Wind power plant wind speed time sequence prediction method based on SCADA data correction and CFD modeling

Abstract

The invention provides a wind power plant wind speed time sequence prediction method based on SCADA data correction and CFD modeling. The wind power plant wind speed time sequence prediction method comprises the steps that S1, historical SCADA data and synchronous reference wind speed data of a built wind power plant are collected and preprocessed; s2, establishing an SCADA wind speed correction model based on the preprocessed data, performing primary correction and secondary correction on the SCADA wind speed, and generating a reduced hub height wind speed time sequence; s3, constructing a CFD computational domain covering the constructed wind power plant and the planned wind power plant area, taking the reduced wind speed time sequence as a time-varying boundary condition, and obtaining a space wind speed transfer function through CFD simulation; and S4, extrapolating the reduced wind speed time sequence to each location point of the planned wind power plant by using a space wind speed transfer function, and generating high-precision wind speed time sequence prediction data. The method is especially suitable for wind resource evaluation scenes of complex fields, and overcomes the limitation that a traditional method is difficult to represent wind resource distribution of the complex fields.

Description

Technical Field

[0001] This invention relates to the field of new energy wind power, and in particular to a method for predicting wind speed time series of wind farms based on SCADA data correction and CFD modeling. Background Technology

[0002] Accurate assessment of wind energy resources is fundamental to wind farm planning, design, and economic analysis. A common method for wind farm resource assessment relies on erecting meteorological towers in the target area and conducting on-site wind measurements for at least one year to obtain representative time-series data such as wind speed and direction. However, this method is limited in application due to the high cost and long construction period of the meteorological towers, and the difficulty of representing the wind resource distribution in complex areas with single-point measurements.

[0003] With the development of big data technology, the SCADA system installed on wind turbines in wind farms continuously collects and stores a large amount of operational data, including wind speed, power, and rotational speed, which can theoretically serve as an alternative data source. However, directly using SCADA wind speed data for resource assessment has serious inherent biases: a) The anemometer is located at the rear of the nacelle, and is affected by the wake disturbance caused by blade rotation and the blockage effect of the nacelle itself, so the measured values ​​are generally lower than the actual incoming wind speed; b) Wind speed data is only recorded when the wind turbine is operating normally (such as between the cut-in and cut-out wind speeds), resulting in a large number of invalid and missing data; c) The data reflects the intervention results of the wind turbine control system (such as yaw and pitch), rather than the original natural wind conditions.

[0004] Therefore, there is an urgent need in this field for a technical solution that can overcome the above-mentioned defects and acquire long-term wind speed data in areas where no wind farms have been built, in a low-cost, efficient and accurate manner. Summary of the Invention

[0005] The main objective of this invention is to provide a wind speed time series prediction method for wind farms based on SCADA data correction and CFD modeling. This method solves the problems in the background technology, such as high construction cost and long construction period of traditional wind measurement tower methods, difficulty in representing the wind resource distribution of complex fields by single-point measurement, and the existence of wake disturbance, data gaps, and deviation caused by turbine control intervention in SCADA wind speed data, which cannot be directly used for long-term wind speed data acquisition of unbuilt wind farms.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by this invention is: a wind farm wind speed time series prediction method based on SCADA data correction and CFD modeling, the method comprising: S1. Collect historical SCADA data and concurrent reference wind speed data of existing wind farms and perform preprocessing. S2. Based on the preprocessed data, establish a SCADA wind speed correction model, perform first and second corrections on the SCADA wind speed, and generate the restored hub height wind speed time series. S3. Construct a CFD computational domain covering the areas of existing and planned wind farms, use the restored wind speed time series as time-varying boundary conditions, and obtain the spatial wind speed transfer function through CFD simulation. S4. Using the spatial wind speed transfer function, the restored wind speed time series is extrapolated to each turbine location in the planned wind farm to generate high-precision wind speed time series prediction data.

[0007] In the preferred embodiment, in step S1, the SCADA data includes time period, hub height wind speed, wind direction, power, temperature and air pressure data; The reference wind speed data for the same period comes from the wind measurement tower or the wind power prediction system; Preprocessing includes time synchronization, invalid value removal, outlier filtering, and missing data imputation.

[0008] In the preferred embodiment, step S2 includes the following correction: The average power is calculated based on the measured power of the wind turbine in the wind speed range of 4 m / s to 9 m / s. The theoretical average wind speed is then inferred from the average power curve of the wind turbine, and a first-order correction factor is calculated. Correct the original cabin wind speed; in, The theoretical average wind speed, This is the average wind speed measured by SCADA.

[0009] In the preferred embodiment, in step S2, the secondary correction only affects the correlation coefficient between the wind speed and the reference wind speed after the primary correction. It is executed in time and calibrated based on the regression relationship between the two.

[0010] In the preferred embodiment, when calculating the average power in the wind speed range of 4 m / s to 9 m / s, a weighted kernel density estimation method is used: The measured power at each wind speed is used as a sample, and a kernel function is constructed using the wind speed frequency as the weight. The probability-weighted mean of the power within the range is calculated through kernel density integration, as shown in the formula: ; in The power is the interval average. This represents the measured power corresponding to wind speed v. The wind speed probability density function is based on kernel density estimation. The Epanechnikov kernel is selected as the kernel function, and the bandwidth is determined by cross-validation.

[0011] In the preferred embodiment, the calibration steps based on the regression relationship between the two include: Wind speed after one correction Compared with reference wind speed The initial regression coefficients were calculated using the M-estimation method for the sample points. , A preliminary regression model was obtained. ; Calculate the residuals for each sample point Construct a weight function based on the absolute value of the residuals. , where s is the median absolute deviation of the residuals; By weight Weight the samples and recalculate the regression coefficients. , The final calibration model is obtained. ,in This is the wind speed after secondary correction.

