Complex terrain wind power plant power prediction method based on CFD flow field pre-calculation and machine learning

Abstract

The invention discloses a complex terrain wind power plant power prediction method based on CFD (computational fluid dynamics) flow field pre-calculation and machine learning. The method comprises the following steps: acquiring elevation information of a wind power plant; fan parameter information is obtained; obtaining wind measurement data of the wind power plant; cleaning the wind measurement data; fitting a terrain curved surface; modeling different wind directions of the wind power plant; establishing a rotary actuating disc model for fan wake flow simulation; establishing a wind power plant three-dimensional numerical calculation model; performing data discretization processing; simulating different wind speeds and directions of the wind power plant; wind speed and wind direction information at the position of each anemometer tower under each working condition; wind speed and wind direction information at the height of a hub of each fan; dividing the target wind power plant into a plurality of sub-regions; obtaining the total power of each sub-region; establishing a grouping power prediction database; and power prediction is realized. The method can effectively improve the accuracy of power prediction of the wind power plant in the complex terrain, is suitable for wind power plants which are newly built and run for a short time, and can effectively improve the economic benefits of the wind power plant in the complex terrain.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a method for predicting wind farm power in complex terrain based on CFD flow field pre-calculation and machine learning. Background Art

[0002] With the rapid development of the global economy and the continuous improvement of people's living standards, total energy consumption has increased dramatically, and problems such as the shortage of traditional fossil energy and environmental pollution have become increasingly prominent. To further promote sustainable development strategies and promote energy transformation, humanity must find alternatives to traditional fossil energy. Wind energy stands out due to its clean, renewable, and abundant resources. Wind power generation, as the primary method of utilizing wind energy, works by converting wind energy into electricity through the rotation of rotor blades, which is then connected to the power grid for users.

[0003] Wind power generation capacity is significantly affected by meteorological conditions and the geographical environment. Due to the fluctuating and random nature of wind speed, wind power output exhibits significant intermittent, random, and uncertain characteristics. Fast and accurate wind power forecasting can effectively mitigate the negative impact of unstable wind power output on the power grid, playing a crucial role in ensuring the safe and stable operation of the power grid and increasing wind farm revenue.

[0004] Wind farms are the main places for wind power generation. They are generally equipped with multiple wind turbines and wind towers. For wind farms in complex terrain, they are affected by terrain effects, which will produce obvious flow separation, exacerbating the intermittent, random and uncertain nature of wind power output, making it more difficult to accurately predict wind power in the entire wind farm.

[0005] Existing methods for predicting wind power at wind farms can be categorized by their principles: traditional statistical methods, artificial intelligence-based methods, and physical methods. Traditional statistical methods suffer from low prediction accuracy when applied to time-series data with strong nonlinearities. With the rapid development of computer technology, artificial intelligence (AI), with its powerful nonlinear modeling capabilities, has attracted widespread attention in the field of wind power prediction. AI-based methods primarily learn from historical data to map input and output data, creating a prediction model. Future data is then fed into this prediction model to predict wind farm power. However, both traditional statistical and AI-based methods require extensive historical data and are unable to accurately predict power for newly built or recently operating wind farms. The core of physical methods is the mesoscale NWP model. Under given initial and boundary conditions, a system of differential equations describing atmospheric motion is solved to obtain key information such as wind speed and direction near the wind farm. The wind farm's generated power is then derived from the wind turbine's power curve. The large grid size of mesoscale models makes it difficult to capture specific flow details, resulting in low prediction accuracy, especially for wind farms located in complex terrain. Summary of the Invention

[0006] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is: how to provide a complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning that has high prediction accuracy and is suitable for newly built wind farms and wind farms with a short operating time.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] The power prediction method for wind farms in complex terrain based on CFD flow field pre-calculation and machine learning includes the following steps:

[0009] Step 1) obtaining elevation information of the target terrain wind farm;

[0010] Step 2) obtaining parameter information of wind turbines in the wind farm;

[0011] Step 3) obtaining wind measurement data of the wind farm;

[0012] Step 4) cleaning the wind farm wind measurement data obtained in step 3);

