Wind power plant short-term power load prediction method and system based on digital twinning

Abstract

The invention relates to the technical field of wind power plant short-term power load prediction, and discloses a wind power plant short-term power load prediction method and system based on digital twinning, and the method comprises the steps: firstly obtaining a weather forecast value, carrying out the correction, dividing the weather forecast value into layered data according to the atmospheric stability, mapping the layered data to a three-dimensional grid, and carrying out the iterative evolution to generate an updated three-dimensional wind field; and combining the updated three-dimensional wind field, the wind power plant data and the fan position to construct a spatial discrete wind field, obtaining a corrected wind field after correction, determining the wake flow influence range of the corrected wind field, and performing adjustment through the adjustment point to obtain a final wind field. Screening out a normal fan set of which the wind speed exceeds a preset wind speed threshold value, fitting a wind speed and power change curve by using a polynomial, generating a power set by wind power plant fans according to the change curve, inputting the power set into an LSTM model to obtain a predicted power set, and correcting and smoothly filtering the predicted power set to form a final power set. The method can accurately describe the wind field and accurately reflect the fan power change of the wind power plant.

Description

Technical Field

[0001] This invention relates to the field of short-term power load forecasting technology for wind farms, and particularly to a method and system for short-term power load forecasting of wind farms based on digital twins. Background Technology

[0002] Currently, wind power, as an important component of clean energy, occupies an increasingly crucial position in the global energy transition. The power generation capacity of wind farms is greatly affected by natural wind resources, exhibiting strong randomness and volatility. This necessitates reliance on industrial data analysis techniques for short-term power load forecasting to achieve higher prediction accuracy, thereby supporting grid dispatch and optimal energy allocation.

[0003] Under current technology, most forecasting methods rely on historical power generation data and basic meteorological elements for statistical modeling. However, in actual complex terrain and wind farms with densely packed multiple turbines, these methods struggle to accurately depict the true spatial distribution of wind speed within the farm. Furthermore, the operation of wind turbines generates a significant wake effect, leading to a substantial reduction in wind speed at downstream turbine locations. Existing forecasting methods often ignore or simplify the interactions between turbines, making it difficult to capture the true trends in turbine power output at different times within a wind farm.

[0004] Therefore, existing technologies have the drawback of being unable to accurately characterize wind fields and accurately reflect changes in wind turbine power. Summary of the Invention

[0005] This invention provides a method and system for predicting short-term power load in wind farms based on digital twins, which can solve the problem that existing technologies cannot accurately characterize wind farms and accurately reflect changes in wind turbine power.

[0006] In a first aspect, to address the aforementioned technical problems, this invention provides a method for short-term power load forecasting of wind farms based on digital twins, comprising:

[0007] Meteorological data is acquired, and the meteorological data is processed into layers to obtain layered data. The layered data is then mapped onto a three-dimensional grid and iteratively evolved to generate an updated three-dimensional wind field.

[0008] Based on the updated three-dimensional wind field and the acquired wind farm data and wind turbine location information, a spatial discrete wind field is constructed. The wind turbine action area in the spatial discrete wind field is located, the momentum source term of the wind turbine action area is calculated, and the spatial discrete wind field is corrected based on the momentum source term to obtain the corrected wind field.

[0009] Determine the wake influence range of the corrected wind field, calculate the velocity loss rate of each grid point in the wake influence range, mark the grid points whose velocity loss rate is greater than a preset loss rate threshold as adjustment points, and adjust the corrected wind field according to the adjustment points to obtain the final wind field;

[0010] In the final wind field, identify all wind turbines whose wind speed is greater than a preset wind speed threshold and integrate them into a normal wind turbine set. Use a polynomial to fit the historical wind speed and power variation patterns of the normal wind turbine set to obtain the variation curves.

[0011] The wind speed of the wind turbines in the wind farm at various times is obtained and the change curve is input to obtain the power set. The power set is then input into a preset long short-term memory neural network model for prediction to obtain the predicted power set.

[0012] The predicted power set is obtained by identifying the wind turbines affected by the wake, and all wake source wind turbines corresponding to the affected wind turbines are also identified and their combined loss is calculated. The predicted power set is then corrected and smoothed based on the combined loss to obtain the final power set.

[0013] In one optional implementation, the steps of acquiring meteorological data, performing layered processing on the meteorological data to obtain layered data, mapping the layered data to a three-dimensional grid and iteratively evolving it to generate an updated three-dimensional wind field include:

[0014] Meteorological forecast values ​​are obtained, and at the same time, measured values ​​are obtained from ground-based meteorological data. The meteorological forecast values ​​are then corrected based on the measured values ​​to obtain corrected meteorological values.

[0015] The Monin-Obukhov length is calculated for the corrected meteorological value. Based on the Monin-Obukhov length, the atmospheric stability of the corrected meteorological value is assessed and stratified to obtain stratified data.

[0016] The vertical wind profiles at each location are calculated based on the layered data and mapped onto a three-dimensional grid to form an initial three-dimensional wind field.

[0017] The initial three-dimensional wind field is iteratively evolved using a preset numerical solver to obtain an updated three-dimensional wind field.

[0018] In one optional implementation, the step of constructing a spatially discrete wind field based on the updated three-dimensional wind field and the acquired wind farm data and wind turbine location information, locating the wind turbine operating area in the spatially discrete wind field, calculating the momentum source term of the wind turbine operating area, and correcting the spatially discrete wind field based on the momentum source term to obtain a corrected wind field includes:

[0019] Acquire wind farm data, wind turbine location information, and wind turbine blade geometry data;

[0020] Based on the updated three-dimensional wind field, the wind farm data, and the wind turbine location information, a spatially discrete wind field is constructed.

[0021] The blade sweep area is calculated based on the wind turbine location information and the wind turbine blade geometry data. The blade sweep area is then mapped to the spatial discrete wind field to obtain the wind turbine action grid.

[0022] The momentum source term of the wind turbine action grid is calculated using a preset actuation disk model. The spatial discrete wind field and the momentum source term are then input into a preset numerical solver to obtain the corrected wind field.

[0023] In one optional implementation, determining the wake influence range of the modified wind field, calculating the velocity deficit rate of each grid point within the wake influence range, marking the grid points with velocity deficit rates greater than a preset deficit rate threshold as adjustment points, and adjusting the modified wind field according to the adjustment points to obtain the final wind field includes:

[0024] The dominant incoming wind direction is determined based on the modified wind field, and the wake influence range is determined based on the dominant incoming wind direction.

[0025] Calculate the velocity loss rate of each grid point in the wake influence range. If the velocity loss rate is greater than a preset loss rate threshold, mark the grid point as an adjustment point to obtain an adjustment point list.

[0026] In the corrected wind field, find the adjustment point in the list of adjustment points, adjust the wind speed of the adjustment point and update the wind speed value to obtain the final wind field.

[0027] In one optional implementation, the step of identifying all wind turbines in the final wind field whose wind speeds exceed a preset wind speed threshold and integrating them into a normal wind turbine set, and using a polynomial to fit the historical wind speed and power variation patterns of the normal wind turbine set to obtain a variation curve, includes:

[0028] In the final wind field, for each wind turbine, the wind speed corresponding to the grid point closest to the wind turbine's location information is searched as the wind turbine's wind speed, thus obtaining a wind speed sequence;

[0029] If the wind speed of a fan in the wind speed sequence is greater than a preset wind speed threshold, then the fan is marked as a normal fan, and a set of normal fans is obtained.