[0012] In the preferred embodiment, in step S3, the CFD computational domain is constructed based on the digital elevation model (DEM) and surface roughness data, and the Navier-Stokes equations are solved to simulate the three-dimensional wind flow field.

[0013] In the preferred scheme, the Navier-Stokes equations are solved using a separate implicit solution algorithm, with the following steps: A1. The computational domain is discretized using the finite volume method. Unstructured tetrahedral meshes are used for the control volume. Boundary layer refinement is applied to the near-ground region. The mesh quality is verified by orthogonality and stretch ratio. A2. The velocity field is initialized based on the logarithmic wind profile, the pressure field is initialized based on the hydrostatic equilibrium, and the turbulence parameters are set based on the initial values ​​of the k-ω SST turbulence model. A3. The time discretization adopts the second-order implicit scheme, the convection term adopts the QUICK scheme, the diffusion term adopts the central difference scheme, and the pressure and velocity coupling adopts the SIMPLEC algorithm. A4. First, solve the momentum prediction equation to obtain the intermediate velocity field. , , Then, the pressure field p' is adjusted using the pressure correction equation, and the velocity field is updated to... , , ; A5. Solve the k-equation and ω-equation of the turbulence equations, update the turbulent viscosity coefficient, and determine whether the calculation residual is less than [the value of the original equation]. If the condition is not met, return to step A4 for iteration until a stable three-dimensional airflow field distribution is obtained.

[0014] In the preferred scheme, the step of extrapolating the restored wind speed time series to each turbine location of the planned wind farm using the spatial wind speed transfer function includes: B1. First, determine the spatial correspondence between the reference points of the existing wind farm and the pre-selected turbine sites of the planned wind farm, clarify the corrected wind speed time series dataset of the reference points, and at the same time determine the number and spatial location of all pre-selected turbine sites of the planned wind farm. B2. Next, call the spatial wind speed transfer function obtained through CFD simulation in step S3; B3. Then, for each pre-selected machine location, the wind speed value at each moment in the wind speed time series after the reference point correction is multiplied by the amplification factor corresponding to that location to obtain the preliminary extrapolated wind speed series. B4. Then, introduce a wind condition time series correlation correction factor, multiply the wind speed value at each moment in the preliminary extrapolated wind speed sequence by the correction factor, and complete the time series feature matching correction. B5. Then, based on the technical parameters of the wind turbines to be selected for the planned wind farm, the time values ​​in the corrected wind speed sequence that exceed the cut-in and cut-out wind speed range are adjusted by using the linear interpolation method of the wind speed at adjacent effective time points to ensure that they meet the actual operating wind speed requirements of the wind turbines. B6. Finally, repeat steps B3-B5 above for the calculation process of a single turbine site to complete the wind speed time series calculation for all pre-selected turbine sites in the planned wind farm and generate high-precision wind speed time series prediction data for each turbine site.

[0015] A wind farm wind speed time series prediction system includes: The data acquisition and preprocessing module is used to acquire and clean SCADA data and reference wind speed data; The data correction and restoration module is used to build and apply the wind speed correction model to generate a restored wind speed sequence. The CFD modeling and extrapolation module is used to establish the computational domain, run CFD simulations, and generate spatial transfer functions. The database and output module is used to store intermediate results and output wind speed time series prediction data for each turbine location in the planned wind farm.

[0016] This invention provides a wind speed time series prediction method for wind farms based on SCADA data correction and CFD modeling. By utilizing historical SCADA data from existing wind farms, combined with data preprocessing and phased correction, it effectively eliminates the inherent biases in SCADA wind speed data caused by wake disturbances, data gaps, and turbine control interventions. This method reconstructs a hub height wind speed time series that closely approximates real natural wind conditions. It eliminates the need for costly and time-consuming single-point measurements from meteorological towers, reducing the cost of wind resource data acquisition and shortening the data acquisition cycle. Simultaneously, by constructing a CFD model covering both existing and planned wind farm areas, and combining high-precision geographic information data with spatial wind speed transfer functions, it achieves accurate spatial extrapolation of wind speed data from existing wind farms to planned wind farms. This provides reliable, high-precision wind speed time series for each turbine location in planned wind farms, offering high-quality data support for early-stage power generation assessment, economic analysis, and layout optimization. This enhances the scientific and economical nature of wind farm planning and design, and is particularly suitable for wind resource assessment scenarios in complex areas, overcoming the limitations of traditional methods that cannot represent the wind resource distribution in complex areas. Attached Figure Description

[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is an overall flowchart of the method of the present invention. Detailed Implementation

[0018] Example 1 like Figure 1 As shown, a wind farm wind speed time series prediction method based on SCADA data correction and CFD modeling is presented. The method includes: S1. Collect historical SCADA data and concurrent reference wind speed data of existing wind farms and perform preprocessing. S2. Based on the preprocessed data, establish a SCADA wind speed correction model, perform first and second corrections on the SCADA wind speed, and generate the restored hub height wind speed time series. S3. Construct a CFD computational domain covering the areas of existing and planned wind farms, use the restored wind speed time series as time-varying boundary conditions, and obtain the spatial wind speed transfer function through CFD simulation. S4. Using the spatial wind speed transfer function, the restored wind speed time series is extrapolated to each turbine location of the planned wind farm to generate high-precision wind speed time series prediction data. It outputs high-precision wind speed time series data for each turbine location in the planned wind farm, which can be used for power generation assessment, economic analysis and wind farm layout optimization.

[0019] First, step S1 is executed to collect historical SCADA data and concurrent reference wind speed data from existing wind farms, and both types of data undergo preprocessing. Historical SCADA data includes time periods, wind speed, wind direction, power, temperature, and air pressure at the turbine hub height; this data reflects the turbine's operating status and the surrounding environment. Reference wind speed data comes from meteorological tower measurements or wind power prediction systems and is primarily used to calibrate SCADA data deviations. During preprocessing, time synchronization is first used to ensure the two types of data match in the time dimension. Then, invalid and outlier values ​​are removed. Invalid values ​​include meaningless wind speed records when the turbine is shut down, and outlier values ​​include wind speed data corresponding to power exceeding the rated range. Finally, missing data is interpolated to form a well-organized dataset, laying the data foundation for subsequent correction steps.