[0013] Step 5) fitting the terrain surface based on the target terrain wind farm elevation information obtained in step 1);

[0014] Step 6) Modeling the wind farm under different wind directions;

[0015] Step 7) Based on the wind turbine parameter information obtained in step 2) and blade element-momentum theory, a rotating actuator disk model for wind turbine wake simulation is established;

[0016] Step 8) establishing a three-dimensional numerical calculation model of the target terrain wind farm based on the data of steps 5), 6) and 7);

[0017] Step 9) discretizing the wind speed and direction data of the target terrain wind farm;

[0018] Step 10) obtaining wind speed and direction information at each wind tower location in the wind farm under each operating condition considering the wake effect through CFD simulation;

[0019] Step 11) obtaining wind speed and direction information at the hub height of each wind turbine in the wind farm taking into account the wake effect through CFD simulation;

[0020] Step 12) Divide the target wind farm into several sub-areas;

[0021] Step 13) obtaining the power of each wind turbine and the total power of each sub-area in the wind farm according to the factory power curve of each wind turbine in the wind farm;

[0022] Step 14) using the wind speed and direction at the wind tower location in the CFD simulation in step 10) as input and the total power of each sub-area of ​​the wind farm as output, to establish a grouped power prediction database;

[0023] Step 15) Input the wind speed and direction at the wind tower location at the prediction time, and obtain the total power of the wind farm through the back propagation neural network model to achieve power prediction.

[0024] Preferably, in step 1), the original elevation data of the target terrain wind farm is obtained, the outliers and invalid values ​​of the original elevation data are filtered, and the missing areas after data filtering are filled by interpolation method. At the same time, the truncated boundary of the target terrain wind farm is smoothed by using the square sine function, and a transition section is set around the target terrain wind farm.

[0025] Preferably, the parameter information of the fan in step 2) includes: the hub height of the fan, the rotor diameter of the fan, the thrust coefficient curve of the fan and the factory power curve of the fan.

[0026] Preferably, in step 3), the wind measurement data of the wind farm includes: wind speed, wind direction and power generation data at the height of the wind turbine hub, the longitude and latitude positions of the wind turbine, and the altitude of the wind turbine; wind speed and wind direction data at different heights of the wind measurement tower, the longitude and latitude positions of the wind measurement tower, and the altitude of the wind measurement tower.

[0027] Preferably, the method for cleaning the wind farm wind measurement data in step 4) is as follows: at a resolution of 1 minute, discontinuous time data is considered as abnormal data; at a resolution of 1 minute, wind speed data is set within a range of 3 to 20 m / s, and data outside this range is considered as abnormal data; at a resolution of 1 minute, at a set height, data with a wind direction angle difference greater than 22.5° is considered as abnormal data;

[0028] At a resolution of 15 minutes, when the rate of change of adjacent wind speed values ​​is greater than 20%, the latter wind speed data is considered abnormal data; at a resolution of 15 minutes, when the difference between the wind speed at a height of 70 meters and the wind speed of the first wind turbine at the same moment is greater than 10 meters per second, the data is considered abnormal data; at a resolution of 15 minutes, when the difference between the wind speed at a height of 70 meters and the wind speed of the second wind turbine at the same moment is greater than 10 meters per second, the data is considered abnormal data, where the first wind turbine and the second wind turbine are the two wind turbines closest to the two sides of the wind measurement tower respectively.

[0029] Preferably, in step 5), based on the target terrain wind farm elevation information obtained in step 1), the terrain surface is fitted, wherein the span numbers of the terrain surface in the U and V directions are both set to 110, and the hardness is set to 5; and based on the fitted terrain surface, a three-dimensional numerical model of the target terrain is obtained by outward expansion, wherein the outward expansion distance is set to 5-10 times the height of the target terrain.

[0030] Preferably, the method for simulating different wind directions in the wind farm in step 6) is: based on the rotating area modeling method, in the geometric modeling process, a cylindrical rotating area is established in the core area of ​​the target terrain wind farm, and the cylindrical rotating area is meshed. After the meshing is completed, the cylindrical rotating area is rotated to realize simulation under different wind directions.