[0030] The historical wind speed and power of each wind turbine in the set of normal wind turbines are obtained, and the variation patterns of the wind speed and power are fitted using a polynomial to obtain the variation curve.

[0031] In one optional implementation, the step of obtaining the wind speed of the wind turbines at various times and inputting the change curve to obtain a power set, and then inputting the power set into a preset long short-term memory neural network model for prediction to obtain a predicted power set, includes:

[0032] Obtain wind farm data for wind turbines within a preset time period, and extract wind speed at each moment within the preset time period;

[0033] Input the wind speed at each time point into the change curve to obtain the power at each time point, and integrate the time points and the power at each time point into a power set;

[0034] The power set is input into a preset long short-term memory neural network model for prediction to obtain the predicted power set.

[0035] In one optional implementation, the step of acquiring the affected wind turbines in the predicted power set, simultaneously acquiring all wake source wind turbines corresponding to the affected wind turbines and calculating the comprehensive loss, and correcting and smoothing the predicted power set based on the comprehensive loss to obtain the final power set includes:

[0036] Obtain the wind direction angle at each moment in the predicted power set, calculate the wake influence range of the wind turbines in the wind farm based on the wind direction angle, and obtain the affected wind turbines;

[0037] Identify the wake source fans that cause wake effects on the affected fans, calculate the overall loss caused by the wake source fans to the affected fans, and correct the predicted power set based on the overall loss to obtain the corrected power set.

[0038] The corrected power set is then smoothed to obtain the final power set.

[0039] Secondly, the present invention provides a short-term power load forecasting system for wind farms based on digital twins, comprising:

[0040] The initial wind field generation module is used to acquire meteorological data, perform layered processing on the meteorological data to obtain layered data, map the layered data to a three-dimensional grid and perform iterative evolution to generate an updated three-dimensional wind field.

[0041] The wind field correction module is used to construct a spatial discrete wind field based on the updated three-dimensional wind field and the acquired wind farm data and wind turbine location information, locate the wind turbine action area in the spatial discrete wind field, calculate the momentum source term of the wind turbine action area, and correct the spatial discrete wind field based on the momentum source term to obtain the corrected wind field.

[0042] The final wind field generation module is used to determine the wake influence range of the corrected wind field, calculate the velocity loss rate of each grid point in the wake influence range, mark the grid points whose velocity loss rate is greater than a preset loss rate threshold as adjustment points, and adjust the corrected wind field according to the adjustment points to obtain the final wind field.

[0043] The regularity fitting module is used to identify all wind turbines in the final wind field whose wind speed is greater than a preset wind speed threshold and integrate them into a normal wind turbine set. The module uses a polynomial to fit the historical wind speed and power variation patterns of the normal wind turbine set to obtain the variation curve.

[0044] The power prediction module is used to obtain the wind speed of the wind turbines in the wind farm at various times and input the change curve to obtain the power set. The power set is then input into a preset long short-term memory neural network model for prediction to obtain the predicted power set.

[0045] The final power generation module is used to acquire the affected wind turbines in the predicted power set, acquire all the wake source wind turbines corresponding to the affected wind turbines and calculate the comprehensive loss, and correct and smooth the predicted power set according to the comprehensive loss to obtain the final power set.

[0046] Thirdly, the present invention also provides an electronic device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the short-term power load forecasting method for wind farms based on digital twins as described above.

[0047] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the short-term power load forecasting method for wind farms based on digital twins as described above.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) This invention generates an initial wind field by correcting meteorological forecast values, layering atmospheric stability and iterative evolution of three-dimensional grids, and then corrects it by combining wind farm data. At the same time, it adjusts the wake effect range to obtain the final wind field. It takes into account meteorological conditions, terrain characteristics and wake effects to accurately depict the real distribution law of wind speed in the field space, and greatly improves the realism and accuracy of wind field simulation.

[0050] (2) In this invention, a set of normal wind turbines is selected based on the final wind field, and wind turbine data with abnormal wind speeds are excluded. The historical wind speed and power data of normal wind turbines are deeply mined by the polynomial fitting method, which can fully characterize the nonlinear output characteristics of wind turbines in different wind speed ranges, so that the fitting curves are close to the actual operating rules, and provide a reliable basis for subsequent power prediction.

[0051] (3) This invention obtains the power set of wind turbines in the wind farm through the change curve, uses the LSTM model to make time-series predictions on the power set, and combines the Jensen wake model to correct the prediction results and smooth the filtering. Finally, the final power set not only conforms to the actual power time-series change law, but also reduces the output uncertainty caused by wake interference, improves the accuracy of wind farm power prediction, and provides reliable technical support for the efficient operation of wind farms. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the short-term power load forecasting method for wind farms based on digital twins provided in the first embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram of the structure of a wind farm short-term power load forecasting system based on digital twins provided in the second embodiment of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Reference Figure 1 The first embodiment of the present invention provides a schematic diagram of a short-term power load forecasting method for wind farms based on digital twins, including steps S11 to S16, as follows:

[0056] S11, acquire meteorological data, perform layered processing on the meteorological data to obtain layered data, map the layered data onto a three-dimensional grid and perform iterative evolution to generate an updated three-dimensional wind field;

[0057] S12, construct a spatial discrete wind field based on the updated three-dimensional wind field and the acquired wind farm data and wind turbine location information, locate the wind turbine action area in the spatial discrete wind field, calculate the momentum source term of the wind turbine action area, and correct the spatial discrete wind field based on the momentum source term to obtain the corrected wind field;

[0058] S13, determine the wake influence range of the corrected wind field, calculate the velocity loss rate of each grid point in the wake influence range, mark the grid points whose velocity loss rate is greater than a preset loss rate threshold as adjustment points, adjust the corrected wind field according to the adjustment points, and obtain the final wind field;

[0059] S14. In the final wind field, find all wind turbines whose wind speed is greater than the preset wind speed threshold and integrate them into a normal wind turbine set. Use a polynomial to fit the historical wind speed and power variation of the normal wind turbine set to obtain the variation curve.

[0060] S15, obtain the wind speed of the wind turbines in the wind farm at each time and input the change curve to obtain the power set, input the power set into the preset long short-term memory neural network model for prediction, and obtain the predicted power set;

[0061] S16, obtain the affected wind turbines in the predicted power set that are affected by the wake, and at the same time obtain all wake source wind turbines corresponding to the affected wind turbines and calculate the comprehensive loss. Based on the comprehensive loss, correct the predicted power set and smooth the filter to obtain the final power set.

[0062] In step S11, meteorological data is acquired, the meteorological data is processed into layers to obtain layered data, the layered data is mapped onto a three-dimensional mesh and iteratively evolved to generate an updated three-dimensional wind field, including:

[0063] Meteorological forecast values ​​are obtained, and at the same time, measured values ​​are obtained from ground-based meteorological data. The meteorological forecast values ​​are then corrected based on the measured values ​​to obtain corrected meteorological values.