[0020] Next, step S2 is performed, establishing a SCADA wind speed correction model based on the preprocessed dataset. Wind speed correction is completed in two stages. The first correction involves calculating the average power curve of wind turbines of the same type, combining this with the average power of the turbines in the effective wind speed range of 4 m / s to 9 m / s to infer the theoretical average wind speed, and then calculating a correction coefficient to correct the original SCADA nacelle wind speed to a preliminary corrected wind speed close to the actual wind conditions. The second correction requires verifying the correlation between the wind speed after the first correction and the reference wind speed. It is only performed when the correlation coefficient R² > 0.81. Then, a regression algorithm is used to establish the correlation between the two, further calibrating the wind speed data, and finally generating the restored hub height wind speed time series. The restored hub height wind speed time series is a long-term wind speed data set that eliminates the inherent biases of the SCADA data. The inherent biases of SCADA data include nacelle wake disturbances and measurement inaccuracies caused by turbine control interventions. This series contains hub height wind speed values ​​at different time points, reflecting natural wind conditions and providing reliable input for CFD modeling.

[0021] Then, step S3 is performed to construct a three-dimensional CFD computational domain covering the areas of existing and planned wind farms. Simultaneously, a high-precision digital elevation model (DEM) and surface roughness data are imported. The DEM reflects the regional topographic relief, while the surface roughness data affects wind flow resistance. Next, the hub height-wind speed time series restored in step S2 is used as the time-varying boundary condition for the CFD model. The time-varying boundary condition refers to the wind input parameters that change over time, simulating the temporal changes of real wind conditions. Finally, the three-dimensional wind flow field within the computational domain is simulated by solving the Navier-Stokes equations, thereby obtaining the spatial wind speed transfer function. The spatial wind speed transfer function is a parameter determined by CFD simulation. It reflects the wind speed correlation between the reference point of an existing wind farm and the pre-selected turbine locations of a planned wind farm. It is usually represented as a wind speed amplification factor matrix. Each element in the matrix corresponds to the wind speed transfer coefficient between a turbine location of the planned wind farm and the reference point. This coefficient is determined by the ratio of the mean steady-state wind speed at the corresponding turbine location in the CFD simulation to the mean steady-state wind speed at the reference point. It can realize the spatial mapping and transformation of wind speed data.

[0022] Finally, step S4 is implemented. Using the aforementioned spatial wind speed transfer function, the wind speed time series restored from the reference points of the existing wind farm is extrapolated to each turbine location in the planned wind farm. Through the coefficients in the transfer function, the wind speed values ​​at different time points of the reference points are converted into the corresponding wind speed values ​​at the planned wind farm turbine locations, ultimately generating high-precision wind speed time series prediction data. This high-precision wind speed time series prediction data is a set of wind speeds at each turbine location in the planned wind farm within the corresponding time period. Each turbine location has a corresponding sequence containing wind speeds at multiple time points, accurately reflecting the wind condition changes at each turbine location and can be directly used for subsequent wind farm design and analysis.

[0023] By performing phased corrections to SCADA data, the inherent biases in SCADA wind speed data caused by wake disturbances, data gaps, and turbine control interventions are effectively resolved. This eliminates the need for long-term on-site wind measurements using traditional meteorological towers, significantly reducing the cost of wind resource data acquisition and shortening the data collection cycle. Furthermore, by leveraging CFD modeling and spatial wind speed transfer functions, reliable wind speed data from existing wind farms can be accurately extrapolated to each turbine location in planned wind farms. This overcomes the limitations of traditional single-point measurements, which cannot represent the wind resource distribution in complex areas. The generated high-precision wind speed time series prediction data accurately reflects the wind characteristics at each location of the planned wind farm, providing high-quality data support for early-stage power generation assessment, economic analysis, and layout optimization. This enhances the scientific rigor and rationality of planning schemes, ensuring the economic efficiency and effectiveness of wind farm construction and operation.

[0024] In the preferred embodiment, in step S1, the SCADA data includes time period, hub height wind speed, wind direction, power, temperature and air pressure data; The reference wind speed data for the same period comes from the wind measurement tower or the wind power prediction system; Preprocessing includes time synchronization, invalid value removal, outlier filtering, and missing data imputation.

[0025] By clearly defining the core components of SCADA data, it is ensured that the collected data comprehensively reflects the operating status of wind turbines and the characteristics of the wind field environment, providing rich basic parameters for subsequent wind speed correction. The introduction of concurrent reference wind speed data provides a reliable basis for calibrating deviations in SCADA wind speed data, solving the problem of insufficient accuracy when SCADA data is used alone. The four preprocessing operations optimize data quality from three dimensions: time matching, data validity, and data integrity. Invalid and abnormal data are removed to avoid interfering with subsequent calculations, and missing data is supplemented to ensure dataset completeness. Ultimately, a well-organized and reliable dataset is formed, providing high-quality input for subsequent steps such as SCADA wind speed correction model construction and CFD modeling, ensuring the accuracy and stability of the entire wind speed prediction scheme. Furthermore, compared to the traditional method that relies solely on wind measurement towers, this reduces dependence on a single data source and lowers data acquisition costs.

[0026] In the preferred embodiment, step S2 includes the following correction: The average power is calculated based on the measured power of the wind turbine in the wind speed range of 4 m / s to 9 m / s. The theoretical average wind speed is then inferred from the average power curve of the wind turbine, and a first-order correction factor is calculated. Correct the original cabin wind speed; in, The theoretical average wind speed, This is the average wind speed measured by SCADA.

