A fan wake flow parameter optimization method based on a data-driven agent model

Abstract

This invention discloses a method for optimizing wind turbine wake parameters based on a data-driven surrogate model, belonging to the field of wind turbine aerodynamic performance optimization technology. By analyzing the influence of key parameters of the dynamic wake model of the wind turbine on the wake velocity field, it reveals the influence of the DWM model on the wake velocity field. k vAmb , k vShr and C NearWake The core parameters are strongly correlated with the accuracy of wake simulation. The Kriging model is used to accurately construct the nonlinear mapping relationship between parameters and flow field errors, and the NSGA-II algorithm is driven to quickly locate the Pareto optimal parameter combination in the global space. The optimization strategy used has a significant effect on improving the prediction accuracy of the DWM model. Its final optimization results are in high agreement with the CFD simulation results, but the overall optimization calculation cost is much lower than the traditional parameter trial and error method and direct CFD iterative verification. It has the advantages of computational efficiency and good engineering application and promotion value in dealing with scenarios that require a lot of simulation calculations, such as wind farm layout optimization and wake interference analysis of wind turbine groups.

Description

Technical Field

[0001] This invention relates to the field of wind turbine aerodynamic performance optimization technology, and in particular to a method for optimizing wind turbine wake parameters based on a data-driven proxy model. Background Technology

[0002] With the rapid development of wind power technology and the continuous expansion of wind farm scale, the wake effect of wind turbines has become a key factor affecting the overall power generation efficiency of wind farms. Accurate simulation of wind turbine wake characteristics is crucial for improving power generation efficiency in wind farm micro-site selection, power prediction, and operation optimization. Dynamic wake models (DWMs) are widely used in engineering practice due to their good balance between computational efficiency and physical rationality; however, their simulation accuracy heavily depends on the reasonable values ​​of key parameters such as the near-wake attenuation coefficient, environmental turbulence coefficient, and shear layer coefficient. Currently, these parameters are often determined in engineering by empirical formulas or manual trial and error, lacking a systematic parameter optimization method. This traditional method is not only time-consuming and labor-intensive but also struggles to guarantee the model's adaptability under different operating conditions, often leading to significant deviations between the velocity distribution predictions in the mid-to-far wake region and the results of high-precision computational fluid dynamics (CFD). Although CFD methods can provide accurate flow field analysis, their enormous computational cost limits their application in parameter sensitivity analysis and rapid optimization. Summary of the Invention

[0003] The purpose of this invention is to provide a method for optimizing wind turbine wake parameters based on a data-driven surrogate model. By constructing a collaborative framework of a surrogate model and a multi-objective optimization algorithm, the method achieves automatic calibration of the near-wake attenuation coefficient, environmental turbulence coefficient, and shear layer coefficient in the DWM model. This systematically improves the prediction accuracy of the DWM model in the mid-to-far wake region, significantly reducing the error between its velocity distribution and high-fidelity CFD results. Furthermore, compared to directly using CFD for parameter sensitivity analysis, the method shortens the optimization calculation time from astronomical to hourly, greatly improving optimization efficiency while ensuring engineering applicability accuracy. This provides a reliable technical means for accurate modeling and efficient optimization of wind farms.

[0004] To achieve the above objectives, this invention provides a method for optimizing wind turbine wake parameters based on a data-driven proxy model, comprising the following steps:

[0005] S1. Determine the key parameters in the Dynamic Wake Model (DWM) that significantly affect the simulation accuracy as input variables for this optimization program, set the range of the key parameters, and set the output response target as the error of the wake velocity distribution between the DWM model and the CFD simulation at the key downstream section.

[0006] S2. Taking into account the impact of sample size on the accuracy and computation time cost of subsequent surrogate models, the Latin hypercube sampling method is adopted to determine the sample size.n This group provides a data foundation for the accurate construction of subsequent agent models;

[0007] S3. Submit the sample set generated in S2 to the DWM model in sequence to perform simulation calculation, obtain the corresponding wake velocity field distribution, and generate the corresponding wake shape diagram to provide a data basis for subsequent error calculation.

[0008] S4. Perform parameter-error calculations for each set of DWM simulation results;

[0009] S5. Use the parameter-error data table obtained in S4 to train the surrogate model to fit the complex nonlinear relationship between the input variables and the output error.

[0010] S6. Based on the established nonlinear mapping relationship between parameters and errors, multi-objective optimization is performed on key parameters to obtain the Pareto optimal parameter solution set that minimizes the difference in simulation results at downstream locations.

