Mesoscale-microscale dynamics-based wind power prediction method and system

Abstract

The invention discloses a wind power prediction method and system based on mesoscale-microscale dynamics, and the method comprises the following steps: constructing a wind power plant microscale aerodynamic model, setting an inflow boundary condition, generating data according to the inflow boundary condition and the wind power plant microscale aerodynamic model, and constructing a database; constructing a mesoscale-microscale aerodynamic coupling model and a prediction network of the wind power plant; taking the mesoscale-microscale aerodynamic coupling model of the wind power plant as a system environment, training a prediction network by using a database, and inputting mesoscale meteorological prediction data and corresponding inflow boundary conditions into the trained prediction network to obtain a wind power prediction result; compared with an existing method, the method is more beneficial to meeting the requirement of wind power prediction, the reliability of wind power prediction is improved, and the speed and precision of wind power prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and specifically to a wind power prediction method and system based on mesoscale-microscale dynamics. Background Art

[0002] With the continuous expansion of the scale of wind power grid connection, the utilization of wind power is gradually developing towards complex environments. Under the combined influence of complex terrains, variable wind fields full of random volatility, etc., the effective utilization and evaluation of wind power resources pose higher requirements for wind farm power prediction. In wind power prediction, the accuracy of meteorological prediction is crucial, and the acquisition of meteorological data is restricted by on-site observations, resulting in low prediction reliability. Accurate wind power prediction can provide reliable wind power data for the power system, while existing methods cannot meet the accuracy and speed requirements of wind farm power prediction in complex environments. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the present invention provides a wind power prediction method and system based on mesoscale-microscale dynamics, aiming to solve the problems in the background art.

[0004] To achieve the above object, the present invention provides the following technical solution: A wind power prediction method based on mesoscale-microscale dynamics, comprising the following steps: Step S1: Construct a microscale aerodynamic model of the wind farm and complete the extraction of relevant features required for model input; Step S2: Set the inflow boundary conditions, and perform simulation based on the inflow boundary conditions and the microscale aerodynamic model of the wind farm to generate turbulence, wind acceleration factors, wind directions, and powers at the positions of each wind turbine. Construct a database based on the generated turbulence, wind acceleration factors, wind directions, and powers at the positions of each wind turbine; Step S3: Construct a mesoscale meteorological prediction model, use the mesoscale meteorological prediction model as the initial condition of the microscale aerodynamic model of the wind farm, and construct a mesoscale-microscale aerodynamic coupling model of the wind farm; Step S4: Combine the block processing technology, time series prediction, and convolutional neural network to construct a wind farm time series prediction PatchTST-CNN neural network; Step S5: Take the meso - microscale aerodynamic coupling model of the wind farm as the system environment, and use the database to train the PatchTST - CNN neural network for wind farm time - series prediction to obtain the trained PatchTST - CNN neural network for wind farm time - series prediction; obtain the mesoscale meteorological prediction data, i.e., the output of the mesoscale meteorological prediction model, and calculate the inflow boundary conditions of the mesoscale meteorological prediction data for the wind farm. Then, input the mesoscale meteorological prediction data and the corresponding inflow boundary conditions into the trained PatchTST - CNN neural network for wind farm time - series prediction to obtain the wind power prediction result.

[0005] Furthermore, adopt the large - eddy simulation method to construct the microscale aerodynamic model of the wind farm: Filter the cubic space with a scale size of to eliminate the turbulence with a scale smaller than , which is expressed as: ; In the formula, represents the filtered turbulence, i.e., the large - scale turbulence; represents the spatial coordinate; represents the time; represents the turbulence variable; represents the integral variable, including the streamwise component , the spanwise component and the vertical component ; represents the filtering kernel function; represents the differential operator; Establish the control equations describing the turbulence in the wind farm, including the continuity equation and the momentum equation, which are respectively expressed as: ; ; In the formula, , respectively represent the large - scale velocity components in the directions and , , representing the streamwise direction, the spanwise direction and the vertical direction respectively; , respectively represent the spatial coordinates in the directions and ; represents the air density; represents the large - scale pressure field; represents the kinematic viscosity; represents the body force; represents the sub - grid - scale stress; Model the wake of a wind turbine using an actuator line model; based on the blade element theory, along the span direction, the lift and drag forces on a single blade element are expressed as: ; ; In the formula, and represent the lift differential and drag differential along the span direction, respectively; and represent the lift coefficient calculation function and drag coefficient calculation function, respectively; represents the angle of attack; represents the length in the span direction; represents the chord length; represents the relative velocity; The relative velocity is expressed as: ; In the formula, and represent the radial velocity and tangential velocity, respectively; represents the angular velocity of blade rotation; represents the blade element radius; Map the lift and drag forces to the nearby grids using a sampled three-dimensional Gaussian distribution: ; In the formula, represents the value of the Gaussian distribution function, describing the distribution intensity at the given position ; represents the Gaussian adaptation coefficient; represents the distance between the center grid point of the flow field and other grid points.