[0027] In the preferred embodiment, in step S2, the secondary correction only affects the correlation coefficient between the wind speed and the reference wind speed after the primary correction. It is executed in time and calibrated based on the regression relationship between the two.

[0028] In the preferred embodiment, when calculating the average power in the wind speed range of 4 m / s to 9 m / s, a weighted kernel density estimation method is used: The measured power at each wind speed is used as a sample, and a kernel function is constructed using the wind speed frequency as the weight. The probability-weighted mean of the power within the range is calculated through kernel density integration, as shown in the formula: ; in The power is the interval average. This represents the measured power corresponding to wind speed v. The wind speed probability density function is based on kernel density estimation. The Epanechnikov kernel is selected as the kernel function, and the bandwidth is determined by cross-validation.

[0029] In the preferred embodiment, the calibration steps based on the regression relationship between the two include: Wind speed after one correction Compared with reference wind speed The initial regression coefficients were calculated using the M-estimation method for the sample points. , A preliminary regression model was obtained. ; Calculate the residuals for each sample point Construct a weight function based on the absolute value of the residuals. , where s is the median absolute deviation of the residuals; By weight Weight the samples and recalculate the regression coefficients. , The final calibration model is obtained. ,in This is the wind speed after secondary correction.

[0030] In the wind speed correction process of step S2, a correction is first performed. Specifically, this involves calculating the average power based on the measured power of the wind turbine within the wind speed range of 4 m / s to 9 m / s, then using the average power curve of the wind turbine to deduce the theoretical average wind speed. Subsequently, a correction factor is calculated and used to correct the original nacelle wind speed. The formula for calculating the correction factor is as follows: ,in This represents a correction factor used to correct the raw cabin wind speed collected by the SCADA system to a wind speed that is closer to the actual wind conditions. The theoretical average wind speed is obtained by inversely calculating the average power curve of the wind turbine and can reflect the state of natural airflow. This represents the SCADA measured average wind speed, which is the raw average wind speed collected by the anemometer at the tail of the wind turbine nacelle within the wind speed range of 4 m / s to 9 m / s.

[0031] In the preferred embodiment, the secondary correction in step S2 has a prerequisite: it is only performed if the correlation coefficient between the wind speed and the reference wind speed is lowered after the first correction. The second correction will only be carried out when the wind speed is corrected, and the second correction is achieved by establishing a regression relationship between the wind speed after the first correction and the reference wind speed.

[0032] Another preferred scheme clarifies the calculation method for the average power in the wind speed range of 4 m / s to 9 m / s. It employs a weighted kernel density estimation method, specifically using the measured power at each wind speed as a sample, constructing a kernel function with wind speed frequency as the weight, and calculating the probability-weighted mean of the power within this range through kernel density integration. The corresponding calculation formula is as follows: . in the formula The average power representing the wind speed range of 4 m / s to 9 m / s is the probability-weighted average of the wind turbine power within this range; The representative wind speed is The measured power corresponding to the wind speed, that is, the power actually generated by the wind turbine under a specific wind speed; The wind speed probability density function obtained based on kernel density estimation is used to describe the probability distribution of different wind speeds occurring within this interval. The Epanechnikov kernel is selected as the kernel function, and the bandwidth of the kernel function is determined by cross-validation to ensure the estimation accuracy of the probability density function.

[0033] The preferred scheme details the steps of secondary calibration based on regression relationships. The first step is to adjust the wind speed after the first correction. Compared with reference wind speed The initial regression coefficients were calculated using the M-estimation method for the sample points. and And a preliminary regression model is established using these two initial coefficients. ,in This represents the wind speed obtained after one correction. This represents the reference wind speed used for calibration. and These represent the intercept coefficient and slope coefficient of the preliminary regression model, respectively. The second step is to calculate the residuals for each sample point. The residual calculation formula is: ,in Representing the The residual for each sample point is the difference between the actual reference wind speed and the predicted reference wind speed calculated using the preliminary regression model. Representing the The actual reference wind speed value for each sample point Representing the Wind speed after one correction for each sample point, followed by a weighting function constructed based on the absolute value of the residuals. ,in Representing the The weights of each sample point are used to adjust the degree of influence of the sample points in subsequent weighted calculations. The median absolute deviation of the residuals is an indicator of the degree of dispersion of the residuals. The third step is to use weights Weight all sample points and recalculate the regression coefficients to obtain... and And based on these two corrected coefficients, a final calibration model is established. ,in This represents the wind speed obtained after a second correction. It is the wind speed that more accurately reflects natural wind conditions after two calibrations. and These represent the intercept coefficient and slope coefficient of the final calibration model, respectively.

[0034] The core of this is the combination of technical scenario adaptability requirements and industry practice verification. When the coefficient is higher than 0.81, the linear correlation between wind speed and reference data such as wind measuring towers after one correction is sufficient to support subsequent calibration. If it is lower than 0.81, it means that there is a systematic bias in the correction model and it needs to be optimized first. Secondly, the technical purpose of matching the secondary correction is to use algorithms such as robust weighted least squares, which requires a high initial correlation to avoid distortion of the calibration results. In similar wind speed correction scenarios in the industry, in order to balance accuracy and efficiency, the high correlation threshold is often set in the range of 0.8-0.85. 0.81 is the optimal value in this range that has been verified by multiple CFD modeling extrapolations, which can ensure that the error is controlled within an acceptable range when the corrected data is used for power generation assessment.

[0035] In the preferred embodiment, in step S3, the CFD computational domain is constructed based on the digital elevation model (DEM) and surface roughness data, and the Navier-Stokes equations are solved to simulate the three-dimensional wind flow field.