[0011] S7. Select a representative optimal parameter combination from the Pareto front as the final optimization result, feed it back into the DWM model for independent verification calculation, and repeat S3 and S4 to obtain the wake distribution characteristic curves of the DWM model and CFD simulation at different locations after data processing under this parameter condition; at the same time, calculate the error at the target section and the comprehensive error.

[0012] Preferably, in S1, the key parameters include the near-wake attenuation coefficient. C NearWake Environmental turbulence coefficient k vAmb and shear layer coefficient k vShr .

[0013] Preferably, in S3, the specific calculation process is as follows:

[0014] S31. Design the DWM model so that the wake is axisymmetric, steady-state, and has zero tangential velocity. V θ =0, the pressure term is ignored, and the Navier–Stokes equations are simplified in the cylindrical coordinate system to obtain a set of control equations for axial and radial velocities. The control equations of the DWM model are as follows:

[0015] (1)

[0016] (2)

[0017] in, It is the axial velocity and axial gradient. It is the radial velocity and radial gradient. It is the radial distance from the calculation point to the origin in the coordinate system;

[0018] S32, The spatial variation of the dynamic eddy viscosity, derived from experiments, approximates the turbulence effect in the wake and is expressed by the following formula:

[0019] (3)

[0020] in, F 1( x )and F 2( x ) is a function related to environmental turbulence and a function related to wake shearing. It is only applied to the near-wake region of the wind turbine, and is generally 1 in the far-wake region.

[0021] S33. Given that this model is applied to the far wake region, reasonable assumptions are made. F 1( x )and F 2( x When ) is 1, equation (3) is equivalent to:

[0022] (4)

[0023] in, Indicates the radius along the given downstream section. V x The minimum value, For the inflow wind speed, k vAmb and k vShr The coefficients represent the effects of environmental turbulence and the wake shear layer on the eddy current viscosity, respectively. R Wake It is the wake half-width. I Amb Turbulence intensity at the hub center;

[0024] The near-wake correction sub-model in the S34 and DWM models calculates the axial and radial wake velocity deficit at the rotor disk surface as the inlet boundary conditions for equations (1) and (2). To improve the accuracy of the far-wake solution, the near-wake correction considers the wind speed decrease and radial expansion of the wake in the pressure gradient region behind the rotor, while the solution in the far-wake ignores this. C T Under conditions <24 / 25, the axial induction coefficient at the radially distributed rotor disk a ( r Average thrust coefficient filtered by low-pass time FiltAzimAvg C T ( rAccording to the BEM theory of momentum region of blade unit momentum:

[0025] (5)

[0026] To avoid unrealistic high-speed losses at the blade tips, equation (5) does not directly consider hub and tip loss corrections; furthermore, FiltAzimAvg C T ( r The upper limit is 24 / 25;

[0027] S35, in n Calculate the axial wake velocity deficit at the +1 section. Radial wake velocity deficit Furthermore, both are radially distributed, and the calculation formula is:

[0028] (6)

[0029] (7)

[0030] (8)

[0031] in, r Plane Is with r Related wake radial expansion; yes r intermediate variables; C NearWake The wake determines how the wind speed decreases at a distance; it expands radially in the pressure gradient region and then recovers at a distance.

[0032] Preferably, in S4, the specific process is as follows:

[0033] S41. Perform nonlinear interpolation on the wake velocity distribution data generated by the DWM model to reconstruct the discrete data into a smooth curve;

[0034] S42. Select at equal intervals on each curve. m Using the group of points as representative samples, the accuracy-precision percentage (PAP) index was used to calculate the accuracy-precision percentage of each sample compared to the CFD reference model. x The velocity distribution error at a specified downstream section corresponds to the output variable. Y 1. Y 2. Y 3… Y x The PAP calculation formula is:

[0035] (9)

[0036] (10)

[0037] in, The ordinate represents the calculated value from the DWM model. The vertical axis represents the true value of the CFD model. m PAP represents the number of sample points; a larger PAP indicates better simulation results with CFD.

[0038] S43, All n The data sets are organized into structured tables, which contain the combination of input parameters and the output error values, providing a dataset for subsequent training of the surrogate model.

[0039] Preferably, when the variance determination coefficient of the surrogate model is set to be greater than 0.9 and the AIC information criterion value is less than 90 in S5, it indicates that the model has excellent fitting accuracy and generalization ability in capturing the nonlinear relationship between input and output.