[0064] The Monin-Obukhov length is calculated for the corrected meteorological value. Based on the Monin-Obukhov length, the atmospheric stability of the corrected meteorological value is assessed and stratified to obtain stratified data.

[0065] The vertical wind profiles at each location are calculated based on the layered data and mapped onto a three-dimensional grid to form an initial three-dimensional wind field.

[0066] The initial three-dimensional wind field is iteratively evolved using a preset numerical solver to obtain an updated three-dimensional wind field.

[0067] It is worth noting that when obtaining weather forecast values, large-scale meteorological data, including basic elements such as temperature, humidity, and wind speed, are acquired through pre-set global numerical weather prediction models, such as GFS or ECMWF. This data is stored in a grid format, covering a large area, but due to limited resolution (e.g., GFS resolution is 0.25°×0.25°, and ECMWF resolution is 0.1°×0.1°), it cannot accurately reflect deviations caused by local topography or the urban heat island effect. Therefore, it is necessary to correct the weather forecast values ​​by combining them with measured meteorological values ​​collected by pre-set ground monitoring devices. These pre-set ground monitoring devices include automatic weather stations, temperature sensors, and humidity sensors, which are generally evenly deployed in the area to be monitored, and can monitor temperature, humidity, etc., at their deployment locations. The specific correction method involves collecting meteorological values ​​from one grid point of the weather forecast, simultaneously collecting meteorological values ​​from all monitoring devices within that grid, calculating the deviation between the measured values ​​of each device and the grid point's forecast value, and finally averaging all deviations. The average deviation is then summed with the forecast value to obtain the corrected meteorological value. For example, grid point A in the GFS model is selected, and the 2-meter temperature forecast value for this grid point is 18.5°C. Then, measured data from three automatic weather stations within this grid area are collected: station 1 is 20.2°C, station 2 is 19.8°C, and station 3 is 20.5°C. The deviations for each station are calculated to be 1.7°C, 1.3°C, and 2.0°C, respectively. The average deviation is then taken as 1.67°C. Finally, this average deviation is added to the original forecast value to obtain the corrected temperature of 20.17°C.

[0068] It is worth noting that the virtual temperature is calculated based on the corrected meteorological data for temperature, humidity, vapor pressure, and wind speed at different altitudes. ,in The absolute temperature is obtained by converting the corrected air temperature from Celsius to Kelvin, where e is the vapor pressure value, derived from the corrected air temperature. (°C) and relative humidity The saturated vapor pressure is obtained by calculation using the formula. Then by formula To obtain the actual water vapor pressure value, Standard atmospheric pressure.

[0069] According to the formula ,in For friction speed, For the average virtual temperature, Let K is the Karman constant, set to 0.4, and g be the acceleration due to gravity. The Monin-Obukhov length L is calculated to represent the virtual temperature flux, which is the covariance between the vertical velocity fluctuations and the virtual temperature fluctuations. Friction velocity and virtual temperature flux are measured using the eddy covariance method, and the average virtual temperature is obtained by averaging the virtual temperature observations at each altitude. Atmospheric stability is assessed based on the sign of the Monin-Obukhov length L, dividing the atmospheric boundary layer into stable, neutral, and unstable layers. A layer is considered stable when L > 0, unstable when L < 0, and neutral when |L| approaches infinity. The calculated values ​​are used to assess atmospheric stability and are then categorized to form a layered data structure. For example, the calculated Monin-Obukhov length L = 850 m at observation point A is positive, indicating a stable atmosphere at that point. Simultaneously, L = −420 m is calculated at observation point B, indicating an unstable layer. L at observation point C approaches infinity, indicating a neutral layer. By integrating these atmospheric stability assessment results according to spatial location, a layered data consisting of stable, neutral, and unstable layers can be formed.

[0070] It is worth noting that the constructed atmospheric stability stratification data requires the selection of corresponding wind speed profile models based on different stability layers. The stability layer is determined using the formula... The correction function This is a correction function based on stability-based partitioning changes, in the stable layer. z is the height. The surface roughness length is given. The unstable layer uses the same formula as the stable layer, and the correction function is... ,in The formula for the neutral layer is: By substituting data from different locations in different stability layers into the corresponding profile formulas, the vertical wind profile from near the ground to the top of the boundary layer is calculated. Finally, the profile is mapped onto three-dimensional grid nodes according to height and horizontal position. The horizontal grid size is 100m and the vertical grid size is 50m, so that each grid point obtains an initial wind speed value that matches its stability category and observation height, forming an initial three-dimensional wind field consistent with the actual atmospheric boundary layer structure.

[0071] For example, for the constructed atmospheric stability layered data, when observation point A is in the stable layer, the stable layer wind speed profile model is selected, and the wind speed is calculated to be 8.27 m / s; when observation point B is in the unstable layer, the unstable layer wind speed profile model is selected, and the wind speed is calculated to be 6.85 m / s. Then, the wind speed values ​​calculated from the vertical wind profiles at different locations and heights, such as 50m, 100m, and 150m, are mapped to the corresponding three-dimensional grid nodes. For example, the node wind speed of the horizontal grid (500, 800, 100) corresponding to observation point A is assigned the value of 8.27 m / s, thus forming the initial three-dimensional wind field.

[0072] It is worth noting that the preset numerical solvers include OpenFOAM and ANSYS Fluent. The three-dimensional wind field is used as the initial wind field input to the numerical solver, and topographic data, surface roughness, temperature, humidity, and air pressure are also input as conditions. Among them, the topographic data, surface roughness, temperature, humidity, and air pressure are obtained from the preset meteorological database. The preset meteorological database consists of multi-source meteorological observation data and topographic survey data, including digital elevation model data of the wind field area, covering information such as altitude, slope, aspect, and topographic relief; roughness lengths corresponding to different surface types, such as grassland, farmland, forest, water area, and built-up area; and temperature, humidity, air pressure, and wind speed data collected continuously over a long period of time by equipment such as wind field anemometer towers, meteorological stations, and automatic weather stations. The solver loads a set of preset fluid dynamics equations, including the continuity equation, momentum equation, and energy equation, and solves them. The solver iterates on the 3D mesh, with a maximum number of iterations of 1000. The root mean square of the residuals of each equation is used as the iteration error. When the iteration error is less than 0.001, the wind field is in a stable state, and the iteration stops. The solvedr then outputs the updated 3D wind field distribution matrix, which updates the 3D wind field.

[0073] For example, OpenFOAM is selected as the numerical solver. The initial 3D wind field generated above is used as input, and the terrain data, surface roughness data, temperature, humidity, and air pressure of each grid are imported as solution conditions. The solver loads the preset fluid dynamics equations and starts iterative calculation. When iterates to 120 steps, the error drops to 0.0008 (less than the preset threshold of 0.001), the wind field distribution tends to stabilize, the iteration stops, and the updated 3D wind field distribution matrix is ​​output. This matrix contains the wind speed values ​​of all grid nodes in a horizontal 100m×100m and a vertical 50m area. For example, the wind speed of grid node (500,800,100) is updated from the initial 8.27m / s to 8.19m / s, and finally the updated 3D wind field is formed.

