Complex terrain wind power output prediction method based on high-precision meteorological and terrain coupling

Abstract

This invention belongs to the field of wind power generation prediction technology and discloses a method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling. The method includes: acquiring digital elevation model (DEM) data and global reanalysis meteorological data; constructing a multi-scale nested meteorological simulation system, embedding the DEM into the reanalysis data, enabling large eddy simulation at the innermost layer, and parameterizing the wind turbine in the form of a momentum sink to generate three-dimensional wind field data; extracting static terrain feature vectors; calculating transient air density based on real-time air pressure, temperature, and humidity to correct the theoretical power of the wind turbine; and inputting the three-dimensional wind field data, static terrain feature vectors, and corrected theoretical power into a hybrid neural network prediction model to output predicted active power. This invention solves the problems of low prediction accuracy and lack of multi-physics coupling in wind power generation in complex terrain, significantly improving prediction accuracy and applicable to wind farm planning and operation and maintenance decisions.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation prediction technology, specifically involving a method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling. Background Technology

[0002] As an important component of clean energy, wind power plays a crucial role in the transformation of the energy structure. With the development and utilization of high-quality wind resources in flat areas such as plains and coastal areas approaching saturation, the focus of wind power development is gradually shifting to complex terrain areas such as mountains, hills, and loess gullies. The development and efficient operation of wind farms in complex terrain has become an important direction for the development of the wind power industry.

[0003] However, the complex terrain and diverse landforms present significant challenges to the prediction of wind power generation in wind farms. In actual operation, it has been found that under the same nominal wind speed, the actual output of wind turbines in complex terrain wind farms is only about 60% of the output or theoretically calculated output of wind farms in adjacent flat terrain. There is a significant "terrain loss" phenomenon, which causes the actual power generation of wind farms to be far lower than expected. At the same time, it greatly increases the error in wind power output prediction, seriously affecting the safe and stable dispatch of the power grid and the economic benefits of wind farms.

[0004] Current wind power output prediction technologies have many shortcomings in application to complex terrain scenarios, and can no longer meet the high-precision prediction requirements of wind fields in complex terrain. The specific problems are as follows: The resolution of meteorological numerical simulation is insufficient. Traditional numerical weather prediction (NWP) uses grids with a precision of 10 kilometers, which is difficult to resolve complex terrain details such as valleys, ridges, and ravines at the hundred-meter level. It cannot accurately reproduce the airflow movement state under complex terrain, resulting in distorted airflow simulation results and bringing basic errors to subsequent power prediction. The assumptions of wake and power calculation models fail in complex terrain. Most existing wind turbine wake models are based on the ideal conditions of flat terrain, ignoring the "high slope effect" caused by complex terrain, namely airflow separation and turbulence. They also do not consider the "two-way wake asymmetry" under complex terrain, as well as the influence of air density changes with altitude on wind energy capture, resulting in a large deviation between the calculated wake loss and theoretical output and the actual situation. Lacking a deep coupling mechanism of multiphysics, traditional wind power output prediction methods fail to organically integrate transient meteorological parameters (such as air density), complex terrain features, and the actual operating characteristics of wind turbines. The parameters are independent of each other and their coupling effect on the comprehensive impact of wind turbine output is not fully considered. This results in persistently high power prediction errors, which not only fail to provide accurate output data for grid dispatch but also affect the formulation of wind farm power generation plans and the optimization of operation and maintenance decisions, thus restricting the development and utilization efficiency of wind farms in complex terrain.

[0005] In summary, in response to the wind power development needs in complex terrain areas, there is an urgent need to develop a wind power output prediction method that can adapt to complex terrain features and improve the accuracy of output prediction, so as to solve the above-mentioned defects of existing technologies and meet the actual operation and grid dispatch needs of wind farms in complex terrain. Summary of the Invention

[0006] The purpose of this invention is to provide a method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling, which solves the technical problems of low prediction accuracy, poor terrain adaptability, and lack of multi-physics coupling mechanism in the existing technology for wind power output in complex terrain.

