds:

[0077] The 360° all-wind-direction range was divided into 16 intervals, each interval being 22.5°. Since the cut-in wind speed of commonly used wind turbines is 3 m / s and the rated wind speed is 12 m / s, the incoming wind speed was selected from the range of 3 to 12 m / s, with intervals of 0.5 m / s, resulting in 19 wind speed groups. Simulation of all wind speeds and directions was achieved by setting Fluent inlet boundary conditions. During CFD simulation, the aerodynamic characteristics of the wind turbines were strictly followed. The turbine thrust coefficient was set individually for each independent operating condition to ensure physical consistency. Finally, a CFD database containing 16 × 19 sets of operating condition data for all wind directions and wind speeds of the wind farm simulation was formed.

[0078] The main difference between the above technical solution and the existing CFD simulation of the all-wind-direction wind field in the virtual lidar wind measurement system for wind farms lies in the wake model. In numerical simulation, there are two common models for evaluating the wake effect of wind turbine clusters: analytical models and turbine solid modeling. Analytical models are simplified mathematical expressions derived from the basic principles of fluid mechanics, describing the wake velocity and turbulence intensity distribution. Analytical models simplify complex fluid dynamic processes, are computationally efficient, and are easy to implement. However, due to the neglect of key nonlinear terms and complex turbulent structures in the Navier-Stokes equations, the accuracy of analytical models is limited. Turbine solid modeling, on the other hand, completely constructs the geometric structure of the wind turbine (tower, nacelle, blades) in the computational domain and sets it as a solid boundary, offering advantages such as high accuracy and strong versatility. However, to represent the complex geometric structure of the wind turbine in detail, the mesh size of solid modeling is enormous, resulting in extremely high simulation computation costs.

[0079] The actuated disk model is a hybrid approach combining analytical models and wind turbine solid modeling. It not only achieves higher accuracy than analytical models but also significantly reduces the computational cost of traditional solid modeling. It utilizes CFD to solve the fluid equations (RANS or LES), but instead of directly analyzing the wind turbine geometry, it simulates the forces acting on the incoming gas through the addition of source terms in the Fluent simulation. Since the wind farm in this study is located amidst complex mountain terrain, affected by topographical undulations and the wake effects between wind turbines, and considering the requirements of large-scale simulation and computational efficiency, this application employs the actuated disk model for wake effect assessment.

[0080] In summary, using the actuated disk model for wake evaluation and establishing a database compensates for the lack of physical mechanisms in the analytical model, corrects the time delay effect of the flow field, and significantly reduces computational costs while improving the accuracy of the database.

[0081] Furthermore, to better illustrate the application of the optimized database proposed in this application in a virtual lidar wind measurement system, in one embodiment, a wind farm virtual lidar wind measurement method based on deep learning and improved actuation disk theory includes:

[0082] Step 1: Collection of GIS topographic data, anemometer wind speed and direction data, and lidar wind measurement data: Collect GIS topographic data with a resolution of 30m, and collect anemometer and lidar measured data.

[0083] Step 2: Mark existing and predicted data, and analyze the correlation between the location to be measured and the location with known data: Record the anemometer and lidar data as T and L respectively. Mark the location as installed (already installed) or unknown (not installed) depending on whether lidar is installed, for example, L... unknown Represents virtual lidar data;

[0084] Simultaneously, the anemometer data is divided into different intervals based on wind direction angle, and within each wind direction interval, T is analyzed. install (Real wind speed and direction indicator) and T unknown (Virtual wind speed and direction instrument) Statistical correlation between wind speed data to find the camera position with the strongest correlation to the camera position without lidar installed;

[0085] Step 3: Construct a CFD full wind speed and direction database (an optimized CFD database proposed in this application): After using ICEM to model the complex terrain at scale and divide it into grids, perform CFD numerical simulations of the full wind direction. Utilize the improved actuator disk theory to fully consider the wake interference effect between wind turbine units and establish a CFD simulation result database of the wind farm with complex flow field information.

[0086] Step 4: Correct the wind speed time history information of all locations with cameras in the entire field: Based on the above correlation analysis, select the nearest reference camera (install) with the strongest correlation for each virtual lidar location (Lunknown); combine the measured anemometer data and CFD simulation database information to correct the wind speed time history information of all locations with cameras in the entire field; here, there are two correction methods, namely the whole field correction based on Linstall and Tinstall, denoted as Model 1 and Model 2; the two models are combined based on inverse distance weighted combination, which is called the combined model;

[0087] Step 5: Virtual LiDAR wind measurement, short-term future wind measurement, and virtual LiDAR short-term intelligent prediction: Based on measured wind speed data, a deep learning model is constructed to predict short-term wind speed. The output of this short-term prediction model is then used as input to the virtual LiDAR data intelligent wind measurement technology, ultimately forming a short-term intelligent prediction technology for wind farm virtual LiDAR wind measurement data, enabling the prediction of future wind speed and direction at unknown locations.