[0040] Preferably, in S6, the optimization process is evaluated using a comprehensive error function:

[0041] (11)

[0042] (12)

[0043] The weight configuration is adjusted according to specific working conditions and actual needs. The optimization process ultimately outputs a set of Pareto optimal solutions, providing a parameter combination with clear physical meaning for subsequent decisions. The larger the value of SP, the closer the values ​​of the DW model and CFD simulation are.

[0044] Therefore, this invention adopts the aforementioned wind turbine wake parameter optimization method based on a data-driven surrogate model. By constructing a Kriging model to accurately map the nonlinear relationship between key parameters and wake errors, and utilizing a multi-objective genetic algorithm to achieve global optimization in the parameter space, this method significantly improves the prediction accuracy of the DWM model in the mid-to-far wake region compared to the traditional empirical trial-and-error method, and significantly reduces the error between its velocity distribution and the CFD reference results. At the same time, compared to directly using CFD for parameter sensitivity analysis, this method reduces the optimization computation cost by several orders of magnitude, achieving a breakthrough in optimization efficiency while ensuring engineering applicability accuracy, and providing reliable technical support for accurate modeling and efficient optimization of wind farms.

[0045] 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

[0046] Figure 1 This is a flowchart of an embodiment of the present invention;

[0047] Figure 2 This is a comparison of the wake distribution characteristic curves of DWM and CFD at different locations in embodiments of the present invention;

[0048] Figure 3 This is a wake morphology diagram generated according to an embodiment of the present invention;

[0049] Figure 4 This is the Pareto optimal solution in the embodiments of the present invention;

[0050] Figure 5 This is a comparison of the wake distribution characteristic curves of the optimized DWM and CFD at different locations according to an embodiment of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0052] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0053] Example 1

[0054] This invention provides a method for optimizing wind turbine wake parameters based on a data-driven proxy model, the process of which is as follows: Figure 1 As shown, it includes the following steps:

[0055] S1. Determine the key parameters in the Dynamic Wake Model (DWM) that significantly affect simulation accuracy as input variables for this optimization program, set the range of the key parameters, and set the output response target as the error in wake velocity distribution between the DWM model and the CFD simulation at the key downstream section; key parameters include the near-wake attenuation coefficient. C NearWake Environmental turbulence coefficient k vAmb and shear layer coefficient k vShr In this embodiment, the range of input variables is set as follows: C NearWake∈[1.0,2.1], k vAmb ∈[0,1.0], k vShr ∈[0,0.031]. The output target is defined as: Y 1 (error at section X / D=4) Y 2 (error at section X / D=5) and Y 3 (error at section X / D=6), where D=1.8m. Figure 2 The study presents a comparison of wake distribution characteristic curves at different locations after data processing of the DWM model and CFD simulation. Figure 2 The values ​​of the parameters are: C NearWake =1.8, k vShr =0.016, k vAmb =0.05, which intuitively shows the difference between the two and provides a clear optimization target for this study.

[0056] S2. Taking into account the impact of sample size on the accuracy and computation time cost of subsequent surrogate models, the Latin hypercube sampling method is adopted to determine the sample size. n Group, in this embodiment will n The value is set to 60 to provide a data foundation for the accurate construction of the subsequent agent model.

[0057] S3. Submit the sample sets generated in S2 sequentially to the DWM model for simulation calculation to obtain the corresponding wake velocity field distribution and generate the corresponding wake morphology diagram, providing a data basis for subsequent error calculation. The specific calculation process is as follows:

[0058] S31. Design the DWM model so that the wake is axisymmetric, steady-state, and has zero tangential velocity. V θ =0, the pressure term is ignored, and the Navier–Stokes equations are simplified in the cylindrical coordinate system to obtain a set of control equations for axial and radial velocities. The control equations of the DWM model are as follows:

[0059] (1)

[0060] (2)

[0061] in, It is the axial velocity and axial gradient. It is the radial velocity and radial gradient. It is the radial distance from the calculation point to the origin in the coordinate system;

[0062] S32, The spatial variation of the dynamic eddy viscosity, derived from experiments, approximates the turbulence effect in the wake and is expressed by the following formula:

[0063] (3)

[0064] in, F 1( x )and F 2( x ) is a function related to environmental turbulence and a function related to wake shearing. It is only applied to the near-wake region of the wind turbine, and is generally 1 in the far-wake region.