[0007] This invention is achieved through the following technical solution: This invention discloses a method for predicting wind power output in complex terrain based on high-precision meteorological and topographic coupling, comprising the following processes: Acquire digital elevation model data and global reanalysis meteorological data for the target wind farm; A multi-scale nested meteorological simulation system is constructed, in which the digital elevation model data is embedded into the global reanalysis meteorological data, large eddy simulation is enabled in the innermost nested layer, and wind turbines are parameterized and embedded into the meteorological model in the form of momentum sinks to generate three-dimensional wind field data covering each wind turbine location. The static terrain feature vector of each wind turbine location is extracted based on the three-dimensional wind field data; The transient air density is calculated based on real-time simulated air pressure, temperature and humidity data, and the theoretical power of the fan is corrected using the transient air density to obtain the corrected theoretical power of the fan. The three-dimensional wind field data, static terrain feature vector, and corrected theoretical power of the wind turbines are input into a pre-trained hybrid neural network prediction model, which outputs the predicted active power value for each wind turbine location.

[0008] Furthermore, the multi-scale nested meteorological simulation system adopts at least a three-layer nested grid architecture, with the outer grid having a resolution of kilometers, the inner grid having a resolution of 100 meters or higher, and a large eddy simulation module enabled in the innermost layer.

[0009] Furthermore, when extracting the static terrain feature vector for each wind turbine location, the extracted features include the slope, roughness length, curvature, and azimuth relative to the prevailing wind direction.

[0010] Furthermore, virtual lidar wind field data at the corresponding wind turbine hub height is generated based on the three-dimensional wind field data. The airflow separation zone and strong turbulence zone at the wind turbine hub height are identified based on the virtual lidar wind field data, and the cut-in wind speed threshold and cut-out wind speed threshold of the wind turbine are corrected based on the identification results.

[0011] Furthermore, the formula for calculating the transient air density is as follows:

[0012] The gas constant for dry air; It is the water vapor pressure; P T represents the real-time air pressure, and T represents the real-time temperature. Transient air density; The theoretical power of the wind turbine is corrected by using the transient air density. Specifically, the calculated transient air density is substituted into the wind turbine power curve, and the theoretical output value is recalculated.

[0013] Furthermore, the hybrid neural network prediction model includes gated recurrent units and fully connected layers; The three-dimensional wind field data, static terrain feature vector, and corrected theoretical wind turbine power are input into a pre-trained hybrid neural network prediction model, which outputs the predicted active power value for each wind turbine location, specifically: The three-dimensional wind field data is sent as a dynamic time-series input to the gated loop unit, and the static terrain feature vector is sent as a static input to the fully connected layer. The output of the gated loop unit, the output of the fully connected layer, and the corrected theoretical power of the wind turbine are fused to obtain the predicted active power value.

[0014] Furthermore, the active power prediction values ​​for each wind turbine location include ultra-short-term prediction results, short-term prediction results, and medium-to-long-term prediction results, wherein the time range of ultra-short-term prediction is 0 to 4 hours, the time range of short-term prediction is 24 hours, and the time range of medium-to-long-term prediction is 240 hours.

[0015] Furthermore, after outputting the predicted active power values ​​for each wind turbine location, the terrain loss factor is calculated by the ratio of the predicted active power value to the corrected theoretical wind turbine power. The optimal generator unit layout scheme is generated by removing generator sites whose predicted active power is lower than the first preset threshold or whose terrain loss factor exceeds the second preset threshold.

[0016] Furthermore, after generating the optimal unit layout scheme, terrain damage warning information for each reserved unit location is output. The terrain damage warning information includes the power loss ratio caused by airflow separation or wake effect and the blade fatigue load risk level.

[0017] Furthermore, the formula for calculating the terrain loss factor is as follows: ; Among them, P ideal This is the corrected theoretical power of the wind turbine; P pred This is the predicted value of active power.