[0088] Existing virtual wind measurement technologies mainly focus on data-driven modeling and physics-AI fusion methods. However, models built using data-driven modeling to predict future wind speeds are overly simplistic and rely solely on pure data, resulting in a lack of physical mechanisms. While physics-AI fusion methods address the simplistic nature of the former, most wind speed prediction AI models are ANN or CNN, which inherently suffer from gradient vanishing and exploding, leading to high computational costs and low efficiency. By integrating the aforementioned optimized database into a virtual lidar wind measurement system for wind farms, utilizing an improved actuator disk for CFD simulation, and combining deep learning algorithms, the proposed virtual lidar wind measurement technology significantly improves the accuracy of existing virtual lidar predictions. Compared to self-predictive models based solely on lidar and anemometer data, the proposed method achieves a significant improvement, with an RMSE error below 2.2565 m / s and a mean absolute error below 1.8789 m / s, representing a 13.3–21.4% improvement in prediction accuracy compared to the aforementioned self-predictive models.

[0089] In one specific embodiment, focusing on a wind farm with 33 wind turbines on complex terrain, the study primarily examines three turbines equipped with Doppler lasers. Data from 14:00 on February 27th to 07:00 on April 6th of a given year (a total duration of 38 days and 17 hours) is used to construct a virtual radar for wind speed prediction. The research background and average wind speed sequence are as follows: Figure 2 and Figure 3 As shown;

[0090] Figure 2 This is a schematic diagram of the wind turbine locations and lidar measurement points. The x and y axes represent distances in meters (m), N is the direction of the map, indicating north, and 6, 10, and 14 are the wind turbine numbers. Only these 24 wind turbines are equipped with both anemometers and lidar instruments. This is also the basis for the following setting conditions: the simulated data can be verified with the measured data.

[0091] Figure 3This is a schematic diagram of wind speed data; the x-axis represents the date (day - number of days), the y-axis represents the equipment at different locations, and the z-axis represents the wind speed (m / s). It is used to visualize the measured wind data. The actual data is a matrix containing three sets of anemometer and wind vane data from three locations and three sets of LiDAR data (LiDAR instruments 6, 10, and 14). The wind speed data was measured by anemometers and three LiDARs. Due to the different start and end times of the data, a common time period was selected for analysis to facilitate engineering applications and model verification: from 14:00 on February 27, 2023 to 07:00 on April 6, 2023, a total duration of 38 days and 17 hours, with a data acquisition frequency of 0.2 Hz. Missing data within this time period was interpolated using interpolation theory to ensure data integrity. The specific three-dimensional measured wind speeds from the anemometers and wind vanes at the three locations are shown below. Figure 3 As shown;

[0092] First, a numerical model was established for the target area, a terrain transition section was set, and CFD simulations were conducted across all wind directions to construct a CFD database. A Realizable k-ε turbulence model was used to simulate wind fields in 16 wind directions. The aim was to consider local terrain effects and capture the correlation between wind speeds at anemometer locations and wind speeds at key grid node locations under different wind directions. The schematic diagrams of the CFD simulation computation domain and the CFD simulation database are shown below. Figure 4 and Figure 5 As shown.

[0093] Figure 4 This is a schematic diagram of the CFD simulation computational domain. The gray rectangles represent localized magnification of the mountain area, with mesh refinement applied. The numbers within the gray rectangles are the serial numbers of all the wind turbines.

[0094] Figure 5 This is a schematic diagram of a CFD simulation database, where the x-axis represents equipment at different locations, the z-axis represents wind speed (m / s), and the y-axis represents wind direction angle.

[0095] After completing modeling and CFD simulation, a self-predictive, virtual radar intelligent wind measurement, and short-term prediction model was developed. Specifically, it uses real lidar data installed at upstream turbine locations as model input, treats designated downstream turbine locations as virtual measurement points, and outputs predicted values ​​for comparison with downstream measured lidar data. To conduct rigorous accuracy verification, three operating conditions were set based on the correlation between wind turbines: Condition 1 uses turbine location 14 as input and turbine location 10 as output; Condition 2 uses turbine location 10 as input and turbine location 6 as output; and Condition 3 uses turbine location 6 as input and turbine location 10 as output. The core verification approach is as follows: using real lidar data installed at upstream turbine locations as model input, treating designated downstream turbine locations as virtual measurement points, and comparing the model's predicted values ​​with downstream measured lidar data. The operating condition settings are shown in the table below.

[0096]

[0097] The virtual lidar deep learning wind measurement technology based on purely measured data is denoted as model D.

[0098] The prediction results of virtual lidar intelligent wind measurement under the three operating conditions are good. The RMSE error of the virtual lidar prediction is less than 2.0788 m / s, the correlation coefficient R is higher than 0.9622, and the coefficient of determination R² is higher than 0.9122. Compared with model D, the prediction accuracy can be improved by 14.6~16.2%. The deep learning intelligent prediction results of anemometer and lidar data from the three locations are excellent. The established model has good overall performance. The prediction error of wind speed is less than 1.4792 m / s, the prediction accuracy exceeds 90.4%, and the prediction error of wind direction is less than 8.3°.