[0074] In step S12, a spatial discrete wind field is constructed based on the updated three-dimensional wind field and the acquired wind farm data and wind turbine location information. The wind turbine operating area in the spatial discrete wind field is located, the momentum source term of the wind turbine operating area is calculated, and the spatial discrete wind field is corrected based on the momentum source term to obtain the corrected wind field, including:

[0075] Acquire wind farm data, wind turbine location information, and wind turbine blade geometry data;

[0076] Based on the updated three-dimensional wind field, the wind farm data, and the wind turbine location information, a spatially discrete wind field is constructed.

[0077] The blade sweep area is calculated based on the wind turbine location information and the wind turbine blade geometry data. The blade sweep area is then mapped to the spatial discrete wind field to obtain the wind turbine action grid.

[0078] The momentum source term of the wind turbine action grid is calculated using a preset actuation disk model. The spatial discrete wind field and the momentum source term are then input into a preset numerical solver to obtain the corrected wind field.

[0079] It is worth noting that the wind farm data is determined by ground-based measurements, recording the wind farm's area, topographic elevation (DEM) data, meteorological background data, etc.; the wind turbine location information is retrieved from the GIS geographic information platform, including the three-dimensional coordinates of the hub center of all wind turbines; the wind turbine blade geometry data is obtained from the wind turbine manufacturer's design manual, including data such as the wind turbine impeller diameter, blade chord length, and hub height.

[0080] It is worth noting that, based on the updated 3D wind field, wind farm data, and wind turbine location information, the wind farm area is used as the computational domain. Combined with the wind turbine location information, a dense and sparse grid area is defined. The dense area is a circular region with a radius of 500 m centered on each wind turbine. The dense grid resolution is set to 10 m × 10 m horizontally, and 2 m vertically within ±50 m of the wind turbine hub height, with a spacing of 5 m for other vertical layers. The sparse area is the region outside the dense area, with a sparse grid resolution of 50 m × 50 m horizontally and a vertical spacing of 10 m. Based on the updated 3D wind speed data and its corresponding coordinate grid, for grid nodes with higher resolution in the dense and sparse areas, if their coordinates are between two adjacent grid nodes in the original updated 3D wind field data, linear interpolation is used to calculate the wind speed value of the target node. Non-interpolated areas directly use the updated 3D wind speed data, forming a spatially discrete wind field where each grid node has a wind speed assigned. For example, a wind farm contains 10 wind turbines. Based on the updated 3D wind field and turbine locations, the wind farm boundary is used as the computational domain to delineate a dense zone and a sparse zone. A circular dense zone with a radius of 500m is delineated for turbine No. 1, with a horizontal resolution of 10m×10m and a vertical resolution of 2m (50m~150m). The vertical spacing between other layers is 5m. The sparse zone is located outside the dense zone, with a horizontal resolution of 50m×50m and a vertical spacing of 10m. If, within the dense zone, the target grid node (515,790,102) is located between the original grid nodes (500,750,100) and (600,850,150), the wind speed at this node is calculated to be 7.68m / s using linear interpolation. This results in a spatially discrete wind field where each grid has a wind speed.

[0081] It is worth noting that, based on the wind turbine location information and the wind turbine blade geometry data, the blade sweep area is calculated. Specifically, the three-dimensional coordinates of the hub center are obtained from the wind turbine location information. The impeller diameter D and hub height are obtained from the geometric data of the wind turbine blades. In the 3D mesh, the swept region of the blade satisfies , ,in Let (x, y, z) be the vertical spacing of the grid, and (x, y, z) be the center coordinates of the grid. Traverse all grids within the discrete wind field and select grid cells that satisfy the above conditions for the grid center coordinates, marking them as the wind turbine's impact grid. For example, if the hub center coordinates of wind turbine No. 1 are (512, 786, 100), the impeller diameter is 120m, and the vertical grid spacing is 2m, then the blade sweep area must satisfy (x−512)²+(y−786)²≤60² and |z−100|≤1. Traverse the discrete wind field grids; grid A, with center coordinates (510, 788, 100), satisfies the blade sweep area condition and is marked as the wind turbine's impact grid.

[0082] It is worth noting that the momentum source term is calculated using the actuator disk model for the wind turbine action grid. Specifically, based on the impeller diameter in the wind turbine blade geometry data, the swept area of ​​all impellers is calculated using the circular area formula. If the swept areas of multiple wind turbines overlap in the spatially discrete wind field, overlapping grid cells must first be identified. All wind turbine action grids are traversed, and grids covered by the swept areas of two or more wind turbines are marked as overlapping grids. The swept area of ​​the overlapping area is then split according to the number of wind turbines. For example, in a grid where the swept areas of two wind turbines overlap, the effective swept area of ​​each wind turbine within that grid is 50% of the original single-turbine swept area. For multiple overlapping areas, the swept area is split equally according to the number of wind turbines. This is combined with the wind speed u and air density of the grid. The thrust coefficient C (typically 0.8 by default, but the specific value needs to be adjusted between 0.7 and 0.9 depending on the wind turbine type and operating conditions) is used to calculate the total thrust. Then, the total thrust is evenly distributed according to the number and volume of the wind turbine's action grids to obtain the momentum source term for each action grid. The calculation method is as follows Where N is the wind turbine action grid and V is the grid volume, following the method of numerical solver calculation in step S11, the spatial discrete wind field and each momentum source term are input into the numerical solver, and finally the numerical solver outputs the corrected wind field.

[0083] For example, the swept area of ​​the impeller of wind turbine No. 1 is 11310 m², the air density is 1.225 kg / m³, the thrust coefficient C = 0.8, the wind speed of the affected grid is 7.68 m / s, and the total thrust is calculated to be 328500 N. If the wind turbine has 150 affected grids, the momentum source term of each affected grid is −10.95 N / m³. If the swept areas of wind turbine No. 1 and wind turbine No. 2 overlap at grid C, the effective swept area of ​​grid C is divided into 50% each, and the thrust and momentum source terms of the two wind turbines at that grid are calculated separately. The momentum source terms of all affected grids and the spatial discrete wind field are input into the numerical solver, and the wind speed at 100 m of wind turbine No. 1 is corrected to 6.95 m / s.

[0084] In step S13, the wake influence range of the corrected wind field is determined, the velocity deficit rate of each grid point within the wake influence range is calculated, and the grid points with velocity deficit rates greater than a preset deficit rate threshold are marked as adjustment points. The corrected wind field is then adjusted based on these adjustment points to obtain the final wind field, including:

[0085] The dominant incoming wind direction is determined based on the modified wind field, and the wake influence range is determined based on the dominant incoming wind direction.

[0086] Calculate the velocity loss rate of each grid point in the wake influence range. If the velocity loss rate is greater than a preset loss rate threshold, mark the grid point as an adjustment point to obtain an adjustment point list.

[0087] In the corrected wind field, find the adjustment point in the list of adjustment points, adjust the wind speed of the adjustment point and update the wind speed value to obtain the final wind field.