[0018] Compared with the prior art, the present invention has the following beneficial technical effects: This invention discloses a method for predicting wind power output in complex terrain based on high-precision meteorological and topographic coupling. Through multi-scale nested simulation, it achieves refined wind field analysis from the global scale to the wind turbine scale, solving the problem of airflow simulation distortion caused by insufficient resolution (10 km level) in traditional numerical weather prediction. By embedding the wind turbine into the meteorological model in the form of a momentum sink, it achieves bidirectional coupling between the wind turbine wake and the atmospheric flow field, breaking through the limitations of wake models in flat terrain. Transient air density correction is introduced to eliminate the systematic bias of power calculation caused by high altitude and seasonal meteorological changes. Through a physics-AI fusion model, physical constraints (correcting theoretical power) are combined with data-driven (dynamic time series + static terrain), making the prediction results both reliable in terms of mechanism and capable of nonlinear fitting, significantly reducing prediction errors.

[0019] Furthermore, by reconstructing the three-dimensional wind field structure at the hub height based on virtual lidar data, it is possible to accurately identify airflow separation zones and strong turbulence zones, thus compensating for the lack of spatial coverage in actual wind measurement data. Based on the identification results, the cut-in / cut-out wind speed thresholds are dynamically corrected to avoid frequent start-ups and shutdowns of the wind turbine or excessive fatigue loads under unsteady airflow conditions, thereby improving the wind turbine's operational safety and lifespan.

[0020] Furthermore, the hybrid neural network model includes a gated recurrent unit (GRU) and a fully connected layer. Three-dimensional wind field data is fed into the GRU as dynamic temporal input, while static terrain feature vectors are fed into the fully connected layer as static input. The GRU output, the fully connected layer output, and the corrected theoretical power are fused to obtain the predicted active power value. The GRU, through update and reset gate mechanisms, effectively captures the long-term and short-term dependencies between meteorological variables and power, solving the gradient vanishing problem of traditional RNNs, and is particularly suitable for processing meteorological data with lags. The fully connected layer extracts a high-dimensional representation of static terrain features, which is then concatenated with the dynamic temporal features to achieve deep fusion of spatiotemporal information. The corrected theoretical power serves as a physical prior constraint, guiding the model to learn in a physically reasonable direction, avoiding overfitting and inconsistency with physics in purely data-driven models.

[0021] Furthermore, after outputting the predicted active power value, this invention calculates the terrain loss factor by the ratio of the predicted value to the corrected theoretical power of the wind turbine; based on the terrain loss factor, turbine locations with predicted active power values ​​below a preset threshold are eliminated, generating an optimal turbine layout scheme. The terrain loss factor quantifies the proportion of power loss caused by terrain and wake, providing a unified quantitative indicator for turbine location selection; combining the predicted power and the loss factor thresholds, inefficient and high-risk turbine locations are eliminated, avoiding the placement of turbines in airflow separation zones and strong turbulence zones; the generated optimal layout scheme can significantly reduce the overall loss rate and increase the total power generation over the entire life cycle. Attached Figure Description

[0022] Figure 1 This is a flowchart of a method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling, according to the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention; that is, the described embodiments are only a part of the embodiments of the present invention, and not all of them.

[0024] The detailed description of the embodiments of the present invention provided in the following figures is not intended to limit the scope of the claimed invention, but merely to illustrate one selected embodiment of the invention. All other embodiments obtained by those skilled in the art based on the figures and embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] Explanation of relevant terms: A momentum sink is a simplified model in computational fluid dynamics that simulates the obstruction and deceleration effects of obstacles such as fans and buildings on airflow as a virtual region that "absorbs" or "removes" momentum from the flow field. This allows for efficient simulation of the reaction force and wake effects on airflow without precisely depicting the geometry of the object.

[0026] Global reanalysis meteorological data refers to a globally covered, spatiotemporally continuous, and physically consistent gridded meteorological dataset generated by fusing historical observation data with model simulation results using a fixed numerical weather prediction model and data assimilation system. This includes, but is not limited to, reanalysis products such as ECMWF ERA5 and NCEP CFSR. This type of data provides initial boundary conditions for the multi-scale nested meteorological simulation system of this invention, ensuring the realism and physical consistency of large-scale meteorological fields.