[0006] Furthermore, the sub-grid stress is expressed as: ; In the formula, represents the product of the large-scale velocity components in the directions and ; The calculation method of the sub-grid stress is: ; ; In the formula, represents the tensor deformation rate of the large-scale turbulence; represents the sign function; represents the isotropic component; represents the sub-grid eddy viscosity coefficient; The sub-grid eddy viscosity coefficient Expressed as: ; In the formula, represents the Smagorinsky coefficient.

[0007] Furthermore, the relevant characteristics required for the input of the micro-scale aerodynamic model of the wind farm include: The thermal stability of the inflow boundary condition, wind speeds at different heights, wind directions, atmospheric pressure, and ground roughness; The topographic map of the wind farm and the positions of the wind turbines for large-eddy simulation; The parameters of the wind turbines for the actuator line model, including blade shape parameters, hub height, cut-in wind speed, and cut-out wind speed; The output of the micro-scale aerodynamic model of the wind farm is the turbulence, wind acceleration factor, wind direction, and power at the positions of each wind turbine.

[0008] Furthermore, the Monin-Obukhov similarity theory is used to set the inflow boundary condition; The inflow boundary condition under constant shear stress is: ; In the formula, represents the turbulent shear stress; represents the air density; represents the height; represents the height The corresponding wind speed; represents the turbulent kinetic energy; represents the dissipation rate; represents a constant; For the correction of the turbulent kinetic energy along the height, the inflow boundary condition with gradually changing shear stress is: ; In the formula, represents the local atmospheric boundary layer thickness; represents the friction velocity; The wall inflow boundary condition is: ; In the formula, represents the wind speed at the center position of the cell adjacent to the wall; represents the height at the center position of the cell adjacent to the wall; represents the von Kármán constant; represents a dimensionless constant; represents the roughness constant; represents the roughness length; represents the shear stress at the wall; The upper surface inflow boundary condition is: ; In the formula, represents the equivalent volume force applied to the cells in the layer adjacent to the upper top surface; represents the additional shear stress at the upper top surface; represents the height of the cell adjacent to the lower bottom surface.

[0009] Furthermore, the mesoscale meteorological prediction model includes a data preprocessing module, a prediction module, a prediction result postprocessing module, and a visualization module. The inputs of the mesoscale meteorological prediction model include the meteorological prediction area, terrain data, grid node data, and interpolated meteorological data; The control equation of the mesoscale meteorological prediction model is: ; In the formula, represents the partial derivative with respect to time ; represents the component of the velocity field in the direction; represents the component of the velocity field in the direction; represents the wind speed component in the direction; represents the partial derivative with respect to ; represents the change of the geopotential height in the normal direction; represents the change of the geopotential height in the direction; represents the forcing term caused by physical processes; represents the wind speed component in the direction; represents the partial derivative with respect to ; represents the change of the geopotential height in the direction; represents the forcing term caused by perturbation mixing; represents the component of the velocity field in the direction; represents the wind speed component in the direction; represents the partial derivative with respect to ; represents the forcing term caused by spherical projection; represents the momentum of the potential temperature field; represents the potential temperature; represents the forcing term caused by the potential temperature; represents the geopotential.