[0065] S33. Given that this model is applied to the far wake region, reasonable assumptions are made. F 1( x )and F 2( x When ) is 1, equation (3) is equivalent to:

[0066] (4)

[0067] in, Indicates the radius along the given downstream section. V x The minimum value, For the inflow wind speed, k vAmb and k vShr The coefficients represent the effects of environmental turbulence and the wake shear layer on the eddy current viscosity, respectively. R Wake It is the wake half-width. I Amb Turbulence intensity at the hub center;

[0068] The near-wake correction sub-model in the S34 and DWM models calculates the axial and radial wake velocity deficit at the rotor disk surface as the inlet boundary conditions for equations (1) and (2). To improve the accuracy of the far-wake solution, the near-wake correction considers the wind speed decrease and radial expansion of the wake in the pressure gradient region behind the rotor, while the solution in the far-wake ignores this. C T Under conditions <24 / 25, the axial induction coefficient at the radially distributed rotor disk a ( r Average thrust coefficient filtered by low-pass time FiltAzimAvg C T ( r According to the BEM theory of momentum region of blade unit momentum:

[0069] (5)

[0070] To avoid unrealistic high-speed losses at the blade tips, equation (5) does not directly consider hub and tip loss corrections; furthermore, FiltAzimAvg C T ( r The upper limit is 24 / 25;

[0071] S35, in n Calculate the axial wake velocity deficit at the +1 section. Radial wake velocity deficit Furthermore, both are radially distributed, and the calculation formula is:

[0072] (6)

[0073] (7)

[0074] (8)

[0075] in, r Plane Is with r Related wake radial expansion; yes r intermediate variables; C NearWake The wake determines how the wind speed decreases at a distance. It expands radially in the pressure gradient region and then recovers at a distance, resulting in an output wake shape as follows: Figure 3 As shown.

[0076] S4. For each set of DWM simulation results, perform parameter-error calculations. The specific process is as follows:

[0077] S41. Perform nonlinear interpolation on the wake velocity distribution data generated by the DWM model to reconstruct the discrete data into a smooth curve;

[0078] S42. Select at equal intervals on each curve. m Using the data points as representative samples, this embodiment selects 20 points, and calculates their accuracy versus precision percentage (PAP) against the CFD reference model based on these data points. x The velocity distribution error at a specified downstream section corresponds to the output variable. Y 1. Y 2. Y 3… Y x The PAP calculation formula is:

[0079] (9)

[0080] (10)

[0081] in, The ordinate represents the calculated value from the DWM model. The vertical axis represents the true value of the CFD model. m The number of sample points is shown in this embodiment. m A PAP value of 20 indicates a better simulation effect with CFD. The PAP value is a comprehensive indicator of the fit between the DWM prediction curve (red) and the CFD baseline curve (black). Figure 2 and Figure 5 A higher PAP value indicates a better fit between the two curves. Specifically, a high PAP value means that the predicted curve not only matches the baseline curve in its overall trend (corresponding to accuracy), but also accurately captures its local fluctuation characteristics (corresponding to precision). This metric, by quantifying the performance of these two dimensions, intuitively reflects how the optimized parameters make the DWM model's calculation results highly close to the high-fidelity CFD results.

[0082] S43, All n The data, consisting of 60 sets, is organized into a structured table, which contains the input parameter combinations and output error values, providing a dataset for subsequent training of the surrogate model.

[0083] S5. The parameter-error data table obtained in S4 is used to train the surrogate model to fit the complex nonlinear relationship between the input variables and the output error. In this embodiment, a Kriging model is selected as the surrogate model architecture. A variance determination coefficient greater than 0.9 and an AIC information criterion value less than 90 indicate that the model has excellent fitting accuracy and generalization ability in capturing the nonlinear relationship between input and output. In this embodiment, the variance determination coefficient reaches 0.93, and the AIC information criterion value is 80.

[0084] S6. The established nonlinear mapping relationship between parameters and errors is used to perform multi-objective optimization on key parameters, obtaining the Pareto optimal parameter solution set that minimizes the synchronous differences in simulation results at downstream locations. This embodiment, based on a trained Kriging surrogate model, employs the NSGA-II non-dominated sorting genetic algorithm with an elitist strategy for multi-objective optimization search to synchronously minimize the error objective. Through fast non-dominated sorting and crowding calculation, the Pareto front representing the optimal trade-off solution is sought. The optimization process is evaluated using a comprehensive error function.