[0027] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0028] Example 1 like Figure 1 As shown, this invention discloses a method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling, comprising the following processes: Acquire digital elevation model data and global reanalysis meteorological data for the target wind farm; A multi-scale nested meteorological simulation system is constructed, in which the digital elevation model data is embedded into the global reanalysis meteorological data, large eddy simulation is enabled in the innermost nested layer, and wind turbines are parameterized and embedded into the meteorological model in the form of momentum sinks to generate three-dimensional wind field data covering each wind turbine location. The static terrain feature vector of each wind turbine location is extracted based on the three-dimensional wind field data; The transient air density is calculated based on real-time simulated air pressure, temperature and humidity data, and the theoretical power of the fan is corrected using the transient air density to obtain the corrected theoretical power of the fan. The three-dimensional wind field data, static terrain feature vector, and corrected theoretical power of the wind turbines are input into a pre-trained hybrid neural network prediction model, which outputs the predicted active power value for each wind turbine location.

[0029] The multi-scale nested meteorological simulation system adopts at least a three-layer nested grid architecture, with the outer grid having a resolution of kilometers, the inner grid having a resolution of 500 meters or higher, and the innermost layer activating a large eddy simulation module.

[0030] Example 2 This embodiment takes a mountain wind farm located in the gully area of ​​the Loess Plateau as an example. The wind farm has an installed capacity of 100MW, with large terrain undulations and a maximum relative height difference of 200 meters, which is a typical complex terrain wind power scenario.

[0031] I. Data Acquisition and Preprocessing First, acquire SRTM (30-meter precision) Digital Elevation Model (DEM) data for the area, covering the wind farm and a 20-kilometer radius around it. Integrate historical operational data from the on-site SCADA system and meteorological tower data, including time-series data for wind speed, wind direction, power, air pressure, temperature, and humidity for each wind turbine location over the past two years. Obtain ECMWF (Early European Wind Power Factor) global reanalysis meteorological data as the initial boundary conditions for the simulation.

[0032] II. Performing multi-scale nested meteorological simulations Launch a scale-nested meteorological simulation system and construct a triple-nested grid architecture: Outer grid: 9 km resolution, covering a 500×500 km area, capturing large-scale weather system evolution.

[0033] Mid-level grid: with a resolution of 3 kilometers, covering a 100×100-kilometer area, it provides a preliminary analysis of the obstruction and acceleration effects of terrain on airflow.

[0034] Inner grid: 100-meter resolution, covering a 10×10 km area, accurately covering all turbine locations of the target wind farm.

[0035] Large eddy simulation (LES) modules are enabled in the innermost grid to analyze the turbulent structure induced by complex terrain. At the same time, each wind turbine in the wind farm is parameterized and embedded into the meteorological model in the form of a "momentum sink". That is, the resistance of the wind turbine to the airflow is calculated based on the turbine thrust coefficient and acts in the opposite direction on the flow field to achieve bidirectional coupling between the turbine wake and the upstream airflow.

[0036] The above simulation generates three-dimensional wind field data covering all wind turbine locations in the wind farm, including parameters such as wind speed, wind direction, turbulence intensity, and wind shear index at hub height (100 meters), with a time resolution of 15 minutes.

[0037] Taking the simulated northwest wind condition as an example, the simulation results show that there is a significant airflow separation zone at positions 3 and 5 located on the leeward side of the ridge, with a wind speed shear index exceeding 0.5, which is much higher than the normal value of 0.2 for flat terrain.

[0038] III. Terrain Feature Extraction and Virtual LiDAR Measurement Based on the high-resolution 3D wind field data generated in step two, the static terrain feature vector of each wind turbine location is extracted: Position 3: Slope 25 degrees, roughness length 0.3 meters, curvature -0.02 (negative curvature indicates it is located on the leeward slope), azimuth angle relative to the prevailing wind direction (northwest wind) is 135 degrees.