[0010] Furthermore, the processing flow of the PatchTST-CNN neural network for wind farm time series prediction is as follows: Divide the input of the PatchTST-CNN neural network for wind farm time series prediction into multiple small blocks; Perform feature mapping on the small blocks to obtain latent space feature vectors, expressed as: ; In the formula, represents the feature mapping matrix; represents the position encoding matrix; represents the th latent space feature vector obtained after feature mapping of the th small block; represents the th small block; Process the latent space feature vectors through the multi-head attention mechanism to obtain attention features;

[0011] The wind power prediction system based on mesoscale-microscale dynamics includes: A wind farm microscale aerodynamic model construction module, which is used to construct a wind farm microscale aerodynamic model and complete the extraction of relevant features required for model input; An inflow boundary condition setting and database construction module, which is used to set the inflow boundary conditions, perform simulation based on the inflow boundary conditions and the wind farm microscale aerodynamic model, generate the turbulence, wind acceleration factor, wind direction, and power at the positions of each wind turbine, and construct a database based on the generated turbulence, wind acceleration factor, wind direction, and power at the positions of each wind turbine; A coupled model construction module, which is used to construct a mesoscale meteorological prediction model, use the mesoscale meteorological prediction model as the initial condition of the wind farm microscale aerodynamic model, and construct a wind farm mesoscale-microscale aerodynamic coupled model; A prediction network construction module, which is used to construct a PatchTST-CNN neural network for wind farm time series prediction by combining block processing technology, time series prediction, and convolutional neural network; The pre-training and prediction module is used to take the mesoscale-microscale aerodynamic coupling model of the wind farm as the system environment, train the PatchTST-CNN neural network for wind farm time series prediction using the database, and obtain the trained PatchTST-CNN neural network for wind farm time series prediction; obtain the mesoscale meteorological prediction data, i.e., the output of the mesoscale meteorological prediction model, calculate the inflow boundary conditions of the mesoscale meteorological prediction data for the wind farm, and then input the mesoscale meteorological prediction data and the corresponding inflow boundary conditions into the trained PatchTST-CNN neural network for wind farm time series prediction for prediction.

[0012] An electronic device includes a processor, a memory, and a bus. The processor and the memory are connected through the bus. Among them, the memory is used to store a set of program codes, and the processor is used to call the program codes stored in the memory to execute the wind power prediction method based on mesoscale-microscale dynamics.

[0013] A non-volatile computer storage medium stores computer-executable instructions, and these computer-executable instructions execute the wind power prediction method based on mesoscale-microscale dynamics.

[0014] Compared with the existing technologies, the present invention has the following beneficial effects: The present invention constructs a mesoscale-microscale aerodynamic coupling model for the wind farm. At the microscale aerodynamic level, a large eddy simulation (LES) method is used to construct a wind farm microscale aerodynamic model, effectively realizing the simulation of the complex terrain environment of the wind farm and enhancing the accuracy of the wind farm numerical simulation; at the mesoscale atmospheric dynamics level, a mesoscale meteorological prediction model is constructed, breaking the limitation of on-site observation of wind farm meteorological data. The construction of the mesoscale-microscale aerodynamic coupling model for the wind farm integrates the atmospheric dynamic characteristics at different scales of the wind farm, realizes the two-way coupling at multiple scales of the wind farm, enhances the overall consistency at multiple scales of the wind farm, reduces the error accumulation between different scales, and significantly improves the accuracy of wind power prediction. The present invention constructs a power prediction model based on the PatchTST-CNN neural network, learns based on a large amount of data, integrates the temporal and spatial features of the data, and improves the robustness of the prediction. Compared with the existing methods, the present invention is more conducive to meeting the requirements of wind power prediction, improves the reliability of wind power prediction, and enhances the speed and accuracy of wind power prediction. Description of the Drawings

[0015] Figure 1 It is the flowchart of the method of the present invention. Detailed Embodiments

[0016] As Figure 1As shown in the figure, the present invention provides a technical solution: a wind power prediction method based on mesoscale-microscale dynamics, comprising the following steps:

[0017] Step S1: Construct a microscale aerodynamic model of a wind farm and complete the extraction of relevant features required for model input.

[0018] The present invention adopts the large eddy simulation method (LES) which is between direct numerical simulation and Reynolds-averaged method for microscale aerodynamic simulation of a wind farm.

[0019] The large eddy simulation method decomposes turbulence into large-scale turbulence that significantly affects the mean flow and small-scale turbulence that dissipates turbulent variables. Among them, the large-scale turbulence can be solved by the Navier-Stokes method, and the small-scale turbulence can be simulated by the sub-grid scale model.