[0036] In the preferred scheme, the Navier-Stokes equations are solved using a separate implicit solution algorithm, with the following steps: A1. The computational domain is discretized using the finite volume method. Unstructured tetrahedral meshes are used for the control volume. Boundary layer refinement is applied to the near-ground region. The mesh quality is verified by orthogonality and stretch ratio. A2. The velocity field is initialized based on the logarithmic wind profile, the pressure field is initialized based on the hydrostatic equilibrium, and the turbulence parameters are set based on the initial values ​​of the k-ω SST turbulence model. A3. The time discretization adopts the second-order implicit scheme, the convection term adopts the QUICK scheme, the diffusion term adopts the central difference scheme, and the pressure and velocity coupling adopts the SIMPLEC algorithm. A4. First, solve the momentum prediction equation to obtain the intermediate velocity field. , , Then, the pressure field p' is adjusted using the pressure correction equation, and the velocity field is updated to... , , ; A5. Solve the k-equation and ω-equation of the turbulence equations, update the turbulent viscosity coefficient, and determine whether the calculation residual is less than [the value of the original equation]. If the condition is not met, return to step A4 for iteration until a stable three-dimensional airflow field distribution is obtained.

[0037] In step S3, the CFD computational domain is constructed based on the digital elevation model (DEM) and surface roughness data. The DEM is used to present the topographic relief features within the computational domain, while the surface roughness data is used to reflect the differences in resistance to wind flow caused by different surface types. Together, they provide the CFD model with basic parameters that fit the actual geographical environment. Subsequently, the three-dimensional wind flow field within the computational domain is simulated by solving the Navier-Stokes equations, thereby obtaining the wind flow pattern in space.

[0038] In the preferred scheme, the Navier-Stokes equations are solved using a segregated implicit solution algorithm, with the specific steps as follows: When performing step A1, the CFD computational domain is discretized using the finite volume method, which divides the continuous computational domain into multiple discrete control volumes. Unstructured tetrahedral meshes are used for the control volumes, as these meshes are better suited to complex terrain boundaries. Boundary layer refinement is applied to the near-surface region to increase the number of meshes in this region and improve the simulation accuracy of near-surface wind field details. The mesh quality is then verified using orthogonality and stretch ratio indices to ensure that the mesh meets the accuracy requirements of subsequent calculations. Orthogonality reflects the perpendicularity between the edges of the mesh cells, and stretch ratio reflects the stretching of the mesh cells. Only when these indices are met can the calculation results be guaranteed to be reliable.

[0039] Proceed to step A2 to set the initial fields required for the calculation. The velocity field is initialized based on the logarithmic wind profile, a classic model that describes the variation of near-surface wind speed with altitude and closely matches the actual wind speed distribution within the atmospheric boundary layer. The pressure field is initialized based on hydrostatic equilibrium, which is hydrostatic equilibrium. Setting the initial pressure field according to this equilibrium relationship makes the initial state closer to the actual atmospheric pressure distribution. The turbulence parameters are set based on the initial values ​​of the k-ω SST turbulence model, a commonly used model for simulating turbulent flow, whose initial parameters provide the starting conditions for turbulence calculations.

[0040] When performing step A3, the discretization scheme and coupling algorithm are determined. For time discretization, a second-order implicit scheme is used, which improves the simulation accuracy of time-varying processes while ensuring computational stability. For spatial discretization, the convection term uses the QUICK scheme, a high-precision discretization scheme that effectively reduces numerical diffusion errors in convection term calculations. The diffusion term uses a central difference scheme, which can calculate the spatial changes of the diffusion term with high accuracy. For pressure and velocity coupling, the SIMPLEC algorithm is used. The SIMPLEC algorithm solves the coupling problem between pressure and velocity through a specific pressure correction method, ensuring that both satisfy the basic equations of fluid motion.

[0041] Perform step A4 to complete the preliminary calculation and update of the velocity and pressure fields. First, solve the momentum prediction equation, which is based on the law of conservation of momentum. After solving, the three components of the intermediate velocity field are obtained. , , These three components correspond to the velocity values ​​in the x, y, and z spatial directions without considering pressure correction. Next, the pressure correction term $p'$ is calculated using the pressure correction equation, which adjusts the initial pressure field to satisfy the fluid continuity equation. Finally, the velocity field is updated using the pressure correction term, with the update formulas as follows: , , ,in , , These represent the final velocity components in the x, y, and z directions after the update, respectively. This represents the time step of the computation, i.e., the time interval between two iterations. This represents the density of the fluid (in this case, air). The gradient operator represents the gradient of the pressure correction term $p'$. This formula describes the rate of change of the pressure correction term in space. It allows the influence of the pressure correction to be incorporated into the velocity field, making the velocity field more closely match the pressure field.

[0042] Perform step A5 to complete turbulence-related calculations and convergence checks. First, solve the k-equation and ω-equation in the turbulence equations. The k-equation is used to calculate the turbulent kinetic energy k, and the ω-equation is used to calculate the specific dissipation rate ω. Solving these two equations yields key parameters of the turbulent flow. Then, update the turbulent viscosity coefficient based on the obtained k and ω. The turbulent viscosity coefficient is an important parameter affecting turbulent flow calculations, and its update improves the accuracy of the turbulence simulation. Finally, check if the calculation residual is less than [a certain value]. The residual is an indicator that measures the deviation between the solution to the equation and the actual value. The smaller the residual, the closer the calculated result is to the true solution. If the residual does not meet the requirements, return to step A4 to solve the momentum equation, correct the pressure, and update the velocity again. This process is repeated iteratively until the residual is less than [the required value]. At this point, the calculation can be considered to have converged, and a stable three-dimensional wind flow field distribution can be obtained, which can accurately reflect the wind flow state within the computational domain.