[0039] Machine position 5: slope 18 degrees, roughness length 0.2 meters, curvature -0.01, azimuth angle 120 degrees.

[0040] Simultaneously, virtual LiDAR wind field data was generated using simulated data. Specifically, the wind speed distribution at the hub height (100 meters) and within 200 meters upstream and downstream of each wind turbine location was extracted to reconstruct the three-dimensional wind field structure. The virtual LiDAR data identified that: turbine location 3 is located in an airflow separation zone, where the actual effective wind speed at the hub height is only 65% ​​of the simulated ambient wind speed; turbine location 5 is affected by wake superposition, resulting in turbulence intensity as high as 25%, exceeding the normal operating threshold of the wind turbine.

[0041] Based on the above identification results, the system automatically corrects the cut-in wind speed threshold (from 3m / s to 4m / s) and cut-out wind speed threshold (from 25m / s to 22m / s) of Unit 3 to avoid the fan frequently starting and stopping or bearing excessive fatigue load under unsteady airflow conditions.

[0042] IV. Transient air density correction Based on the real-time air pressure, temperature, and humidity data obtained from step two, calculate the transient air density at each fan location. The calculation formula is as follows:

[0043] in, The gas constant for dry air; It is the water vapor pressure; P T represents the real-time air pressure, and T represents the real-time temperature. This refers to the transient air density.

[0044] Actual water vapor pressure e =RH× e s saturated vapor pressure e s Using Magnus's formula:

[0045] Where t is the temperature in Celsius.

[0046] Taking winter operating conditions as an example, on a certain day, the air pressure is 850 hPa, the temperature is -10℃, and the humidity is 22%. The calculated air density is ρ = 1.125 kg / m³. 3 Compared to the standard value (1.225 kg / m³), 3 The density is approximately 8.2% lower. Taking summer conditions as an example, with an air pressure of 840 hPa, a temperature of 25℃, and a humidity of 65%, the calculated air density ρ = 0.972 kg / m³. 3 It is about 20.7% lower than the standard value.

[0047] The calculated transient air density is substituted into the wind turbine power curve to correct the theoretical output. Due to the wind energy power... Proportional to transient air density ( This correction eliminates systematic biases caused by high altitude and seasonal weather changes. After the correction, the theoretical output in winter is reduced by 8.2%, and the theoretical output in summer is reduced by 20.7%, making the power curve more consistent with actual operating conditions.

[0048] V. Physics-AI Fusion Prediction Construct a hybrid neural network prediction model, with the following model structure: Input layer: contains two branches: dynamic timing input and static input.

[0049] Dynamic time series processing branch: A two-layer gated recurrent unit (GRU) network is used, with each layer containing 128 hidden units. The input is the three-dimensional wind field data generated in step two (the time window is the past 72 hours).

[0050] Static feature processing branch: A two-layer fully connected network is used, with 64 neurons in each layer. The input is a static terrain feature vector.

[0051] Fusion layer: The output features of the GRU branch are concatenated with the output features of the fully connected branch, and then weighted and fused with the corrected theoretical power of the wind turbine.

[0052] Output layer: A fully connected layer is used to output the predicted active power values ​​for the ultra-short term (0-4 hours), short term (24 hours), and medium-to-long term (240 hours).

[0053] The model training used historical SCADA data (the past two years) as the supervision signal, the loss function was the root mean square error (RMSE) between the predicted power and the actual power, the optimizer was Adam, the learning rate was set to 0.001, the batch size was 32, and the training epochs were 200.

[0054] The trained model was deployed to a real-world operating environment and validated using data from January to March of a given year. The prediction results of the traditional method (using only NWP wind speed to input the power curve) are compared with those of the present invention, as shown in Table 1. Table 1

[0055] Taking Unit 3 as an example, the traditional method predicts that the unit will output 2.5MW under a northwest wind of 10m / s, but the actual output is only 1.6MW, with an error of 36%. The method of this invention predicts an output of 1.7MW, with an error of only 6.25%. The system also generates a "terrain damage warning", indicating that Unit 3 suffers a 36% power loss due to airflow separation, and suggests that maintenance personnel pay attention to the blade fatigue load in this area.