[0085] (11)

[0086] (12)

[0087] The weight configuration is adjusted according to specific working conditions and actual needs. The optimization process ultimately outputs a set of Pareto optimal solutions, providing a parameter combination with clear physical meaning for subsequent decision-making. As can be seen from the expression and definition of PAP, the larger the value of SP, the closer the values ​​of the DW model and CFD simulation are. In this embodiment, considering that the flow field characteristics in the far wake region (X=6D section) have a more significant impact on engineering applications, the weight coefficient is set to... , To highlight Y 3. Its dominant role in the overall objective. This weighting can be flexibly adjusted by the reader according to specific working conditions and actual needs. The optimization process ultimately outputs a set of Pareto optimal solutions, such as... Figure 4 As shown, it can be observed that it is close to Y The large curvature at the vertices of the three axes indicates that in this region, Y A tiny improvement of 3 requires... Y 1 and Y The significant decrease in 2 comes at the cost of indicating that in the weighted error function... Y 3. The accuracy of the far wake region, which is the highest weight given to engineering decisions, is reasonable and provides a combination of parameters with clear physical meaning for subsequent decisions.

[0088] S7. Select a representative optimal parameter combination from the Pareto front as the final optimization result. In this embodiment, C NearWake =1.831, k vAmb =0.074, k vShr =0.00164, which is fed back into the DWM model for independent verification calculation. S3 and S4 are repeated to obtain the wake distribution characteristic curves of the DWM model and CFD simulation at different locations under this parameter condition, after data processing. Figure 5 As shown. Simultaneously, calculate according to equation (10). Y 1. Y 2. Y 3. Calculate the comprehensive error function according to equation (11). In this embodiment, the error values ​​of this parameter combination at the three target sections are respectively Y 1 = 0.477 Y 2 = 0.099 Y =3=0.574, and the overall error SP=0.4121 is calculated according to the predetermined weighting coefficient. Compared with the initial parameters, the optimized DWM model has significantly improved the simulation accuracy of each key section, especially in the far wake region X=6D, which fully verifies the effectiveness of this optimization process in reducing the error between the model and the CFD theoretical value, and demonstrates good engineering practical value.

[0089] Therefore, this invention adopts the aforementioned wind turbine wake parameter optimization method based on a data-driven surrogate model. By constructing a Kriging model to accurately map the nonlinear relationship between key parameters and wake errors, and utilizing a multi-objective genetic algorithm to achieve global optimization in the parameter space, this method significantly improves the prediction accuracy of the DWM model in the mid-to-far wake region compared to the traditional empirical trial-and-error method, and significantly reduces the error between its velocity distribution and the CFD reference results. At the same time, compared to directly using CFD for parameter sensitivity analysis, this method reduces the optimization computation cost by several orders of magnitude, achieving a breakthrough in optimization efficiency while ensuring engineering applicability accuracy, and providing reliable technical support for accurate modeling and efficient optimization of wind farms.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for optimizing wind turbine wake parameters based on a data-driven surrogate model, characterized in that, Includes the following steps: S1. Determine the key parameters in the Dynamic Wake Model (DWM) that significantly affect the simulation accuracy as input variables for this optimization program, set the range of the key parameters, and set the output response target as the error of the wake velocity distribution between the DWM model and the CFD simulation at the key downstream section. S2. Taking into account the impact of sample size on the accuracy and computation time cost of subsequent surrogate models, the Latin hypercube sampling method is adopted to determine the sample size. n This group provides a data foundation for the accurate construction of subsequent agent models; S3. Submit the sample sets generated in S2 sequentially to the DWM model for simulation calculation to obtain the corresponding wake velocity field distribution and generate the corresponding wake morphology diagram, providing a data basis for subsequent error calculation; the specific calculation process is as follows: S31. Design the DWM model so that the wake is axisymmetric, steady-state, and has zero tangential velocity. V θ =0, the pressure term is ignored, and the Navier–Stokes equations are simplified in the cylindrical coordinate system to obtain a set of control equations for axial and radial velocities. The control equations of the DWM model are as follows: ； (1) ； (2) in, It is the axial velocity and axial gradient. It is the radial velocity and radial gradient. It is the radial distance from the calculation point to the origin in the coordinate system; S32, The spatial variation of the dynamic eddy viscosity, derived from experiments, approximates the turbulence effect in the wake and is expressed by the following formula: ； (3) in, F 1( x )and F 2( x ) is a function related to environmental turbulence and a function related to wake shearing. It is only applied to the near-wake region of the wind turbine, and is generally 1 in the far-wake region. S33. Given that this model is applied to the far wake region, reasonable assumptions are made. F 1( x )and F 2( x When ) is 1, equation (3) is equivalent to: ； (4) in, Indicates the radius along the given downstream section. V x The minimum value, For the inflow wind speed, k vAmb and k vShr The coefficients represent the effects of environmental turbulence and the wake shear layer on the eddy current viscosity, respectively. R Wake It is the wake half-width. I Amb Turbulence intensity at the hub center; The near-wake correction sub-model in the S34 and DWM models calculates the axial and radial wake velocity deficit at the rotor disk surface as the inlet boundary conditions for equations (1) and (2). To improve the accuracy of the far-wake solution, the near-wake correction considers the wind speed decrease and radial expansion of the wake in the pressure gradient region behind the rotor, while the solution in the far-wake ignores this. C T Under conditions <24 / 25, the axial induction coefficient at the radially distributed rotor disk a ( r Average thrust coefficient filtered by low-pass time FiltAzimAvg C T ( r According to the BEM theory of momentum region of blade unit momentum: ； (5) To avoid unrealistic high-speed losses at the blade tips, equation (5) does not directly consider hub and tip loss corrections; furthermore, FiltAzimAvg C T ( r The upper limit is 24 / 25; S35, in n Calculate the axial wake velocity deficit at the +1 section. Radial wake velocity deficit Furthermore, both are radially distributed, and the calculation formula is: ； (6) ； (7) ； (8) in, r Plane Is with r Related wake radial expansion; yes r intermediate variables; C NearWake The wake expands radially in the pressure gradient region and then recovers at a distance, determining how the wind speed decreases at a distance; S4. Perform parameter-error calculations for each set of DWM simulation results; S5. Use the parameter-error data table obtained in S4 to train the surrogate model to fit the complex nonlinear relationship between the input variables and the output error. S6. Based on the established nonlinear mapping relationship between parameters and errors, multi-objective optimization is performed on key parameters to obtain the Pareto optimal parameter solution set that minimizes the difference in simulation results at downstream locations. S7. Select a representative optimal parameter combination from the Pareto front as the final optimization result, feed it back into the DWM model for independent verification calculation, and repeat S3 and S4 to obtain the wake distribution characteristic curves of the DWM model and CFD simulation at different locations after data processing under this parameter condition; at the same time, calculate the error at the target section and the comprehensive error.