[0031] Preferably, in step 9), the method for discretizing the wind speed and direction data of the target terrain wind farm is as follows: dividing the wind direction of 0 to 360° into multiple fan-shaped wind direction areas; dividing the wind speed into two wind speed areas, and discretizing each wind speed area according to different intervals, and then combining each wind speed and wind direction area to correspond to a wind condition, thereby completing the discretization of the wind speed and direction data of the target terrain wind farm.

[0032] Preferably, in step 12), based on the results of the clustering algorithm, the target terrain wind farm is divided into several sub-areas. The specific method is as follows: obtain the following four characteristic parameters of each wind turbine: the average annual wind speed U at the height of the wind turbine hub; mean , the average power value of the whole year P mean, the horizontal coordinate X of the fan position, the vertical coordinate Y of the fan position, and the four characteristic parameters of each fan are normalized; the normalized characteristic parameters are used as input, and the K-means++ algorithm is used for clustering grouping, and the silhouette coefficient and the sum of squares of the intra-cluster distance are used as judgment indicators for different grouping schemes.

[0033] Preferably, the method further includes step 16), wherein the power prediction accuracy C is calculated according to the following formula: R :

[0034]

[0035] Where: n is the total number of time periods within the error statistical time interval minus the number of time periods exempted from assessment; P Pi is the predicted average power in period i; P Mi is the actual average power in period i; C i is the total startup capacity in time period i.

[0036] Compared to existing technologies, this invention develops a power prediction method for wind farms in complex terrain. This method preempts time-consuming CFD flow field calculations and establishes a prediction database for the relationship between incoming wind speed, wind direction, and output power. First, the wind turbines in the wind farm are clustered using the K-means++ clustering algorithm. Then, a backpropagation artificial neural network is used to extract the logical relationship between incoming wind speed, wind direction, and output power in the prediction database. This method can effectively improve the accuracy of power predictions for wind farms in complex terrain and is suitable for newly built wind farms and those with relatively short operating times, effectively enhancing the economic benefits of wind farms in complex terrain. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Attachment Figure 1 This is a flow chart of the complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning of the present invention;

[0038] Attachment Figure 2 It is a three-dimensional visualization cloud map of the time-averaged downstream velocity at the hub height of the target terrain wind farm;

[0039] Attachment Figure 3 Schematic diagram of wind turbine layout of a target wind farm in the first embodiment of the present invention;

[0040] Attachment Figure 4 This is a comparison result diagram of the power prediction data and the measured data of the target terrain wind farm sub-area I in Example 1 of the present invention;

[0041] Attachment Figure 5 This is a schematic diagram of the average accuracy of wind farm power prediction in the first embodiment of the present invention;

[0042] Attachment Figure 6Schematic diagram of the comparison results of the accuracy of different power prediction methods in Example 1 of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0044] In this specific embodiment, a method for predicting wind farm power in complex terrain based on CFD flow field pre-calculation and machine learning is provided, as shown in the attached figure. Figure 1 As shown, the following steps are included:

[0045] Step 1) Obtain the elevation information of the target terrain wind farm. Specifically, obtain the original elevation data of the target terrain wind farm. The source of the original elevation data can be the terrain elevation map data in dwg format drawn by the surveying and mapping unit in the early stage of wind farm design or the terrain elevation data obtained by satellite remote sensing data (such as SRTM), with a resolution of not less than 30m×30m. Filter outliers and invalid values ​​of the original elevation data, and fill in the missing areas after data filtering by interpolation method. In order to avoid the significant impact of truncated terrain on the flow field, the square sine function is used to smooth the truncated boundary of the target terrain wind farm, and a transition section is set around the target terrain wind farm.

[0046] Step 2) Obtain parameter information of the wind turbines in the wind farm, which mainly includes: the hub height of the wind turbine, the rotor diameter of the wind turbine, the thrust coefficient curve of the wind turbine, and the factory power curve of the wind turbine.

[0047] Step 3) Obtain wind farm wind data. The wind farm wind data primarily includes: wind speed, wind direction, and power generation data at the turbine hub height, the longitude and latitude of the wind turbine, and the wind turbine's altitude, as recorded by the SCADA system; wind speed and direction data at different heights of a wind tower (10 m, 30 m, 50 m, and 70 m in this embodiment), the longitude and latitude of the wind tower, and the altitude of the wind tower.