[0056] Example 3: Application of wind farm micro-situation optimization methods This embodiment takes a wind farm in a loess plateau gully area that is yet to be developed as an example. The area covers about 50 square kilometers and has a large topographic relief. The optimal layout scheme needs to be selected from 100 candidate turbine locations.

[0057] I. Data Acquisition and Multi-Scale Nested Simulation Acquire SRTM 30-meter precision digital elevation model (DEM) data and nearly 10 years of ECMWF global reanalysis meteorological data for the candidate region. Construct a triple-nested meteorological simulation system (9km-3km-100m), and enable large eddy simulation in the innermost layer to generate high-resolution three-dimensional wind field data covering the entire candidate region.

[0058] II. Candidate Machine Site Feature Extraction and Virtual Measurement Static terrain feature vectors were extracted for 100 candidate aircraft sites, including: slope, roughness length, curvature, and azimuth relative to the prevailing wind direction. Simultaneously, virtual lidar wind field data at the hub height (100 meters) of each candidate site was generated using simulated data to identify the airflow separation risk, turbulence intensity, and wake influence at each location.

[0059] III. Calculation and Prediction of Transient Air Density Based on the simulated air pressure, temperature, and humidity data, the transient air density at each candidate point is calculated using the formula described in Example 1 (considering seasonal changes in winter and summer). The calculated transient air density is then substituted into the fan power curve to correct the theoretical output and obtain the corrected theoretical fan power.

[0060] All the above data (high-resolution three-dimensional wind field data, static terrain feature vector, and corrected theoretical power of wind turbines) are input into the hybrid neural network prediction model trained in Example 2, and the predicted active power value of each wind turbine location is output.

[0061] After outputting the predicted active power values ​​for each wind turbine location, the terrain loss factor is calculated by the ratio of the predicted active power value to the corrected theoretical wind turbine power; the formula for calculating the terrain loss factor is as follows:

[0062] Among them, P ideal This is the corrected theoretical power of the wind turbine; P pred This is the predicted value of active power.

[0063] IV. Camera Position Selection and Optimal Layout Set the filter threshold: Candidates whose predicted active power is less than 60% of the rated power are directly eliminated; Candidate points with a terrain loss factor greater than 0.4 (i.e., loss exceeding 40%) are marked as high risk.

[0064] After screening, 35 points were retained out of 100 candidate points, of which: The predicted power was highest at 10 locations in the ridge area (reaching more than 85% of the rated power), and the terrain loss factor was less than 0.15. Fifteen locations in the leeward slope area were excluded due to severe airflow separation; Ten locations in the wake superposition area were excluded due to excessive turbulence.

[0065] Based on the reserved locations and the principle of minimizing wake effects, the optimal unit layout scheme is generated: 8 units are arranged along the main ridge line and 27 units are arranged on the side slopes. The total installed capacity is adjusted from the original plan of 100MW to 87.5MW, but the overall loss rate is reduced from 35% to 12%, and the expected power generation over the whole life cycle is increased by 18%.

[0066] V. Output Site Selection Report and Topographic Loss Early Warning The system automatically outputs a micro-location report, which includes: The geographical coordinates, predicted power curves, topographic loss factors, turbulence risk levels, and wake impact assessments of the retained points are preserved. Terrain-induced power loss warning information: For the reserved points on the leeward slope, the power loss ratio caused by airflow separation is indicated (e.g., 36% loss at point 3) and the blade fatigue load risk level (high risk, medium risk, low risk).