[0020] Filter a cubic space with a scale size of to eliminate the turbulence with a scale smaller than , which is expressed as: ; In the formula, represents the filtered turbulence, that is, the large-scale turbulence; represents the spatial coordinate; represents the time; represents the turbulent variable; represents the integral variable, including the streamwise component , the spanwise component and the vertical component ; represents the filtering kernel function; represents the differential operator.

[0021] The turbulence in the wind farm belongs to an incompressible fluid. Establish the control equations describing the turbulence in the wind farm, including the continuity equation and the momentum equation, which can be respectively expressed as: ; ; In the formula, , respectively represent the large-scale velocity components in the direction and the direction , , respectively represent the streamwise direction, the spanwise direction and the vertical direction; , respectively represent the spatial coordinates in the direction and the direction ; represents the air density; represents the large-scale pressure field; represents the kinematic viscosity; represents the body force; represents the sub-grid stress.

[0022] Sub-grid stress can be expressed as: ; In the formula, represents the direction and the direction is the product of the large-scale velocity components in the direction.

[0023] The sub-grid stress characterizes the momentum transport of small-scale fluctuations to large-scale turbulent motion and can be calculated using Smagorinsky's eddy viscosity model: ; ; In the formula, represents the tensor deformation rate of large-scale turbulence; represents the sign function; represents the isotropic component; represents the sub-grid eddy viscosity coefficient.

[0024] Sub-grid eddy viscosity coefficient can be expressed as: ; In the formula, represents the Smagorinsky coefficient, which determines the influence degree of the sub-grid stress on large-scale turbulence.

[0025] The actuator line model is used to model the wind turbine wake. It can consider the aerodynamic characteristics of the wind turbine blades, the wake effect between wind turbines, and has a moderate computational complexity. Based on the blade element theory, along the span direction, the lift and drag forces on a single blade element can be calculated as follows: ; ; In the formula, , respectively represent the lift differential and drag differential along the span direction; , respectively represent the lift and drag coefficient calculation functions; represents the angle of attack; represents the length in the span direction; represents the chord length; represents the relative velocity.

[0026] Relative velocity It can be expressed as: ; In the formula, , respectively represent the radial and tangential velocities; represents the angular velocity of blade rotation; represents the blade element radius.

[0027] To avoid numerical oscillations, this patent selects a three-dimensional Gaussian distribution to map the above lift and drag to the nearby grids, and adds it as a source term to the Navier-Stokes equation. The Gaussian distribution function is as follows: ; In the formula, represents the value of the Gaussian distribution function, describing the distribution intensity at the given position ; represents the Gaussian adaptation coefficient; represents the distance between the center grid point of the flow field and other grid points.

[0028] For the wind farm power prediction problem, the relevant features required for the input of the micro-scale aerodynamic model of the wind farm include: the thermal stability of the inflow boundary condition, the wind speed, wind direction, atmospheric pressure, and ground roughness at different heights; the topographic map of the wind farm and the fan positions for large eddy simulation; the fan parameters of the actuator line model, including blade shape parameters, hub height, cut-in wind speed, and cut-out wind speed.

[0029] The output of the micro-scale aerodynamic model of the wind farm is the turbulence, wind acceleration factor, wind direction, and power at the positions of each fan.

[0030] Step S2: Set the inflow boundary condition, and perform simulation according to the inflow boundary condition and the micro-scale aerodynamic model of the wind farm to generate the turbulence, wind acceleration factor, wind direction, and power at the positions of each fan, and construct a database based on the generated turbulence, wind acceleration factor, wind direction, and power at the positions of each fan.

[0031] When using the large eddy simulation (LES) method for micro-scale aerodynamic calculation of the wind farm, it should be ensured that the wind speed, turbulent kinetic energy, and its dissipation rate at the inlet maintain a horizontally uniform inflow boundary condition.

[0032] The Monin-Obukhov similarity theory is used to set the inflow boundary condition, which comprehensively considers the wind shear characteristics and turbulent intensity influence of the atmospheric boundary layer.

[0033] The wind profile, as one of the key parameters of the inflow boundary condition, can be expressed as: ; In the formula, represents the wind speed at height ; represents the friction velocity; represents the von Kármán constant; represents the surface roughness; represents the atmospheric stability correction function, represents the Monin-Obukhov length, which represents a stable atmosphere when and an unstable atmosphere when .