[0048] Step 4) Clean the wind farm wind measurement data obtained in step 3). During the actual operation of the wind farm, the SCADA system and the wind tower will be subject to various interferences, resulting in missing or invalid abnormal data. Therefore, the wind farm wind measurement data must be cleaned. The cleaning method is: at a resolution of 1 minute, the time-discontinuous data is regarded as abnormal data; at a resolution of 1 minute, the wind speed data is set within the range of 3 to 20 m / s, and the wind speed data outside this range is regarded as abnormal data; at a resolution of 1 minute, at the set heights of 30m, 50m, and 70m, the data with a wind direction angle difference greater than 22.5° is regarded as abnormal data;

[0049] At a resolution of 15 minutes, when the rate of change of adjacent wind speed values ​​is greater than 20%, the latter wind speed data is considered abnormal data; at a resolution of 15 minutes, data where the wind speed at a height of 70 meters differs from the wind speed of the first wind turbine at the same moment by more than 10 meters per second is considered abnormal data; at a resolution of 15 minutes, data where the wind speed at a height of 70 meters differs from the wind speed of the second wind turbine at the same moment by more than 10 meters per second is considered abnormal data. The first wind turbine and the second wind turbine are the two wind turbines closest to the two sides of the wind tower, respectively. In this specific embodiment, the first wind turbine is WT6 and the second wind turbine is WT7.

[0050] Step 5) Fitting a terrain surface based on the target terrain wind farm elevation information obtained in step 1). Specifically, fitting the terrain surface based on the target terrain wind farm elevation information obtained in step 1) is performed, wherein the span number of the terrain surface in the U and V directions is set to 110 and the hardness is set to 5. A three-dimensional numerical model of the target terrain is obtained by outwardly expanding the fitted terrain surface, wherein the outward expansion distance is set to 5-10 times the height of the target terrain.

[0051] Step 6) Modeling of wind farms in different wind directions. Traditional methods rotate the original elevation data according to the wind direction angle to obtain different coordinates and model them separately. Traditional methods involve a lot of repetitive work and are time-consuming and labor-intensive. Here, a method based on rotating region modeling is adopted. During the geometric modeling process, a cylindrical rotating region is established in the core area of ​​the target terrain wind farm, and the cylindrical rotating region is meshed. After the meshing is completed, the cylindrical rotating region is rotated to achieve simulation under different wind directions.

[0052] Step 7) Based on the wind turbine parameter information obtained in step 2) and the blade element-momentum theory, a rotating actuator disk model for wind turbine wake simulation is established.

[0053] Step 8) Based on the data from Step 5), Step 6) and Step 7), a three-dimensional numerical calculation model of the target terrain wind farm is established. The RANS simulation method and the Realizable k-ε turbulence model are used to perform steady-state simulation of the complex terrain wind farm; the second-order upwind scheme is used to discretize the pressure, momentum, turbulent kinetic energy and turbulent dissipation rate. Figure 2 Shown is a three-dimensional visualization cloud map of the downstream time-averaged velocity at the hub height of the wind farm on the target terrain.

[0054] Step 9) Discretize the wind speed and direction data for the target terrain wind farm. Divide the wind direction from 0 to 360 degrees into multiple fan-shaped wind direction regions; divide the wind speed into two wind speed regions, and discretize each wind speed region at different intervals. Then, combine each wind speed and wind direction region to correspond to a wind condition, completing the discretization of the wind speed and direction data for the target terrain wind farm.

[0055] Specifically, the wind direction of 0 to 360 degrees is divided into 16 sectors, each sector is 22.5 degrees. The wind speed is discretized in the area that contributes most to the wind turbine's power generation. According to the distribution characteristics of the wind turbine power curve obtained in step 2), it is divided into two wind speed regions of 1-11m / s and 11-19m / s. The wind speed region of 1-11m / s is discretized at intervals of 2m / s, and the wind speed region of 11-19m / s is discretized at intervals of 4m / s, that is, the discretization is into 8 wind speed values ​​of 1, 3, 5, 7, 9, 11, 15 and 19m / s. Each combination of wind speed and wind direction corresponds to a wind condition, and there are 128 wind conditions in total.

