A method for predicting extreme weather power based on typhoon path deduction

Abstract

This invention relates to the field of wind power generation technology, specifically to an extreme weather power prediction method based on typhoon path extrapolation, comprising the following steps: S100: acquiring initial typhoon path and sea level pressure data, inputting the data into a pre-trained spatiotemporal graph convolutional neural network coupled with a parameterized typhoon wind field model, and extrapolating to generate a macroscopic three-dimensional typhoon wind field spatiotemporal sequence covering the target wind farm area. This invention overturns the fixed wind speed shutdown judgment logic, introduces a single-unit dynamic cut-out probability model driven by multi-dimensional physical characteristics such as wind speed, turbulence, and vibration, and innovatively integrates it as a penalty factor into the site-level wake topology correction, perfectly restoring the real physical process of "gradual shutdown" and "probabilistic disappearance of wake" of wind turbine units during typhoon passage, and completely eliminating the "cliff-like" error in power prediction under extreme weather conditions.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and more specifically to an extreme weather power prediction method based on typhoon path extrapolation. Background Technology

[0002] Wind power, as an important new clean energy source, exhibits highly random and fluctuating power generation. In wind-rich areas such as the southeastern coast and complex mountainous regions of my country, extreme weather events like typhoons are frequent. During typhoons, wind farms are highly susceptible to large-scale grid disconnection or sudden power fluctuations, posing significant challenges to the safe and stable operation of the power grid and real-time power dispatch. Therefore, establishing a high-precision wind power forecasting system specifically designed for typhoon-related extreme weather events is crucial.

[0003] However, existing wind power forecasting methods suffer from the following three insurmountable technical shortcomings when dealing with extreme weather conditions such as typhoons: There is a scale disconnect between macro-level weather forecasts and micro-level topographic wind fields: most existing forecasting systems rely directly on macro-level numerical weather prediction (NWP) data released by meteorological bureaus. Due to the low spatial resolution of NWP, it cannot characterize the wind field distortions (such as acceleration effects and wind shear) that occur when the large-scale circulation of a typhoon encounters complex terrain and vegetation within a wind farm. While traditional computational fluid dynamics (CFD) can perform downscaling calculations, it is extremely time-consuming and cannot meet the timeliness requirements for advanced real-time power grid dispatching during typhoons.

[0004] Fixed wind speed cutoff logic leads to precipitous prediction errors: Existing prediction models generally employ a "one-size-fits-all" fixed wind speed threshold logic when dealing with extreme weather shutdowns (e.g., setting the predicted power to instantly drop to zero when the wind speed exceeds 25 m / s). However, in the real physical world, wind turbine anti-typhoon protection strategies are extremely complex, often constrained by multiple factors such as sudden changes in turbulence intensity, abnormal nacelle vibration, and yaw errors. This simplistic, rigid cutoff decision results in a severe precipitous error between the predicted and actual power curves.

[0005] Wake models under extreme weather conditions are severely distorted: Existing wake attenuation models (such as the classic Jensen model) are all based on the static assumption that "upstream units are always operating and absorbing wind energy." During typhoons, many upstream units may have already triggered their protection mechanisms to shut down and feather, and their wake-blocking effect on downstream units will disappear, causing the actual available wind speed and power of downstream units to recover significantly. Current technology ignores this "probabilistic wake weakening due to shutdown," leading to a serious underestimation of the total predicted power for the entire field. Summary of the Invention

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention provides an extreme meteorological power prediction method based on typhoon path extrapolation, comprising the following steps: S100: acquiring initial typhoon path and sea level pressure data, inputting the data into a pre-trained spatiotemporal graph convolutional neural network coupled with a parameterized typhoon wind field model, and extrapolating to generate a macroscopic three-dimensional typhoon wind field spatiotemporal sequence covering the target wind farm area; S200: Construct a three-dimensional terrain and surface roughness model of the target wind farm, and use the macroscopic three-dimensional typhoon wind field spatiotemporal sequence as boundary conditions. Input the model into a computational fluid dynamics proxy model inspired by physical mechanisms, and perform downscaling calculations to obtain the time series data of microscopic wind speed, wind direction and turbulence intensity at the hub height of each wind turbine in the target wind farm. S300: Extract the real-time operating characteristic data of each wind turbine, combine the corresponding micro wind speed and turbulence intensity, and calculate the dynamic cut-out probability of each wind turbine triggering shutdown protection at a future predicted time through a pre-built dynamic cut-out probability model. The real-time operating characteristic data includes at least the nacelle vibration acceleration that reflects the load state of the turbine. S400: Based on the theoretical wind speed-power curve of a single unit and the corresponding micro wind speed, calculate the ideal predicted power of a single unit; use the dynamic cut-out probability to mathematically correct the ideal predicted power of a single unit to obtain the expected power of a single unit; and combine the wake topology correction matrix under the current micro wind direction to aggregate the expected power of a single unit of all the wind turbines in the field, and output the final predicted total power of the target wind farm during the typhoon.