[0088] It is worth noting that the horizontal wind speed vector components of all grid nodes in the corrected wind field are read, namely the vertical u component and the horizontal v component, and the arithmetic mean of all u and v components is calculated. Through vector synthesis method The dominant incoming wind direction angle is obtained, i.e., the dominant incoming wind direction, where (.) represents the arctangent function. The wake influence range is defined along the positive direction of the prevailing wind direction, with the hub center of each wind turbine as the origin. Specifically, it is the area from 0 to 20 times the turbine impeller diameter in the horizontal direction and the area from the hub height to 1 times the impeller diameter in the vertical direction for each wind turbine. For example, if the wind farm's corrected wind field has 10,000 grid nodes, the arithmetic mean of the horizontal wind speed vector components u and v of all nodes is calculated. 6.8 m / s The velocity is -2.1 m / s; the prevailing wind direction angle is calculated using the vector synthesis method. The angle is 162°, meaning the prevailing wind direction is southwest-southwest. Using the hub center (500, 800, 100) of wind turbine No. 3 as the origin, with an impeller diameter of 120m and a hub height of 100m, the wake influence area is defined along the 162° direction. The horizontal range is 0 to 20 times the impeller diameter, i.e., 0 to 2400m, and the vertical range is 100m ± 120m, ranging from -20m to 220m. Since the area below -20m to 0m is underground and has no practical wind field significance, only the effective height range above the ground surface is used in the calculation. The underground area is directly truncated and not included in the subsequent analysis process. Therefore, the actual vertical value is 0 to 220m, thus determining the wake influence area of ​​the wind turbine.

[0089] It is worth noting that an area 2 to 5 times the diameter of the wind turbine impeller is selected as the region unaffected by the wake, and the average wind speed in this region is calculated as the incoming wind speed. Iterate through all grid points within the wake's influence range and substitute each point into the formula. Calculation speed loss rate ,in The actual wind speed at a grid point with coordinates (x, y, z) in the corrected wind field. A preset deficit rate threshold of 10% is set. The deficit rate of each grid point is compared with the deficit rate threshold. If the deficit rate is greater than the threshold, the grid point is marked as a wind speed adjustment point. The three-dimensional coordinates and corresponding deficit rate values ​​of the adjustment points are recorded to form a list of adjustment points. The deficit rate threshold of 10% is determined based on historical wake observation data of the wind farm and the calibration of wind turbine power characteristic curves. In historical operating data, in 95% of the samples, when the speed deficit rate is less than 10%, the corresponding wind speed attenuation is a reasonable effect caused by wind turbine wake obstruction, which is consistent with the aerodynamic characteristics of wind turbines and wake propagation laws. However, when the speed deficit rate exceeds 10%, the excess is not caused by the actual wake effect. It is due to the superposition of interference factors such as terrain disturbance, measurement error, and background turbulence, which will cause an abnormal decrease in the output power of wind turbines in the wake influence range, deviating from the change law under the actual wake effect.

[0090] For example, the area from 240m to 600m of wind turbine No. 4 is selected as the region unaffected by the wake, and the average wind speed in this region is calculated to be 7.2m / s. Traversing all grid points within the wake influence range of this wind turbine, for the grid point with coordinates (650, 750, 100), the actual wind speed in the corrected wind field is 6.3m / s. Substituting this into the formula, the velocity deficit rate is calculated to be 12.5%. Since 12.5% ​​is greater than the preset deficit rate threshold of 10%, this grid point is marked as a wind speed adjustment point, and its three-dimensional coordinates (650, 750, 100) and velocity deficit rate of 12.5% ​​are recorded and added to the adjustment point list.

[0091] It is worth noting that, based on the three-dimensional positions of the adjustment points in the adjustment point list, the incoming wind speed of all adjustment points is found in the corrected wind field. ,pass ×(1- The adjusted wind speed value was calculated. The process involves replacing the original wind speed at the corresponding adjustment point in the corrected wind field with the adjusted wind speed value. For non-adjustment points, the original wind speed in the corrected wind field remains unchanged. After updating the wind speeds at all adjustment points, the final wind field is generated. For example, based on the adjustment point list, an adjustment point with an incoming wind speed of 8.0 m / s and a marked loss rate of 12.5% ​​is found in the corrected wind field. The calculated adjusted wind speed is 7.2 m / s. The original wind speed value of 8.0 m / s at this grid point in the corrected wind field is replaced with 7.2 m / s, completing the wind speed update for that point. The same calculation and replacement operation is performed on all points in the adjustment point list to obtain the updated final wind field.

[0092] In step S14, all wind turbines in the final wind field with wind speeds exceeding a preset wind speed threshold are identified and integrated into a normal wind turbine set. A polynomial is used to fit the historical wind speed and power variation patterns of this normal wind turbine set to obtain variation curves, including:

[0093] In the final wind field, for each wind turbine, the wind speed corresponding to the grid point closest to the wind turbine's location information is searched as the wind turbine's wind speed, thus obtaining a wind speed sequence;

[0094] If the wind speed of a fan in the wind speed sequence is greater than a preset wind speed threshold, then the fan is marked as a normal fan, and a set of normal fans is obtained.

[0095] The historical wind speed and power of each wind turbine in the set of normal wind turbines are obtained, and the variation patterns of the wind speed and power are fitted using a polynomial to obtain the variation curve.

[0096] It is worth noting that in the final wind farm, each wind turbine, based on its location information (hub center coordinates), searches for the nearest grid node in the final wind farm grid. This node has the smallest Euclidean distance to the hub center coordinates, and the wind speed at that node is taken as the wind speed of that turbine. Integrating the wind speeds of all turbines yields the wind speed sequence. For example, a wind farm contains three wind turbines with hub center coordinates of turbine 1 (512, 786, 100), turbine 2 (628, 915, 100), and turbine 3 (403, 851, 100). Searching for the nearest grid node for each turbine in the final wind farm, the Euclidean distance from turbine 1 to its surrounding grid nodes is calculated. The nearest node is found to be (500, 800, 100), approximately 18.4 meters away. The wind speed at the node is 7.6 m / s. The nearest grid node to wind turbine No. 2 is (650, 900, 100), which is about 25.2 m away. The wind speed at this node is 8.1 m / s. The nearest grid node to wind turbine No. 3 is (400, 850, 100), which is about 3.2 m away. The wind speed at this node is 7.3 m / s. By integrating the wind speeds of the three wind turbines, the wind speed sequence at this moment is [7.6 m / s, 8.1 m / s, 7.3 m / s].

[0097] It's worth noting that the preset wind speed threshold is 3 m / s. This threshold is determined based on the wind speed characteristics of the wind turbines and the requirements for identifying their operating status. In historical operating data, 95% of the samples showed that when the wind speed was below 3 m / s, the wind turbines were either not running or in a low-wind-speed standby state, with power output mostly at or near zero, failing to reflect the true power generation characteristics of the wind turbines within the effective wind speed range. Therefore, this threshold is set to skip this step if the wind speed does not meet the preset threshold. The wind speed value of each wind turbine in the wind speed sequence is iterated. If the wind speed value of a turbine is greater than the preset threshold, the turbine is marked as a normal turbine. All normal turbines are then integrated to obtain a set of normal turbines. For example, in a wind speed sequence [3.2, 3.5, 2.9, 3.1], the wind speeds of turbines 1, 2, and 4 are higher than the threshold, so turbines 1, 2, and 4 are marked as normal turbines.