[0056] The inlet velocity is expressed as an exponential rate:

[0057] U=U ref (z / z g ) a

[0058] Where U ref is the distance from the ground surface g Average wind speed at height, z g =350m; α is the surface roughness index, α = 0.15. By changing U ref to simulate different wind speed conditions.

[0059] The turbulent kinetic energy and turbulent dissipation rate at the inlet are expressed as:

[0060]

[0061] Where u * is the friction velocity, u * =0.6m / s;h g is the atmospheric boundary layer height, hg =891.82m; z0 is the surface roughness, z0=0.03; κ is the Karman constant, κ=0.4.

[0062] Step 10) The wind speed and direction information at each wind tower position in the wind farm under each working condition considering the wake effect is obtained through CFD simulation.

[0063] Step 11) The wind speed and direction information at the hub height of each wind turbine in the wind farm is obtained by CFD simulation considering the wake effect.

[0064] Step 12) Divide the target wind farm into several sub-areas. Specifically, based on the results of the K-means++ clustering algorithm, the target terrain wind farm is divided into several sub-areas, wherein the K-means++ clustering algorithm is to minimize the distance between the samples in the same cluster and the cluster center by calculating the Euclidean distance; two n-dimensional space points (x1, x2, ..., x n ) and (y1,y2,…,y n ) can be expressed as:

[0065]

[0066] Specifically: Get the following four characteristic parameters of each wind turbine: the average wind speed U measured at the hub height throughout the year mean , the average power value of the whole year P mean , the horizontal coordinate X of the fan position, the vertical coordinate Y of the fan position, and normalize the four characteristic parameters of each fan.

[0067] The normalized 13×4 dimensional feature parameters are used as input, and the K-means++ algorithm is used for clustering. The silhouette coefficient and the sum of squared distances within the cluster are used as the judgment indicators for different grouping schemes. The silhouette coefficient and the sum of squared distances within the cluster can be expressed as:

[0068]

[0069] Where a(i) is the average distance between sample point i and all other sample points in the same cluster; b(i) is the minimum average distance between sample point i and all sample points in other clusters. The silhouette coefficient (Sico) is a key indicator for describing the differences between inside and outside a cluster. As shown in the above formula, Sico ranges from -1 to 1. The closer Sico is to 1, the better the clustering effect, while the closer it is to -1, the worse the clustering effect.

[0070]

[0071] Where m is the number of samples in each cluster; k is the number of samples in each cluster; jThe jth cluster, j = 1, 2, ..., k. A larger value of k results in more clusters and a smaller sum of squared distances (CSS) between each sample point within the cluster and the cluster center. The value of k is determined by determining the inflection point of the curve where the CSS decreases as k increases. Here, the optimal number of clusters is k = 3.

[0072] Step 13) The power of each wind turbine and the total power of each sub-area in the wind farm are obtained according to the factory power curve of each wind turbine in the wind farm.

[0073] In step 14), the wind speed and direction at the wind tower position in the CFD simulation in step 10) are used as input and the total power of each sub-area of ​​the wind farm is used as output to establish a group power prediction database.

[0074] Step 15) Input the wind speed and direction at the wind tower location at the prediction time, and obtain the total power of the wind farm through the back propagation neural network model (BPANN model) to achieve power prediction.

[0075] BPANN needs to predict the total power of the wind farm by inputting the wind speed and direction at the wind tower. Therefore, the input layer has two neurons and the output layer has one neuron. Based on the number of neurons in the input and output layers and after multiple tests, the network structure with one hidden layer and one hidden neuron was finally determined. The specific training parameters are as follows: the learning rate is set to 0.01, the learning rate decay factor is 0.7, the learning rate growth factor is 1.03, the maximum number of iterations is 2000, and the minimum gradient requirement is 1×10 -10 .