[0099] Finally, the output of this short-term prediction model is used as input information and substituted into the above-mentioned virtual machine lidar data intelligent wind measurement technology to ultimately form a short-term intelligent prediction technology for wind farm virtual lidar wind measurement data.

[0100] The resulting virtual lidar short-term intelligent prediction technology achieves high accuracy, with an RMSE error below 2.2565 m / s and a mean absolute error below 1.8789 m / s. Compared to model D, the prediction accuracy is improved by 13.3%–21.4%. The box plots of the combined model and the self-predictive model are shown below. Figure 6 As shown.

[0101] Figure 6 This is a box plot of the short-term prediction error of a virtual lidar system, including operating conditions 1, 2, and 3. The outer frame represents the interquartile range (IQR) of the data set, which is the range from the 25th percentile (Q1) to the 75th percentile (Q3).

[0102] The height of the outer frame reflects the degree of data dispersion: the higher the frame, the more dispersed the error distribution.

[0103] The horizontal line in the middle of the outer frame represents the median.

[0104] Figure 6 The inner frame represents the mean and its confidence interval. Usually, the small square in the center is the location of the mean, and the range of the inner frame reflects the confidence interval of the mean (such as the 95% confidence interval).

[0105] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A CFD database optimization method for a virtual lidar wind measurement system, characterized in that, include: The target wind farm was geometrically modeled and meshed using ICEM CFD to obtain the model mesh. The model mesh is an unstructured mesh, and its near-surface region has six boundary layer meshes. Based on the improved actuator disk model, the improved actuator disks of multiple wind turbines are constructed on the model mesh to obtain the target model mesh; wherein, the target model mesh is used for all wind directions simulation, and the computational domain during geometric modeling and mesh generation is set as a square prism; The improvement of the actuation disk model is that when simulating multiple wind turbines, the wind speed at each actuation disk is used to correct the momentum source term, and a radial distribution function is introduced to correct the distribution of the momentum source term. This is to prevent the influence of the wake of the upstream unit on the downstream unit when simulating multiple wind turbines at the same time from causing the inflow wind speed of the downstream unit to be undetermined. The improved momentum source term is: In the formula, , and These represent the momentum source terms in the improved x, y, and z directions, respectively. U represents the air density, and U represents the magnitude of the fluid velocity vector, with a value of 1. Where u represents the velocity component in the x-direction, v represents the velocity component in the y-direction, and w represents the velocity component in the z-direction. Indicates the thickness of the actuator disk. , and These represent the drag coefficients of the actuator disk in the x, y, and z directions, respectively. The radial distribution function is represented by r, which is the ratio of the distance between the grid center point and the actuation disk center to the actuation disk radius R0. and These are model parameters; The first boundary condition for CFD simulation and the second boundary condition for the improved actuator disk model are set to perform CFD simulation on the target model mesh. During the simulation, the full wind direction and full wind speed are simulated by setting the Fluent inlet boundary condition until the flow field reaches a steady state, forming a wind farm simulation CFD database containing multiple sets of operating data for the full wind direction and full wind speed. The second boundary condition is set to interior, and the momentum source term is introduced in Fluent to simulate the obstruction and deflection effect of the wind turbine on the incoming wind. The first boundary condition for CFD simulation is set as follows: the incoming flow boundary is set as a velocity inlet, the outlet boundary is set as a pressure outlet, the ground adopts a no-slip wall condition, and the top and side boundaries are set as symmetrical boundaries. When the incoming flow direction is not perpendicular to the inlet surface, the inflow is oblique by specifying the incoming flow angle, and both outlets are set as pressure outlets.

2. The CFD database optimization method for a virtual lidar wind measurement system according to claim 1, characterized in that: For CFD simulations, a Realizable k-ε turbulence model was selected, and the SIMPLE algorithm was used to handle pressure-velocity coupling. Simultaneously, the discretization schemes in Fluent software were configured as follows: the gradient term used the least squares element method, the pressure term used a second-order scheme, the momentum term used a second-order upwind scheme, and both the turbulent kinetic energy and turbulent dissipation rate equations were discretized using second-order upwind discretization. The maximum number of iterations was set to 1200, and the convergence residual criterion for velocity and pressure was set to [value missing]. .

3. The CFD database optimization method for a virtual lidar wind measurement system according to claim 1, characterized in that: When performing CFD simulations, the corresponding turbine thrust coefficient is set separately for each independent operating condition to ensure physical consistency.

4. The CFD database optimization method for a virtual lidar wind measurement system according to claim 1, characterized in that: When performing CFD simulation, a monitoring point is set at the outlet. When the monitored wind speed meets the preset stability conditions, the flow field is determined to have reached a stable state.

Images