[0098] It is worth noting that the historical operating data of each normal wind turbine in the set over the past year, i.e., the power data corresponding to the historical wind speed, is obtained. This power data is collected in real-time by the wind farm's SCADA system. Wind speed v (unit: m / s) is used as the independent variable, and power P (unit: kW) as the dependent variable. A third-order polynomial is constructed to fit the variation law of wind speed and power. All historical wind speed and corresponding power samples are input into this model. The mean square error between the fitted power and the measured power is minimized using the least squares method to obtain the coefficients of the third-order polynomial. Substituting these coefficients into the third-order polynomial yields the variation curve describing the wind speed and power variation law of that wind turbine. When using this variation curve to output power, the units of the input wind speed data must first be converted to the fitting stage, and only the numerical part should be used when substituting it into the model calculation to ensure that the input dimensions of the model are consistent with those of the training stage, avoiding calculation errors caused by dimensional mismatch. For example, the historical operating data from the past year contains 1000 sets of data on wind speed and power corresponding to normal operation of wind turbines. The wind speed range for normal operation is v ≥ 3.0 m / s. By minimizing the mean square error between the fitted power and the power using the least squares method, an optimal set of coefficients can be obtained. =0.025、 =-0.41、 =3.2、 =-5.8; Substituting these coefficients into the third-order polynomial, we obtain the variation curve as follows: .

[0099] In step S15, the wind speed of the wind turbines in the wind farm at various times is obtained and input into the change curve to obtain a power set. The power set is then input into a preset long short-term memory neural network model for prediction to obtain a predicted power set, including:

[0100] Obtain wind farm data for wind turbines within a preset time period, and extract wind speed at each moment within the preset time period;

[0101] Input the wind speed at each time point into the change curve to obtain the power at each time point, and integrate the time points and the power at each time point into a power set;

[0102] The power set is input into a preset long short-term memory neural network model for prediction to obtain the predicted power set.

[0103] It's worth noting that the preset time period is from 9 AM to 9 PM every day of the week. Following steps S11, S12, and S13, wind field data is generated every hour throughout the day. Simultaneously, based on the hub center coordinates of each wind turbine in the wind farm, the nearest grid node is searched within the corresponding wind field grid, and the wind speed value of that node is extracted as the wind speed of the turbine at that moment. By iterating through all time points within the preset time period, the wind speed data for each normal wind turbine at each moment can be obtained. For example, if the preset time period is from 9 AM to 9 PM every day from January 1st to January 7th, the same steps S11, S12, and S13 are used, generating the corresponding wind field data every hour throughout the day. The wind farm includes turbines No. 1 and No. 2. For wind turbine #1, the nearest node with a wind speed of 7.6 m / s was found in the wind field grid at 9:00 AM on the 1st, and 7.8 m / s at 10:00 AM. The wind speeds at all times were extracted sequentially, resulting in the wind speed sequence for wind turbine #1 on the 1st: [7.6, 7.8, 7.5, 7.9, 8.1, 7.7, 7.4, 7.6, 7.8, 7.9, 8.0, 8.2, 7.6]. This process was repeated to obtain the corresponding wind speeds for wind turbine #1 from the 1st to the 7th. Similarly, the wind speeds for wind turbine #2 were extracted at each time, completing the extraction of wind speeds for all normal wind turbines.

[0104] It is worth noting that the extracted wind speeds at each moment are input into the variation curve one by one. By substituting these values ​​into the variation curve, the power value of each wind turbine at the corresponding moment is calculated. Then, the time points and the calculated power values ​​are integrated, and the power values ​​are labeled with time points in chronological order, thus forming a power set. For example, the variation curve might be... Substituting the wind speed of 7.6 m / s at 9:00 AM on the 1st into the formula, the power is calculated to be 5.81 kW. Substituting the wind speed of 7.8 m / s at 10:00 AM, the power is calculated to be 6.08 kW. The power values ​​corresponding to the wind speed at all times on the 1st are calculated sequentially, resulting in the power sequence of wind turbine No. 1 on the 1st as [5.81, 6.08, 6.81, 7.83, 8.72, 5.98, 6.21, 7.63, 6.43, 7.08, 6.35, 6.02, 7.22] kW. By analogy, the power values ​​of wind turbine No. 1 at each time from the 1st to the 7th are calculated. Similarly, the power values ​​corresponding to the wind speed at each time are calculated for wind turbine No. 2. The time points such as "January 1st 9:00 - 5.81kW" and "January 1st 10:00 - 6.08kW" are integrated with their corresponding power values ​​one by one. Each power value is marked with its specific time point in chronological order to form a power set including wind turbines No. 1 and No. 2.

[0105] It is worth noting that this invention employs a Long Short-Term Memory (LSTM) neural network to construct the prediction model. The input layer of this network receives a time-series sequence of power values, using the power values ​​at each consecutive time step as the input feature vector. The hidden layer contains multiple LSTM units, each of which selectively updates, maintains, or outputs its cell state through input gates, forget gates, and output gates, thereby effectively capturing long-term dependencies and time-varying features in the power sequence. At each time step, the current input and the hidden state from the previous time step undergo gating computation and state transformation to generate the current hidden state and output. This process iterates along the time steps, gradually extracting the dynamic patterns of the sequence. The output layer maps the predicted power values ​​for future time steps through a fully connected layer. The training dataset is derived from the wind turbine power time series data in the historical operation data collected from wind farms. During the training process, key hyperparameters are configured, with the input sequence time step set to 84 steps and the output prediction step set to 12 steps. The iterative training is set to 100 rounds, with the root mean square error as the loss function. The Adam optimizer is used to dynamically adjust the network parameters. During training, the network parameters are adaptively updated using the optimizer through gradients. After each round of iteration, the iteration loss is calculated. If the loss on the validation set does not decrease significantly for three consecutive rounds, an early stopping mechanism is triggered to avoid overfitting of the model, until 100 rounds of iteration are completed or the model converges early. At the same time, the 5-fold cross-validation method is used to validate and optimize the model. For example, inputting the power set [5.81, 6.08, 6.81, 7.83, 8.72, 5.98, 6.21, 7.63, 6.43, 7.08, 6.35, 6.02, 7.22, ...] into an LSTM model will output the predicted power values ​​for January 8th from 9:00 AM to 2:00 PM as [6.56, 7.51, 7.08, 6.68, 6.15, 5.09] kW. By mapping the predicted time points to the corresponding predicted power values ​​and organizing them in chronological order, we obtain the predicted power set.

[0106] In step S16, the affected wind turbines in the predicted power set are obtained, and all wake source wind turbines corresponding to the affected wind turbines are also obtained and the comprehensive loss is calculated. The predicted power set is then corrected and smoothed based on the comprehensive loss to obtain the final power set, including:

[0107] Obtain the wind direction angle at each moment in the predicted power set, calculate the wake influence range of the wind turbines in the wind farm based on the wind direction angle, and obtain the affected wind turbines;

[0108] Identify the wake source fans that cause wake effects on the affected fans, calculate the overall loss caused by the wake source fans to the affected fans, and correct the predicted power set based on the overall loss to obtain the corrected power set.

[0109] The corrected power set is then smoothed to obtain the final power set.