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting wind power output in complex terrain based on high-precision meteorological and topographic coupling, characterized in that, Includes the following processes: Acquire digital elevation model data and global reanalysis meteorological data for the target wind farm; A multi-scale nested meteorological simulation system is constructed, in which the digital elevation model data is embedded into the global reanalysis meteorological data, large eddy simulation is enabled in the innermost nested layer, and wind turbines are parameterized and embedded into the meteorological model in the form of momentum sinks to generate three-dimensional wind field data covering each wind turbine location. The static terrain feature vector of each wind turbine location is extracted based on the three-dimensional wind field data; The transient air density is calculated based on real-time simulated air pressure, temperature and humidity data, and the theoretical power of the fan is corrected using the transient air density to obtain the corrected theoretical power of the fan. The three-dimensional wind field data, static terrain feature vector, and corrected theoretical power of the wind turbines are input into a pre-trained hybrid neural network prediction model, which outputs the predicted active power value for each wind turbine location.

2. The method for predicting wind power output in complex terrain based on high-precision meteorological and topographic coupling as described in claim 1, characterized in that, The multi-scale nested meteorological simulation system adopts a nested grid architecture of at least three layers, with the outer grid having a resolution of kilometers, the inner grid having a resolution of 100 meters or higher, and the large eddy simulation module being enabled in the innermost layer.

3. The method for predicting wind power output in complex terrain based on high-precision meteorological and topographic coupling as described in claim 1, characterized in that, When extracting the static terrain feature vector for each wind turbine location, the extracted features include the slope, roughness length, curvature, and azimuth relative to the prevailing wind direction.

4. The method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling as described in claim 1, characterized in that, Based on the three-dimensional wind field data, virtual lidar wind field data is generated at the corresponding wind turbine hub height. Based on the virtual lidar wind field data, the airflow separation zone and strong turbulence zone at the wind turbine hub height are identified, and the cut-in wind speed threshold and cut-out wind speed threshold of the wind turbine are corrected based on the identification results.

5. The method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling as described in claim 1, characterized in that, The formula for calculating the transient air density is as follows: The gas constant for dry air; It is the water vapor pressure; P T represents the real-time air pressure, and T represents the real-time temperature. Transient air density; The theoretical power of the wind turbine is corrected by using the transient air density. Specifically, the calculated transient air density is substituted into the wind turbine power curve, and the theoretical output value is recalculated.

6. The method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling as described in claim 1, characterized in that, The hybrid neural network prediction model includes gated recurrent units and fully connected layers; The three-dimensional wind field data, static terrain feature vector, and corrected theoretical wind turbine power are input into a pre-trained hybrid neural network prediction model, which outputs the predicted active power value for each wind turbine location, specifically: The three-dimensional wind field data is sent as a dynamic time-series input to the gated loop unit, and the static terrain feature vector is sent as a static input to the fully connected layer. The output of the gated loop unit, the output of the fully connected layer, and the corrected theoretical power of the wind turbine are fused to obtain the predicted active power value.

7. The method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling as described in claim 1, characterized in that, The predicted active power values ​​for each wind turbine location include ultra-short-term prediction results, short-term prediction results, and medium-to-long-term prediction results. The time range for ultra-short-term prediction is 0 to 4 hours, the time range for short-term prediction is 24 hours, and the time range for medium-to-long-term prediction is 240 hours.

8. The method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling as described in claim 1, characterized in that, After outputting the predicted active power values ​​for each wind turbine location, the terrain loss factor is calculated by the ratio of the predicted active power value to the corrected theoretical wind turbine power. The optimal generator unit layout scheme is generated by removing generator sites whose predicted active power is lower than the first preset threshold or whose terrain loss factor exceeds the second preset threshold.

9. The method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling as described in claim 8, characterized in that, After generating the optimal unit layout scheme, the terrain damage warning information for each reserved unit location is further output. The terrain damage warning information includes the power loss ratio caused by airflow separation or wake effect and the blade fatigue load risk level.

10. The method for predicting wind power output in complex terrain based on high-precision meteorological and terrain coupling as described in claim 8, characterized in that, The formula for calculating the terrain loss factor is as follows: ； Among them, P ideal This is the corrected theoretical power of the wind turbine; P pred This is the predicted value of active power.

Images