[0034] In the case of a stable atmosphere such as at night, the atmospheric stability correction function can be expressed as: ; In the case of an unstable atmosphere such as on a sunny day, the atmospheric stability correction function can be expressed as: ; In the case of a neutral atmosphere, the atmospheric stability correction function can be expressed as: .

[0035] The inflow boundary condition under a constant shear stress is: ; In the formula, represents the turbulent shear stress; represents the air density; represents the height; represents the height corresponding wind speed; represents the turbulent kinetic energy; represents the dissipation rate; represents the constant 0.99.

[0036] For the correction of the turbulent kinetic energy along the height and the inflow boundary condition with a gradually changing shear stress is: ; In the formula, represents the local atmospheric boundary layer thickness; represents the friction velocity.

[0037] The wall inflow boundary condition is: ; In the formula, represents the wind speed at the center position of the cell adjacent to the wall; represents the height at the center position of the cell adjacent to the wall; represents the von Kármán constant, taking 0.4187; represents the dimensionless constant, taking 9.793; represents the roughness constant, with a value ranging from 0.5 to 1; represents the roughness length; represents the shear stress at the wall surface.

[0038] The inflow boundary condition for the upper top surface is: ; In the formula, represents the equivalent body force applied in the cells adjacent to the upper top surface; represents the additional shear stress at the upper top surface; represents the height of the cells adjacent to the lower bottom surface.

[0039] For the microscale aerodynamic model of the wind farm and the designed inflow boundary conditions, the open-source OpenFOAM software is used for simulation, and calibration is carried out through the data of the wind turbine laboratory. The simulation data is stored using the distributed database technology MongoDB, and the hash index is used to improve the retrieval speed. Based on the large amount of microscale aerodynamic turbulence and power of the wind farm considering the boundary conditions generated by the simulation, a database is constructed.

[0040] Step S3: Construct a mesoscale meteorological prediction model, use the mesoscale meteorological prediction model as the initial condition of the microscale aerodynamic model of the wind farm, and construct a mesoscale-microscale aerodynamic coupling model of the wind farm.

[0041] Among them, the mesoscale meteorological prediction model takes the meteorological observation data at a certain moment as the initial value and obtains the meteorological forecast by solving with numerical methods; the mesoscale meteorological prediction model includes a data preprocessing module, a prediction module, a prediction result postprocessing module, and a visualization module. The input of the mesoscale meteorological prediction model includes the meteorological prediction area, terrain data, grid node data, and interpolated meteorological data.

[0042] Among them, the governing equation of the mesoscale meteorological prediction model is: ; In the formula, represents the partial derivative with respect to time ; represents the component of the velocity field in the direction; represents the component of the velocity field in the direction; represents the wind speed component along the direction; represents the partial derivative with respect to ; represents the change of the geopotential height in the normal direction; Indicates the change in geopotential height in direction; Indicates the forcing term caused by physical processes; Indicates along wind speed component in the direction; Indicates the partial derivative with respect to ; Indicates the change in geopotential height in direction; Indicates the forcing term caused by perturbation mixing; Indicates component of the velocity field in the direction; Indicates along wind speed component in the direction; Indicates the partial derivative with respect to ; Indicates the mass of the air column per unit area of the model grid point; Indicates the forcing term caused by spherical projection; Indicates the momentum of the potential temperature field; Indicates the potential temperature; Indicates the forcing term caused by the potential temperature; Indicates the geopotential.

[0043] Using a mesoscale meteorological prediction model as the initial condition of the microscale aerodynamic model of a wind farm, a mesoscale-microscale aerodynamic coupling model of the wind farm is constructed to realize the coupled calculation of two different scale models.

[0044] Step S4: Combine the Patch technology, Time Series Forecasting, and Convolutional Neural Network (CNN) to construct the PatchTST-CNN neural network for wind farm time series prediction.

[0045] The processing flow of the PatchTST-CNN neural network for wind farm time series prediction is as follows:

[0046] Divide the input of the PatchTST-CNN neural network for wind farm time series prediction into multiple small patches (Patch).