[0110] It is worth noting that, following the method in step S13, the dominant incoming wind direction (i.e., wind direction angle) of the wind field at each moment in the predicted power set is obtained. Then, following the method in step S13 to divide the wake influence range, the wake influence range of each wind turbine in the wind farm is obtained. Wind turbines within the wake influence range are marked as influencing turbines. For example, the wind direction angle at 9:00 AM on January 8th in the predicted power set is 162°. The wake influence range of turbine 1 is defined as 0~2400m downstream and -20m~+120m vertically along the 162° direction. Since the area below -20m to 0m is underground space and has no actual wind field significance, only the effective height range above the ground surface is used in the calculation. The underground area is directly truncated and not included in the subsequent analysis process. Turbine 2 (628,915,100) falls within this range, therefore turbine 2 is determined to be an influencing turbine of turbine 1 at 9:00 AM on January 8th.

[0111] It is worth noting that for each influencing wind turbine, the wind turbines in the wind farm that generate wake effects on the influencing turbine are counted as wake source turbines. If an influencing turbine is affected by multiple wake source turbines, the wind speeds of the multiple wake source turbines are recorded. Based on the Jensen wake model, the wind speed at which each wake source turbine generates a wake effect on the influencing turbine is calculated. ,in The wind speed of the wake source fan. This is the thrust coefficient of the wind turbine, with a default value of 0.8. The radius of the wind turbine impeller. Where is the distance between the wake source fan and the affected fan, and k is the wake diffusion coefficient, with a default value of 0.05. and The ratio of the wake influence factor is obtained. Calculate the loss caused by each wake source fan to the affected fans. According to the formula Calculate the overall loss D, and finally obtain the corrected power. The original power values ​​in the predicted power set are replaced with the corrected power values, while the power values ​​of the remaining turbines unaffected by the wake remain unchanged. This update forms the corrected power set. For example, the predicted power value of turbine No. 2 at 9:00 AM on January 8th is 8.60 kW. This turbine is only affected by the wake of turbine No. 1. Using the Jensen model, the calculated overall loss is 0.08, so the corrected power value is 7.91 kW. The 8.60 kW value for turbine No. 2 at 9:00 AM on January 8th in the predicted power set is replaced with 7.91 kW, resulting in the corrected power set containing the corrected power values ​​of all turbines.

[0112] It is worth noting that the power sequence in the modified power set is smoothed by a moving average filtering method. Specifically, a fixed window size is set, such as 5 consecutive time periods. The average power value within the window is calculated for each time period. The original power value is replaced with this average value. The power values ​​of all time periods in the modified power set are traversed to complete the filtering, thereby eliminating random fluctuations and noise in the power sequence and finally obtaining the smoothed power sequence, which is the final power set. For example, the power sequence from 9:00 to 13:00 on January 8th in the corrected set of wind turbine No. 2 is [8.58, 9.75, 8.89, 9.36, 10.02]. With the moving average window size set to 3, the calculated power is 9.07kW at 9:00, 9.33kW at 10:00, and 9.42kW at 11:00. Smoothing calculations are performed for all times in sequence to obtain the filtered sequence [9.07, 9.33, 9.42, 9.46, 9.79]kW. This sequence is the final power set for wind turbine No. 2 during the corresponding time period. Similarly, after processing all wind turbines, the overall final power set is obtained.

[0113] In summary, this invention discloses a short-term power load forecasting method for wind farms based on digital twins. It accurately depicts the real distribution of wind speed within the wind farm by considering multiple factors such as meteorological conditions, terrain characteristics, and wake effects, resulting in a realistic and accurate wind farm representation. Furthermore, by employing a polynomial fitting method to deeply mine historical wind speed and power data of normal wind turbines, it can fully characterize the nonlinear output characteristics of wind turbines in different wind speed ranges. Based on the variation curves, the power set of the wind farm turbines is obtained. The final power set not only conforms to the actual power time-series variation pattern but also reduces the output uncertainty caused by wake interference, improving the accuracy of wind farm power forecasting. This provides reliable technical support for the efficient operation of wind farms and overcomes the shortcomings of existing technologies that cannot accurately characterize wind farms and accurately reflect wind turbine power changes.

[0114] Reference Figure 2 The second embodiment of the present invention provides a short-term power load forecasting system for wind farms based on digital twins, comprising:

[0115] The initial wind field generation module is used to acquire meteorological data, perform layered processing on the meteorological data to obtain layered data, map the layered data to a three-dimensional grid and perform iterative evolution to generate an updated three-dimensional wind field.

[0116] The wind field correction module is used to construct a spatial discrete wind field based on the updated three-dimensional wind field and the acquired wind farm data and wind turbine location information, locate the wind turbine action area in the spatial discrete wind field, calculate the momentum source term of the wind turbine action area, and correct the spatial discrete wind field based on the momentum source term to obtain the corrected wind field.

[0117] The final wind field generation module is used to determine the wake influence range of the corrected wind field, calculate the velocity loss rate of each grid point in the wake influence range, mark the grid points whose velocity loss rate is greater than a preset loss rate threshold as adjustment points, and adjust the corrected wind field according to the adjustment points to obtain the final wind field.

[0118] The regularity fitting module is used to identify all wind turbines in the final wind field whose wind speed is greater than a preset wind speed threshold and integrate them into a normal wind turbine set. The module uses a polynomial to fit the historical wind speed and power variation patterns of the normal wind turbine set to obtain the variation curve.

[0119] The power prediction module is used to obtain the wind speed of the wind turbines in the wind farm at various times and input the change curve to obtain the power set. The power set is then input into a preset long short-term memory neural network model for prediction to obtain the predicted power set.

[0120] The final power generation module is used to acquire the affected wind turbines in the predicted power set, acquire all the wake source wind turbines corresponding to the affected wind turbines and calculate the comprehensive loss, and correct and smooth the predicted power set according to the comprehensive loss to obtain the final power set.

[0121] It should be noted that the short-term power load forecasting system for wind farms based on digital twins provided in this embodiment of the invention is used to execute all the process steps of the short-term power load forecasting method for wind farms based on digital twins in the above embodiments. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0122] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

[0123] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps described in the method embodiments above.

[0124] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0125] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0126] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the electronic device through various interfaces and lines.

[0127] The memory can be used to store the computer programs and modules. The processor implements various functions of the electronic device by running or executing the computer programs and modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, SmartMediaCard (SMC), Secure Digital (SD) card, FlashCard, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0128] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0129] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0130] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for short-term power load forecasting of wind farms based on digital twins, characterized in that, include: Meteorological data is acquired, and the meteorological data is processed into layers to obtain layered data. The layered data is then mapped onto a three-dimensional grid and iteratively evolved to generate an updated three-dimensional wind field. Based on the updated three-dimensional wind field and the acquired wind farm data and wind turbine location information, a spatial discrete wind field is constructed. The wind turbine action area in the spatial discrete wind field is located, the momentum source term of the wind turbine action area is calculated, and the spatial discrete wind field is corrected based on the momentum source term to obtain the corrected wind field. Determine the wake influence range of the corrected wind field, calculate the velocity loss rate of each grid point in the wake influence range, mark the grid points whose velocity loss rate is greater than a preset loss rate threshold as adjustment points, and adjust the corrected wind field according to the adjustment points to obtain the final wind field; In the final wind field, identify all wind turbines whose wind speed is greater than a preset wind speed threshold and integrate them into a normal wind turbine set. Use a polynomial to fit the historical wind speed and power variation patterns of the normal wind turbine set to obtain the variation curves. The wind speed of the wind turbines in the wind farm at various times is obtained and the change curve is input to obtain the power set. The power set is then input into a preset long short-term memory neural network model for prediction to obtain the predicted power set. The predicted power set is obtained by identifying the wind turbines affected by the wake, and all wake source wind turbines corresponding to the affected wind turbines are also identified and their combined loss is calculated. The predicted power set is then corrected and smoothed based on the combined loss to obtain the final power set.