[0076] Step 16) In order to evaluate the accuracy of the proposed power prediction method, the prediction accuracy recommended in the specification is used for comparison. Specifically, the power prediction accuracy C is calculated according to the following formula: R :

[0077]

[0078] Where: n is the total number of time periods within the error statistical time interval minus the number of time periods exempted from assessment; P Pi is the predicted average power in period i; P Mi is the actual average power in period i; C i is the total startup capacity in time period i.

[0079] Example 1:

[0080] The method of the present invention is tested below using a specific wind farm as an example:

[0081] As attached Figure 3 Shown is a diagram of the arrangement of wind turbines in a wind farm in this specific embodiment.

[0082] The power prediction method for complex terrain wind farms proposed in this invention is used to predict the power of wind farms. The prediction results are compared with the measured data of the wind farm to verify the prediction accuracy of this method. When the time resolution is 15 minutes, the comparison results of the power prediction data of the complex terrain wind farm sub-area I and the measured data are shown in the attached figure. Figure 4 Due to space limitations, only the comparison results for sub-region I are given. The average accuracy results for the entire wind farm are shown in the attached Figure 5 It can be found that the proposed complex terrain wind farm power prediction method has a high accuracy rate, with an annual average accuracy rate of 90.10%, which is more than 7% higher than the regulatory requirements.

[0083] In order to demonstrate the advancedness of the proposed CFD-K-means++-BPANN power prediction method, Figure 6 The comparison results of the accuracy of different power prediction methods are given. It can be found that compared with the interpolation method, the accuracy of the proposed power prediction method is improved by about 10.78%, and compared with the non-clustering method, the accuracy of the proposed power prediction method is improved by about 4.50%.

[0084] Compared to existing technologies, this invention develops a power prediction method for wind farms in complex terrain. This method preempts time-consuming CFD flow field calculations and establishes a prediction database for the relationship between incoming wind speed, wind direction, and output power. First, the wind turbines in the wind farm are clustered using the K-means++ clustering algorithm. Then, a backpropagation artificial neural network is used to extract the logical relationship between incoming wind speed, wind direction, and output power in the prediction database. This method can effectively improve the accuracy of power predictions for wind farms in complex terrain and is suitable for newly built wind farms and those with relatively short operating times, effectively enhancing the economic benefits of wind farms in complex terrain.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the technical solutions. Those skilled in the art should understand that modifications or equivalent replacements of the technical solutions of the present invention that do not depart from the purpose and scope of the technical solutions of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A wind farm power prediction method for complex terrain based on CFD flow field pre-calculation and machine learning, characterized by: The following steps are involved: Step 1) obtaining elevation information of the target terrain wind farm; Step 2) obtaining parameter information of wind turbines in the wind farm; Step 3) obtaining wind measurement data of the wind farm; Step 4) cleaning the wind farm wind measurement data obtained in step 3); Step 5) fitting the terrain surface based on the target terrain wind farm elevation information obtained in step 1); Step 6) Modeling the wind farm under different wind directions; Step 7) Based on the wind turbine parameter information obtained in step 2) and blade element-momentum theory, a rotating actuator disk model for wind turbine wake simulation is established; Step 8) establishing a three-dimensional numerical calculation model of the target terrain wind farm based on the data of steps 5), 6) and 7); Step 9) discretizing the wind speed and direction data of the target terrain wind farm; Step 10) obtaining wind speed and direction information at each wind tower location in the wind farm under each operating condition considering the wake effect through CFD simulation; Step 11) obtaining wind speed and direction information at the hub height of each wind turbine in the wind farm taking into account the wake effect through CFD simulation; Step 12) Divide the target wind farm into several sub-areas; Step 13) obtaining the power of each wind turbine and the total power of each sub-area in the wind farm according to the factory power curve of each wind turbine in the wind farm; Step 14) using the wind speed and direction at the wind tower location in the CFD simulation in step 10) as input and the total power of each sub-area of ​​the wind farm as output, to establish a grouped power prediction database; Step 15) Input the wind speed and direction at the wind tower location at the prediction time, and obtain the total power of the wind farm through the back propagation neural network model to achieve power prediction.

2. The complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning according to claim 1 is characterized in that: In step 1), the original elevation data of the target terrain wind farm is obtained, the outliers and invalid values ​​of the original elevation data are filtered, and the missing areas after data filtering are filled by interpolation method. At the same time, the truncation boundary of the target terrain wind farm is smoothed by using square sine function, and a transition section is set around the target terrain wind farm.