[0043] Example 2 Further explanation in conjunction with Example 1, such as Figure 1 As shown, in the preferred scheme, the step of extrapolating the restored wind speed time series to each turbine location of the planned wind farm using the spatial wind speed transfer function includes: B1. First, determine the spatial correspondence between the reference points of the existing wind farm and the pre-selected turbine sites of the planned wind farm, clarify the corrected wind speed time series dataset of the reference points, and at the same time determine the number and spatial location of all pre-selected turbine sites of the planned wind farm. B2. Next, call the spatial wind speed transfer function obtained through CFD simulation in step S3; B3. Then, for each pre-selected machine location, the wind speed value at each moment in the wind speed time series after the reference point correction is multiplied by the amplification factor corresponding to that location to obtain the preliminary extrapolated wind speed series. B4. Then, introduce a wind condition time series correlation correction factor, multiply the wind speed value at each moment in the preliminary extrapolated wind speed sequence by the correction factor, and complete the time series feature matching correction. B5. Then, based on the technical parameters of the wind turbines to be selected for the planned wind farm, the time values ​​in the corrected wind speed sequence that exceed the cut-in and cut-out wind speed range are adjusted by using the linear interpolation method of the wind speed at adjacent effective time points to ensure that they meet the actual operating wind speed requirements of the wind turbines. B6. Finally, repeat steps B3-B5 above for the calculation process of a single turbine site to complete the wind speed time series calculation for all pre-selected turbine sites in the planned wind farm and generate high-precision wind speed time series prediction data for each turbine site.

[0044] The specific algorithm steps in step S4, which involve extrapolating the restored wind speed time series to each turbine location in the planned wind farm using the spatial wind speed transfer function to generate high-precision wind speed time series prediction data, include: In step S4, the restored wind speed time series is extrapolated to each turbine location of the planned wind farm using the spatial wind speed transfer function, and high-precision wind speed time series prediction data is generated. The specific algorithm steps are as follows: When performing step B1, first determine the spatial correspondence between the reference points of existing wind farms and the pre-selected turbine sites of planned wind farms, clarifying their geographic spatial relationship. Then, establish a corrected wind speed time series dataset at the reference points. ,in Represents the time dimension, with values ​​ranging from 1 to... , This refers to the total number of samples in the time series, and the time interval of this dataset must be consistent with the original SCADA data acquisition interval. Outliers that still deviate from three standard deviations after correction must be removed to ensure the continuity and validity of the series. Furthermore, the set of pre-selected turbine locations for planned wind farms is clearly defined as follows: , This represents the total number of pre-selected machine sites.

[0045] Proceed to step B2, and call the spatial wind speed transfer function obtained in step S3. This function specifically uses the wind speed amplification coefficient matrix. Presented in the form of a matrix with dimensions of The first in the matrix element Corresponding pre-selected machine sites Wind speed amplification factor between reference point and reference point The value range is 1 to ,and Through CFD simulation The ratio of the steady-state wind speed at one location to the steady-state wind speed at the reference point is calculated and used to quantify the wind speed transmission relationship between the two locations.

[0046] When carrying out step B3, for each pre-selected site Based on its corresponding magnification factor Time series of wind speeds at reference point According to the formula The preliminary extrapolated wind speed sequence was calculated. ,in represent The preliminary extrapolated wind speed sequence for the location is calculated by multiplying the wind speed at each moment at the reference point by an amplification factor, thus representing the wind speed amplitude from the reference point to... The basic spatial mapping of the site.

[0047] When performing step B4, a wind condition time series correlation correction factor is introduced. To match the temporal variation characteristics of wind conditions, the specific operation is divided into three steps: First, the wind speed sequence after correction at the reference point is... and CFD simulation wind speed sequence A 10-minute moving average was applied to obtain the average wind speed sequence. and ,in yes The sequence after moving average yes The sequence after moving average; the second step, using the formula , The reference point and the calculation are obtained Wind speed pulsation component at the location , , This represents the fluctuation portion of the wind speed sequence at the reference point that deviates from its average wind speed. represent The third step is to determine the fluctuations in the CFD-simulated wind speed sequence that deviate from its average wind speed; Sure ,in The function representing the cross-correlation coefficient is used to calculate... and The correlation obtained This is the wind condition time series correlation correction factor, after which the initial extrapolated wind speed series is corrected to... , Represents the result after time-series correlation correction Site wind speed sequence.

[0048] When performing step B5, the corrected wind speed sequence is... Boundary constraint processing: First, determine the cut-in and cut-out wind velocities of the wind turbines to be selected for the planned wind farm based on their technical parameters. If the model of the wind turbine to be selected is not specified, then the industry-standard values ​​for the same type of wind turbine are adopted, with the cut-in wind velocity ranging from 3m / s to 4m / s and the cut-out wind velocity ranging from 25m / s to 28m / s; if At a certain moment If the wind speed value is lower than the cut-in wind speed or higher than the cut-out wind speed, the wind speed value at that moment is adjusted by linear interpolation of the wind speed at the adjacent valid moment to ensure that the corrected wind speed sequence meets the wind speed range requirements of the actual operation of the wind turbine.

[0049] Finally, perform step B6, repeating steps B3 to B5, sequentially checking all pre-selected turbine sites for the planned wind farm. Wind speed time series calculations are performed, ultimately generating high-precision wind speed time series prediction data for each machine location. ,in represent High-precision time-series wind speed prediction data for the location. The value range is 1 to It covers all sample moments of the entire time series.

[0050] Example 3 Further explanation in conjunction with Example 1, such as... Figure 1 The structure shown is a wind farm wind speed time series prediction system, including: The data acquisition and preprocessing module is used to acquire and clean SCADA data and reference wind speed data; The data correction and restoration module is used to build and apply the wind speed correction model to generate a restored wind speed sequence. The CFD modeling and extrapolation module is used to establish the computational domain, run CFD simulations, and generate spatial transfer functions. The database and output module is used to store intermediate results and output wind speed time series prediction data for each turbine location in the planned wind farm.