[0047] Perform feature mapping on the small patches (Patch) to obtain the latent space feature vectors, denoted as: ; In the formula, Indicates the feature mapping matrix; Indicates the position encoding matrix; Indicates the th latent space feature vector obtained after feature mapping of the small patch (Patch); Indicates the th patch.

[0048] Process the latent space feature vector through the multi-head attention mechanism to obtain the attention feature.

[0049] Input the attention feature into the convolutional neural network for processing to obtain the wind power prediction result, which is the output of the PatchTST-CNN neural network for wind farm time series prediction.

[0050] Step S5: Use the mesoscale-microscale aerodynamic coupling model of the wind farm as the system environment, train the PatchTST-CNN neural network for wind farm time series prediction using the database to obtain a trained PatchTST-CNN neural network for wind farm time series prediction; obtain the mesoscale meteorological prediction data, which is the output of the mesoscale meteorological prediction model, and calculate the inflow boundary conditions for the wind farm for the mesoscale meteorological prediction data, and then input the mesoscale meteorological prediction data and the corresponding inflow boundary conditions into the trained PatchTST-CNN neural network for wind farm time series prediction to obtain the wind power prediction result.

[0051] Among them, the mean squared error (MSE) loss is used to measure the difference between the wind farm power prediction and the actual wind farm power value, and normalization processing is performed.

[0052] A wind power prediction system based on mesoscale-microscale dynamics includes:

[0053] A wind farm microscale aerodynamic model construction module, which is used to construct a wind farm microscale aerodynamic model and complete the extraction of relevant features required for model input.

[0054] An inflow boundary condition setting and database construction module, which is used to set the inflow boundary conditions, perform simulation based on the inflow boundary conditions and the wind farm microscale aerodynamic model to generate turbulence, wind acceleration factors, wind directions, and powers at the positions of each wind turbine, and construct a database based on the generated turbulence, wind acceleration factors, wind directions, and powers at the positions of each wind turbine.

[0055] A coupling model construction module, which is used to construct a mesoscale meteorological prediction model, use the mesoscale meteorological prediction model as the initial condition of the wind farm microscale aerodynamic model, and construct a wind farm mesoscale-microscale aerodynamic coupling model.

[0056] A prediction network construction module, which is used to construct a PatchTST-CNN neural network for wind farm time series prediction by combining the patch processing technology, time series forecasting, and convolutional neural network (CNN).

[0057] A pre-training and prediction module, which uses the meso-scale micro-scale aerodynamic coupling model in the wind farm as the system environment, trains the PatchTST-CNN neural network for wind farm time series prediction using a database, and obtains a trained PatchTST-CNN neural network for wind farm time series prediction; obtains meso-scale meteorological prediction data, i.e., the output of the meso-scale meteorological prediction model, calculates the inflow boundary conditions of the meso-scale meteorological prediction data for the wind farm, and then inputs the meso-scale meteorological prediction data and the corresponding inflow boundary conditions into the trained PatchTST-CNN neural network for wind farm time series prediction for prediction.

[0058] An electronic device includes a processor, a memory, and a bus. The processor and the memory are connected through the bus. Among them, the memory is used to store a set of program codes, and the processor is used to call the program codes stored in the memory to execute a wind power prediction method based on meso-scale micro-scale dynamics.

[0059] A non-volatile computer storage medium stores computer-executable instructions, and these computer-executable instructions execute a wind power prediction method based on meso-scale micro-scale dynamics.

[0060] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A wind power prediction method based on mesoscale-microscale dynamics, characterized in that: The steps include: Step S1: construct a micro-scale aerodynamic model of the wind farm and extract relevant features required for model input; Step S2: setting inflow boundary conditions, performing simulation according to the inflow boundary conditions and the micro-scale aerodynamic model of the wind farm, generating turbulence, wind acceleration factor, wind direction, and power at the location of each wind turbine, and constructing a database based on the generated turbulence, wind acceleration factor, wind direction, and power at the location of each wind turbine; Step S3: constructing a mesoscale meteorological prediction model, taking the mesoscale meteorological prediction model as the initial condition of the microscale aerodynamic model of the wind farm, and constructing a mesoscale-microscale aerodynamic coupling model of the wind farm; Step S4: combining block processing technology, time series prediction and convolutional neural network to construct a wind farm time series prediction PatchTST-CNN neural network; Step S5: Using the mesoscale-microscale aerodynamic coupling model of the wind farm as the system environment, and using the database to train the wind farm time series prediction PatchTST-CNN neural network to obtain a trained wind farm time series prediction PatchTST-CNN neural network; obtaining the mesoscale meteorological forecast data, i.e., the output of the mesoscale meteorological forecast model, and calculating the inflow boundary conditions of the mesoscale meteorological forecast data for the wind farm, and then inputting the mesoscale meteorological forecast data and the corresponding inflow boundary conditions into the trained wind farm time series prediction PatchTST-CNN neural network to obtain the wind power prediction result.