3. The complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning according to claim 1 is characterized in that: The parameter information of the fan in step 2) includes: the hub height of the fan, the rotor diameter of the fan, the thrust coefficient curve of the fan, and the factory power curve of the fan.

4. The complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning according to claim 1 is characterized in that: In step 3), the wind farm wind measurement data includes: wind speed, wind direction and power generation data at the height of the wind turbine hub, the longitude and latitude of the wind turbine, and the altitude of the wind turbine; wind speed and wind direction data at different heights of the wind tower, the longitude and latitude of the wind tower, and the altitude of the wind tower.

5. The complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning according to claim 1 is characterized in that: The method for cleaning the wind farm wind measurement data in step 4) is as follows: at a resolution of 1 minute, discontinuous time data is considered abnormal data; at a resolution of 1 minute, wind speed data is set within the range of 3 to 20 m / s, and data outside this range is considered abnormal data; at a resolution of 1 minute, at a set height, data with a wind direction angle difference greater than 22.5° is considered abnormal data; At a resolution of 15 minutes, when the rate of change of adjacent wind speed values ​​is greater than 20%, the latter wind speed data is considered abnormal data; at a resolution of 15 minutes, when the difference between the wind speed at a height of 70 meters and the wind speed of the first wind turbine at the same moment is greater than 10 meters per second, the data is considered abnormal data; at a resolution of 15 minutes, when the difference between the wind speed at a height of 70 meters and the wind speed of the second wind turbine at the same moment is greater than 10 meters per second, the data is considered abnormal data, where the first wind turbine and the second wind turbine are the two wind turbines closest to the two sides of the wind measurement tower respectively.

6. The complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning according to claim 1 is characterized in that: In step 5), based on the target terrain wind farm elevation information obtained in step 1), the terrain surface is fitted, where the span numbers in the U and V directions of the terrain surface are both set to 110, and the hardness is set to 5; and based on the fitted terrain surface, a three-dimensional numerical model of the target terrain is obtained by outward expansion, where the outward expansion distance is set to 5-10 times the height of the target terrain.

7. The complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning according to claim 1 is characterized in that: The method for simulating different wind directions in the wind farm in step 6) is: based on the rotating area modeling method, in the geometric modeling process, a cylindrical rotating area is established in the core area of ​​the target terrain wind farm, and the cylindrical rotating area is meshed. After the meshing is completed, the cylindrical rotating area is rotated to realize simulation under different wind directions.

8. The complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning according to claim 1 is characterized in that: In step 9), the method for discretizing the wind speed and direction data of the target terrain wind farm is as follows: dividing the wind direction of 0 to 360 degrees into multiple fan-shaped wind direction areas; dividing the wind speed into two wind speed areas, and discretizing each wind speed area according to different intervals; then combining each wind speed and wind direction area to correspond to a wind condition, thereby completing the discretization of the wind speed and direction data of the target terrain wind farm.

9. The complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning according to claim 1 is characterized in that: In step 12), based on the results of the clustering algorithm, the target terrain wind farm is divided into several sub-areas. The specific method is as follows: obtain the following four characteristic parameters of each wind turbine: the average annual wind speed U at the height of the wind turbine hub; mean , the average power value of the whole year P mean , the horizontal coordinate X of the fan position, the vertical coordinate Y of the fan position, and the four characteristic parameters of each fan are normalized; the normalized characteristic parameters are used as input, and the K-means++ algorithm is used for clustering grouping, and the silhouette coefficient and the sum of squares of the intra-cluster distance are used as judgment indicators for different grouping schemes.

10. The complex terrain wind farm power prediction method based on CFD flow field pre-calculation and machine learning according to claim 1, characterized in that: Also includes step 16), calculate the power prediction accuracy C according to the following formula R : Where: n is the total number of time periods within the error statistical time interval minus the number of time periods exempted from assessment; P Pi is the predicted average power in period i; P Mi is the actual average power in period i; C i is the total startup capacity in time period i.

Images