[0051] The wind farm wind speed time series prediction system comprises four core functional modules. The functions and collaborative logic of each module are as follows: The core role of the data acquisition and preprocessing module is to acquire and clean two types of key data: one is SCADA data of existing wind farms, and the other is synchronous reference wind speed data. After acquiring the data, the raw data is processed into a regular and effective dataset through cleaning operations such as time synchronization, invalid value removal, outlier filtering, and missing data interpolation, providing high-quality data input for subsequent modules. The data correction and restoration module constructs a SCADA wind speed correction model based on the regularized dataset output by the data acquisition and preprocessing module. This model includes primary and secondary correction processes. By calculating the average power curve, theoretical average wind speed, correction coefficient, and regression calibration, the SCADA wind speed data with inherent biases is corrected into a restored wind speed sequence that reflects natural wind conditions, providing reliable wind speed input for CFD modeling. The CFD modeling and extrapolation module, based on the restored wind speed sequence generated by the data correction and restoration module, first constructs a CFD calculation domain covering the areas of existing and planned wind farms. It then imports digital elevation models and surface roughness data, and uses the restored wind speed sequence as time-varying boundary conditions to run CFD simulations. Solving the Navier-Stokes equations yields a three-dimensional wind flow field, which in turn generates a spatial transfer function reflecting the correlation between wind speeds at reference points in existing wind farms and at various turbine locations in planned wind farms. The database and output module stores intermediate results generated during the operation of each module, including preprocessed datasets, coefficients and model parameters from the correction process, and intermediate calculation data from the CFD simulation. It also applies the spatial transfer function generated by the CFD modeling and extrapolation module to the restored wind speed sequence, obtaining and outputting the predicted wind speed time series data for each turbine location in the planned wind farms.

[0052] Through the coordinated operation of four modules, a complete technical chain is formed from data acquisition to prediction result output: the data acquisition and preprocessing module ensures the validity of basic data and avoids interference from invalid or abnormal data in subsequent processes; the data correction and restoration module eliminates the inherent bias of SCADA data and solves the problem of insufficient accuracy when SCADA data is used alone, providing reliable data for subsequent modeling; the CFD modeling and extrapolation calculation module, with the help of high-precision modeling and simulation technology, realizes the spatial extrapolation of wind speed data from existing wind farms to planned wind farms, overcoming the limitation of traditional single-point measurement in covering complex areas; the database and output module facilitates process traceability and problem investigation by storing intermediate results, while accurately outputting prediction data for each turbine location in the planned wind farm, providing direct support for power generation assessment, economic analysis and layout optimization in the early planning of wind farms. The overall system does not rely on high-cost, long-cycle wind measurement towers, which greatly reduces the cost of wind resource assessment and improves planning efficiency and accuracy.

[0053] Example 4 Further explanation in conjunction with Example 1, such as Figure 1 This paper takes an offshore wind power project B in XX region as an example to introduce the specific implementation steps and methods of the present invention in detail. There is an existing wind farm A to the west of the project site, which has been in operation for more than a year, and SCADA data of the wind turbines in the existing site has been collected.

[0054] Data Acquisition: Collect SCADA data (at 10-minute intervals) for wind farm A over the past year, including time series, wind speed, wind direction, power, and temperature for each wind turbine. Simultaneously, obtain anemometer tower or wind power prediction data for the same period and region as reference data.

[0055] Data Correction: Time alignment and organization of SCADA and reference data were performed. Abnormal data segments with power values ​​of 0 or greater than rated power were removed. The data was divided into sectors by wind direction (16 sectors) and wind speed (each sector is 2 m / s). A correction model was constructed using SCADA-measured wind speed, power, and wind direction as features, and wind speed values ​​from the reference data as the target. The validated model was applied to correct all historical SCADA wind speeds for wind field A, resulting in a restored 1-year 10-minute wind speed time series.

[0056] CFD extrapolation: Construct a CFD computational domain with a radius of 20km, covering the A and B fields, and import the 30m resolution DEM and the surface roughness map retrieved from satellite remote sensing.

[0057] A wind turbine at the center of field A is selected as the reference point. The corrected wind speed time series obtained in step 2 is used as the time-varying boundary condition at the CFD model inlet. Using commercial CFD software (Meteodyn WT6.5), the wind speed amplification factor matrix (spatial transfer function) from the reference point in field A to each pre-selected turbine location in field B is calculated. Using this matrix, the 1-year wind speed time series of the reference point is extrapolated to each turbine location in field B, generating the 1-year wind speed prediction series corresponding to each point.

[0058] Application: Input the predicted wind speed sequence of each turbine location in Field B into industry-standard power generation calculation software (such as WAsP or Openwind) to calculate the annual power generation (AEP) of the planned wind farm, and complete the investment return analysis accordingly.

[0059] This project, through the method of this invention, completed the wind resource assessment work that would have taken more than a year to complete in one week without erecting a wind measurement tower. The assessment results are highly consistent with the actual operation data after construction.

[0060] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A wind farm wind speed time series prediction method based on SCADA data correction and CFD modeling, characterized by: The method includes: S1. Collect historical SCADA data and concurrent reference wind speed data of existing wind farms and perform preprocessing. S2. Based on the preprocessed data, establish a SCADA wind speed correction model, perform first and second corrections on the SCADA wind speed, and generate the restored hub height wind speed time series. S3. Construct a CFD computational domain covering the areas of existing and planned wind farms, use the restored wind speed time series as time-varying boundary conditions, and obtain the spatial wind speed transfer function through CFD simulation. S4. Using the spatial wind speed transfer function, the restored wind speed time series is extrapolated to each turbine location in the planned wind farm to generate high-precision wind speed time series prediction data.