2. The method for optimizing wind turbine wake parameters based on a data-driven surrogate model according to claim 1, characterized in that, In S1, key parameters include the near-wake attenuation coefficient. C NearWake Environmental turbulence coefficient k vAmb and shear layer coefficient k vShr .

3. The method for optimizing wind turbine wake parameters based on a data-driven proxy model according to claim 2, characterized in that, In S4, the specific process is as follows: S41. Perform nonlinear interpolation on the wake velocity distribution data generated by the DWM model to reconstruct the discrete data into a smooth curve; S42. Select at equal intervals on each curve. m Using the group of points as representative samples, the accuracy-precision percentage (PAP) index was used to calculate the accuracy-precision percentage of each sample compared to the CFD reference model. x The velocity distribution error at a specified downstream section corresponds to the output variable. Y 1. Y 2. Y 3… Y x The PAP calculation formula is: ； (9) )； (10) in, The ordinate represents the calculated value from the DWM model. The vertical axis represents the true value of the CFD model. m PAP represents the number of sample points; a larger PAP indicates better simulation results with CFD. S43, All n The data sets are organized into structured tables, which contain the combination of input parameters and the output error values, providing a dataset for subsequent training of the surrogate model.

4. The method for optimizing wind turbine wake parameters based on a data-driven proxy model according to claim 3, characterized in that, In S5, when the variance determination coefficient of the surrogate model is set to be greater than 0.9 and the AIC information criterion value is less than 90, it indicates that the model has excellent fitting accuracy and generalization ability in capturing the nonlinear relationship between input and output.

5. The method for optimizing wind turbine wake parameters based on a data-driven proxy model according to claim 4, characterized in that, In S6, the optimization process is evaluated using a comprehensive error function: ； (11) ； (12) in, , , ... Indicates the weighting coefficient. , , ... This represents the velocity distribution error value at the target section. The weight configuration is adjusted according to the specific working conditions and actual needs. The optimization process ultimately outputs a set of Pareto optimal solutions, providing a parameter combination with clear physical meaning for subsequent decisions. The larger the value of SP, the closer the values ​​of the DW model and CFD simulation are.

Images