2. The wind power prediction method based on mesoscale-microscale dynamics according to claim 1 is characterized in that: Using large eddy simulation method, a micro-scale aerodynamic model of wind farm is constructed: For the scale size Filter the cube space and filter the scale smaller than The turbulence elimination is expressed as: ； In the formula, represents the filtered turbulence, i.e., large-scale turbulence; Represents spatial coordinates; Indicates time; represents the turbulence variable; represents the integral variable, including the flow direction component , spanwise component and the vertical component ; represents the filter kernel function; represents a differential operator; The governing equations describing the turbulence in the wind farm are established, including the continuity equation and the momentum equation, which are expressed as: ； ； In the formula, , Respectively indicate the direction and direction The large-scale velocity component on , represent the streamwise direction, spanwise direction and vertical direction respectively; , Respectively indicate the direction and direction The spatial coordinates on ; Indicates the air density; represents the large-scale pressure field; Indicates the degree of motion; represents body force; represents the sub-grid stress; The actuating line model is used to model the fan wake. Based on the blade element theory, the lift and drag of a single blade element along the blade span direction are expressed as: ； ； In the formula, , represent the lift differential and drag differential along the span direction respectively; , represent lift and drag coefficient calculation functions respectively; represents the angle of attack; Indicates the length of the leaf in the span direction; Indicates the length of the chord; Indicates relative speed; Relative speed It is expressed as: ； In the formula, , denote radial and tangential velocities, respectively; represents the blade rotation angular velocity; represents the blade element radius; Sample the 3D Gaussian distribution to map lift and drag to nearby grids: ； In the formula, Represents the value of the Gaussian distribution function, describing the The distribution intensity at represents the Gaussian adaptation coefficient; Represents the distance between the center grid point of the flow field and other grid points.

3. The method for wind power prediction based on mesoscale-microscale dynamics according to claim 2 is characterized in that: Subgrid stress It is expressed as: ； In the formula, Indicates direction and direction The product of the large-scale velocity components on ; The subgrid stress is calculated as: ； ； In the formula, represents the tensor deformation rate of large-scale turbulence; represents a symbolic function; represents the isotropic component; represents the subgrid eddy viscosity coefficient; Subgrid eddy viscosity coefficient It is expressed as: ； In the formula, Represents the Smagorinsky coefficient.

4. The method for wind power prediction based on mesoscale-microscale dynamics according to claim 3 is characterized in that: The relevant features required for wind farm microscale aerodynamic model input include: Thermal stability of inflow boundary conditions, wind speed at different heights, wind direction, atmospheric pressure, and ground roughness; Simulated wind farm topography and wind turbine locations; Wind turbine parameters of the actuation line model, including blade shape parameters, hub height, cut-in wind speed, and cut-out wind speed; The output of the micro-scale aerodynamic model of a wind farm is the turbulence, wind acceleration factor, wind direction, and power at the location of each wind turbine.

5. The wind power prediction method based on mesoscale-microscale dynamics according to claim 4 is characterized in that: The Monin-Obukhov similarity theory is used to set the inflow boundary conditions; The inflow boundary condition under constant shear stress is: ； In the formula, represents the turbulent shear stress; Indicates the air density; Indicates height; Indicates height The corresponding wind speed; represents turbulent kinetic energy; represents the dissipation rate; represents a constant; The inflow boundary condition with the gradual change of shear stress and the correction of turbulent kinetic energy along the height is: ； In the formula, represents the thickness of the local atmospheric boundary layer; represents the friction speed; The boundary condition for wall inflow is: ； In the formula, It represents the wind speed at the center of the cell next to the wall; Indicates the height of the center position of the cell next to the wall; represents the von Karman constant; represents a dimensionless constant; represents the roughness constant; represents the roughness length; represents the shear stress at the wall; The boundary condition for inflow on the upper surface is: ； In the formula, It represents the equivalent volume force applied in a layer of cells close to the top surface; represents the additional shear stress at the upper top surface; Indicates the height of the cell next to the top base.