2. The wind speed time series prediction method for wind farms based on SCADA data correction and CFD modeling as described in claim 1, characterized in that: In step S1, the SCADA data includes time period, hub height wind speed, wind direction, power, temperature and air pressure data; The reference wind speed data for the same period comes from the wind measurement tower or wind power prediction system; Preprocessing includes time synchronization, invalid value removal, outlier filtering, and missing data imputation.

3. The wind speed time series prediction method for wind farms based on SCADA data correction and CFD modeling as described in claim 1, characterized in that: In step S2, one correction includes: The average power is calculated based on the measured power of the wind turbine in the wind speed range of 4 m / s to 9 m / s. The theoretical average wind speed is then inferred from the average power curve of the wind turbine, and a first-order correction factor is calculated. Correct the original cabin wind speed; in, The theoretical average wind speed, This is the average wind speed measured by SCADA.

4. The wind speed time series prediction method for wind farms based on SCADA data correction and CFD modeling as described in claim 3, characterized in that: In step S2, the secondary correction only affects the correlation coefficient between the wind speed and the reference wind speed after the primary correction. It is executed in time and calibrated based on the regression relationship between the two.

5. The wind speed time series prediction method for wind farms based on SCADA data correction and CFD modeling as described in claim 3, characterized in that: When calculating the average power in the wind speed range of 4 m / s to 9 m / s, a weighted kernel density estimation method is used: The measured power at each wind speed is used as a sample, and a kernel function is constructed with wind speed frequency as the weight. The probability-weighted mean of the power within the range is calculated by integrating the kernel density. The formula is as follows: ； in The power is the interval average. This represents the measured power corresponding to wind speed v. The wind speed probability density function is based on kernel density estimation. The Epanechnikov kernel is selected as the kernel function, and the bandwidth is determined by cross-validation.

6. The wind speed time series prediction method for wind farms based on SCADA data correction and CFD modeling as described in claim 4, characterized in that: The steps for calibration based on the regression relationship between the two include: Wind speed after one correction Compared with reference wind speed The initial regression coefficients were calculated using the M-estimation method for the sample points. , A preliminary regression model was obtained. ; Calculate the residuals for each sample point Construct a weight function based on the absolute value of the residuals. , where s is the median absolute deviation of the residuals; By weight Weight the samples and recalculate the regression coefficients. , The final calibration model is obtained. ,in This is the wind speed after secondary correction.

7. The wind speed time series prediction method for wind farms based on SCADA data correction and CFD modeling as described in claim 1, characterized in that: In step S3, the CFD computational domain is constructed based on the digital elevation model (DEM) and surface roughness data, and the Navier-Stokes equations are solved to simulate the three-dimensional wind flow field.

8. The wind speed time series prediction method for wind farms based on SCADA data correction and CFD modeling as described in claim 7, characterized in that: The Navier-Stokes equations are solved using a separate implicit solution algorithm, with the following steps: A1. The computational domain is discretized using the finite volume method. Unstructured tetrahedral meshes are used for the control volume. Boundary layer refinement is applied to the near-ground region. The mesh quality is verified by orthogonality and stretch ratio. A2. The velocity field is initialized based on the logarithmic wind profile, the pressure field is initialized based on the hydrostatic equilibrium, and the turbulence parameters are set based on the initial values ​​of the k-ω SST turbulence model. A3. The time discretization adopts the second-order implicit scheme, the convection term adopts the QUICK scheme, the diffusion term adopts the central difference scheme, and the pressure and velocity coupling adopts the SIMPLEC algorithm. A4. First, solve the momentum prediction equation to obtain the intermediate velocity field. , , Then, the pressure field p' is adjusted using the pressure correction equation, and the velocity field is updated to... , , ; A5. Solve the k-equation and ω-equation of the turbulence equations, update the turbulent viscosity coefficient, and determine whether the calculation residual is less than [the value of the original equation]. If the condition is not met, return to step A4 for iteration until a stable three-dimensional airflow field distribution is obtained.

9. The wind speed time series prediction method for wind farms based on SCADA data correction and CFD modeling as described in claim 1, characterized in that: The steps for extrapolating the restored wind speed time series to each turbine location in the planned wind farm using the spatial wind speed transfer function include: B1. First, determine the spatial correspondence between the reference points of the existing wind farm and the pre-selected turbine sites of the planned wind farm, clarify the corrected wind speed time series dataset of the reference points, and at the same time determine the number and spatial location of all pre-selected turbine sites of the planned wind farm. B2. Next, call the spatial wind speed transfer function obtained through CFD simulation in step S3; B3. Then, for each pre-selected machine location, the wind speed value at each moment in the wind speed time series after the reference point correction is multiplied by the amplification factor corresponding to that location to obtain the preliminary extrapolated wind speed series. B4. Then, introduce a wind condition time series correlation correction factor, multiply the wind speed value at each moment in the preliminary extrapolated wind speed sequence by the correction factor, and complete the time series feature matching correction. B5. Then, based on the technical parameters of the wind turbines to be selected for the planned wind farm, the time values ​​in the corrected wind speed sequence that exceed the cut-in and cut-out wind speed ranges are adjusted by using the linear interpolation method of the wind speed at adjacent effective time points to ensure that they meet the actual operating wind speed requirements of the wind turbines. B6. Finally, repeat steps B3-B5 above for the calculation process of a single turbine site to complete the wind speed time series calculation for all pre-selected turbine sites in the planned wind farm and generate high-precision wind speed time series prediction data for each turbine site.

10. A wind farm wind speed time series prediction system implementing the method of any one of claims 1-9, characterized in that, include: The data acquisition and preprocessing module is used to acquire and clean SCADA data and reference wind speed data; The data correction and restoration module is used to build and apply the wind speed correction model to generate a restored wind speed sequence. The CFD modeling and extrapolation module is used to establish the computational domain, run CFD simulations, and generate spatial transfer functions. The database and output module is used to store intermediate results and output wind speed time series prediction data for each turbine location in the planned wind farm.

Images