2. The method for short-term power load forecasting of wind farms based on digital twins according to claim 1, characterized in that, The process of acquiring meteorological data, performing layered processing on the meteorological data to obtain layered data, mapping the layered data onto a three-dimensional mesh and iteratively evolving it to generate an updated three-dimensional wind field includes: Meteorological forecast values ​​are obtained, and at the same time, measured values ​​are obtained from ground-based meteorological data. The meteorological forecast values ​​are then corrected based on the measured values ​​to obtain corrected meteorological values. The Monin-Obukhov length is calculated for the corrected meteorological value. Based on the Monin-Obukhov length, the atmospheric stability of the corrected meteorological value is assessed and stratified to obtain stratified data. The vertical wind profiles at each location are calculated based on the layered data and mapped onto a three-dimensional grid to form an initial three-dimensional wind field. The initial three-dimensional wind field is iteratively evolved using a preset numerical solver to obtain an updated three-dimensional wind field.

3. The method for short-term power load forecasting of wind farms based on digital twins according to claim 1, characterized in that, The process of constructing a spatially discrete wind field based on the updated three-dimensional wind field and the acquired wind farm data and wind turbine location information, locating the wind turbine operating area in the spatially discrete wind field, calculating the momentum source term of the wind turbine operating area, and correcting the spatially discrete wind field based on the momentum source term to obtain a corrected wind field includes: Acquire wind farm data, wind turbine location information, and wind turbine blade geometry data; Based on the updated three-dimensional wind field, the wind farm data, and the wind turbine location information, a spatially discrete wind field is constructed. The blade sweep area is calculated based on the wind turbine location information and the wind turbine blade geometry data. The blade sweep area is then mapped to the spatial discrete wind field to obtain the wind turbine action grid. The momentum source term of the wind turbine action grid is calculated using a preset actuation disk model. The spatial discrete wind field and the momentum source term are then input into a preset numerical solver to obtain the corrected wind field.

4. The method for short-term power load forecasting of wind farms based on digital twins according to claim 1, characterized in that, The process of determining the wake influence range of the modified wind field, calculating the velocity deficit rate of each grid point within the wake influence range, marking grid points with velocity deficit rates greater than a preset deficit rate threshold as adjustment points, and adjusting the modified wind field according to the adjustment points to obtain the final wind field includes: The dominant incoming wind direction is determined based on the modified wind field, and the wake influence range is determined based on the dominant incoming wind direction. Calculate the velocity loss rate of each grid point in the wake influence range. If the velocity loss rate is greater than a preset loss rate threshold, mark the grid point as an adjustment point to obtain an adjustment point list. In the corrected wind field, find the adjustment point in the list of adjustment points, adjust the wind speed of the adjustment point and update the wind speed value to obtain the final wind field.

5. The method for short-term power load forecasting of wind farms based on digital twins according to claim 1, characterized in that, The process involves identifying all wind turbines in the final wind field whose wind speeds exceed a preset wind speed threshold and integrating them into a normal wind turbine set. A polynomial is then used to fit the historical wind speed and power variation patterns of this normal wind turbine set, resulting in variation curves, including: In the final wind field, for each wind turbine, the wind speed corresponding to the grid point closest to the wind turbine's location information is searched as the wind turbine's wind speed, thus obtaining a wind speed sequence; If the wind speed of a fan in the wind speed sequence is greater than a preset wind speed threshold, then the fan is marked as a normal fan, and a set of normal fans is obtained. The historical wind speed and power of each wind turbine in the normal wind turbine set are obtained, and the variation patterns of wind speed and power are fitted using a polynomial to obtain the variation curve.

6. The method for short-term power load forecasting of wind farms based on digital twins according to claim 1, characterized in that, The process involves acquiring the wind speed of the wind turbines at various times and inputting it into the change curve to obtain a power set. This power set is then input into a preset long short-term memory neural network model for prediction, resulting in a predicted power set, including: Obtain wind farm data for wind turbines within a preset time period, and extract wind speed at each moment within the preset time period; Input the wind speed at each time point into the change curve to obtain the power at each time point, and integrate the time points and the power at each time point into a power set; The power set is input into a preset long short-term memory neural network model for prediction to obtain the predicted power set.

7. The method for short-term power load forecasting of wind farms based on digital twins according to claim 1, characterized in that, The process involves acquiring the affected wind turbines from the predicted power set, simultaneously acquiring all wake source wind turbines corresponding to the affected turbines and calculating the overall loss, and then correcting and smoothing the predicted power set based on the overall loss to obtain the final power set, including: Obtain the wind direction angle at each moment in the predicted power set, calculate the wake influence range of the wind turbines in the wind farm based on the wind direction angle, and obtain the affected wind turbines; Identify the wake source fans that cause wake effects on the affected fans, calculate the overall loss caused by the wake source fans to the affected fans, and correct the predicted power set based on the overall loss to obtain the corrected power set. The corrected power set is then smoothed to obtain the final power set.

8. A system using the short-term power load forecasting method for wind farms based on digital twins according to any one of claims 1 to 7, characterized in that, include: The initial wind field generation module is used to acquire meteorological data, perform layered processing on the meteorological data to obtain layered data, map the layered data to a three-dimensional grid and perform iterative evolution to generate an updated three-dimensional wind field. The wind field correction module is used to construct a spatial discrete wind field based on the updated three-dimensional wind field and the acquired wind farm data and wind turbine location information, locate the wind turbine action area in the spatial discrete wind field, calculate the momentum source term of the wind turbine action area, and correct the spatial discrete wind field based on the momentum source term to obtain the corrected wind field. The final wind field generation module is used to determine the wake influence range of the corrected wind field, calculate the velocity loss rate of each grid point in the wake influence range, mark the grid points whose velocity loss rate is greater than a preset loss rate threshold as adjustment points, and adjust the corrected wind field according to the adjustment points to obtain the final wind field. The regularity fitting module is used to identify all wind turbines in the final wind field whose wind speed is greater than a preset wind speed threshold and integrate them into a normal wind turbine set. The module uses a polynomial to fit the historical wind speed and power variation patterns of the normal wind turbine set to obtain the variation curve. The power prediction module is used to obtain the wind speed of the wind turbines in the wind farm at various times and input the change curve to obtain the power set. The power set is then input into a preset long short-term memory neural network model for prediction to obtain the predicted power set. The final power generation module is used to acquire the affected wind turbines in the predicted power set, acquire all the wake source wind turbines corresponding to the affected wind turbines and calculate the comprehensive loss, and correct and smooth the predicted power set according to the comprehensive loss to obtain the final power set.

Images