6. The method for wind power prediction based on mesoscale-microscale dynamics according to claim 5 is characterized in that: The mesoscale meteorological prediction model includes a data preprocessing module, a prediction module, a prediction result post-processing module and a visualization module. The input of the mesoscale meteorological prediction model includes the meteorological prediction area, terrain data, grid node data and interpolated meteorological data; The governing equation of the mesoscale meteorological prediction model is: ； In the formula, Indicates time The partial derivative of express Components of the directional velocity field; represents the acceleration due to gravity; express Components of the directional velocity field; Indicates along Directional wind speed component; Express The partial derivative of Indicates the gas pressure state; It represents the change of potential height in normal direction; Indicates the potential height at Change of direction; represents the forcing term caused by the physical process; Indicates along Directional wind speed component; Express The partial derivative of Indicates the potential height at Change of direction; represents the forcing term caused by the perturbation mixing; express Components of the directional velocity field; Indicates along Directional wind speed component; Express The partial derivative of It represents the mass of air column per unit area at the model grid point; represents the forcing term caused by the spherical projection; represents the momentum of the potential temperature field; Indicates potential temperature; represents the forcing term caused by potential temperature; Indicates potential.

7. The method for wind power prediction based on mesoscale-microscale dynamics according to claim 6, characterized in that: The processing flow of the PatchTST-CNN neural network for wind farm time series prediction is as follows: Divide the input of the wind farm time series prediction PatchTST-CNN neural network into multiple small blocks; Perform feature mapping on the small block to obtain the latent space feature vector, which is expressed as: ； In the formula, represents the feature mapping matrix; represents the position encoding matrix; Indicates The latent space feature vector obtained after feature mapping of each small block; Indicates a small piece; The latent space feature vector is processed by a multi-head attention mechanism to obtain the attention feature; The attention features are then input into the convolutional neural network for processing to obtain the wind power prediction result, which is the output of the wind farm time series prediction PatchTST-CNN neural network.

8. A wind power prediction system based on mesoscale-microscale dynamics, characterized in that: include: Wind farm micro-scale aerodynamic model construction module, used to construct a wind farm micro-scale aerodynamic model and extract relevant features required for model input; The inflow boundary condition setting and database construction module is used to set the inflow boundary conditions, simulate the inflow boundary conditions and the micro-scale aerodynamic model of the wind farm, generate the turbulence, wind acceleration factor, wind direction, and power of each wind turbine, and construct a database based on the generated turbulence, wind acceleration factor, wind direction, and power of each wind turbine; The coupled model building module is used to build a mesoscale meteorological prediction model, which is used as the initial condition of the microscale aerodynamic model of the wind farm to build a mesoscale-microscale aerodynamic coupling model of the wind farm; The prediction network building module is used to combine block processing technology, time series prediction and convolutional neural network to build the wind farm time series prediction PatchTST-CNN neural network; The pre-training and prediction module is used to use the mesoscale-microscale aerodynamic coupling model of the wind farm as the system environment, use the database to train the wind farm time series prediction PatchTST-CNN neural network, and obtain the trained wind farm time series prediction PatchTST-CNN neural network; obtain the mesoscale meteorological forecast data, that is, the output of the mesoscale meteorological forecast model, and calculate the inflow boundary conditions of the mesoscale meteorological forecast data for the wind farm, and then input the mesoscale meteorological forecast data and the corresponding inflow boundary conditions into the trained wind farm time series prediction PatchTST-CNN neural network for prediction.

9. An electronic device, characterized in that: It includes a processor, a memory and a bus, wherein the processor and the memory are connected via the bus, wherein the memory is used to store a set of program codes, and the processor is used to call the program codes stored in the memory to execute the wind power prediction method based on mesoscale-microscale dynamics as described in any one of claims 1 to 7.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions execute the wind power prediction method based on mesoscale-microscale dynamics as described in any one of claims 1-7.

Images