[0007] Further, in step S100, a macroscopic three-dimensional typhoon wind field spatiotemporal sequence covering the target wind farm area is generated, which specifically includes the following sub-steps: S101: Obtain historical typhoon observation data, which includes the latitude and longitude of the typhoon center, the lowest central pressure, and the radius of the maximum wind speed at multiple historical moments; S102: Construct the spatiotemporal graph convolutional neural network, taking the historical observation data and meteorological grid data of the target sea area as input; extract the spatial topological correlation features between the typhoon center and its surrounding meteorological grids through the spatial graph convolutional layer in the network, and extract the temporal evolution features of the typhoon's movement path and intensity through the temporal convolutional layer. S103: Output macroscopic typhoon state parameters for each moment within the future forecast period. The macroscopic typhoon state parameters include: the latitude and longitude of the predicted typhoon center, the predicted minimum air pressure at the center, and the radius of the predicted maximum wind speed.

[0008] Further, in step S100, after obtaining the macroscopic typhoon state parameters, they are input into the parameterized typhoon wind field model to generate a macroscopic three-dimensional typhoon wind field spatiotemporal sequence. The specific calculation logic is as follows: Based on the Holland parametric wind pressure model formula, the radius of the predicted typhoon center is calculated. Free atmospheric gradient wind speed : in, This refers to the typhoon shape coefficient. air density, The radius of the predicted maximum wind speed. The ambient air pressure, The lowest air pressure at the predicted center, For Coriolis parameters.

[0009] Further, in step S200, the process of constructing and training the computational fluid dynamics surrogate model inspired by physical mechanisms includes: A deep neural network is established, comprising an input layer, hidden layers, and an output layer, wherein the input layer receives arbitrary spatial coordinates within the target wind farm. and time parameters The output layer outputs the three-dimensional microscopic wind speed vector corresponding to that point. and micro pressure ; Construct a joint loss function that includes data fitting loss and physical constraint loss; The partial derivatives of the output layer results with respect to the input spatial coordinates and time are calculated using automatic differentiation techniques, and the weights and biases of the deep neural network are iteratively optimized by minimizing the joint loss function.

[0010] Furthermore, the physical constraint loss in the joint loss function It is constructed based on the residuals of the mass conservation equation and the incompressible Navier-Stokes equations in fluid mechanics, and its mathematical expression is: in, The first term represents the divergence of the wind speed vector, and the second term represents the residual of mass conservation. air density, The kinematic viscosity coefficient, and An adaptive penalty weighting coefficient to balance the strength of physical constraints.

[0011] Furthermore, in step S200, the specific steps for performing the downscaling calculation are as follows: The elevation data and surface roughness data of the three-dimensional terrain model are mapped to the bottom no-slip boundary condition, and the spatiotemporal sequence of the macroscopic three-dimensional typhoon wind field is spatially interpolated and used as the lateral boundary condition and top initial boundary condition of the computational domain. The longitude, latitude, and hub height coordinates of each wind turbine are input into the trained computational fluid dynamics surrogate model, which directly outputs the corresponding three-dimensional microscopic wind speed vector through forward inference. ; For the three-dimensional microscopic wind speed vector The system is decomposed to obtain the microscopic wind speed and direction on the horizontal plane; and the ratio of the fluctuation standard deviation of the microscopic wind speed to the average wind speed within a time step is extracted to calculate the time series data of the turbulence intensity of the corresponding unit.

[0012] Furthermore, in step S3, specifically in step S300, the specific process of constructing and calculating the dynamic cut-out probability includes: Obtain the wind turbine at the predicted future time. micro wind speed With turbulence intensity ; Extract cabin vibration acceleration from the real-time operational feature data. and the yaw error rate of the crew ; The above features are input into a pre-built dynamic cut-out probability model based on multidimensional logistic regression to calculate the first... Typhoon turbine units at all times Dynamic handover probability of triggering shutdown protection The calculation formula is as follows: in, Basic bias term, These are the weight coefficients of the corresponding feature variables.

[0013] Furthermore, in step S4, the weight coefficients of each feature variable in the dynamic cut-out probability model are obtained through the following training steps: Collect SCADA system operation records of the target wind farm during historical typhoons, extract historical shutdown events and their corresponding micro wind speed, turbulence intensity, nacelle vibration acceleration and yaw error rate, and construct a historical sample set; The historical sample set is processed by binary labeling: the status label of the wind turbine being normally connected to the grid and generating electricity is set to 0, and the status label of the wind turbine being shut down and disconnected from the grid due to the triggering of the anti-typhoon self-protection mechanism is set to 1. The maximum likelihood estimation method is used to iteratively optimize the labeled historical sample set, aiming to minimize the cross-entropy loss function, and to fit and solve for the weight coefficients that maximize the probability of historical shutdown events. .

[0014] Furthermore, in step S400, the specific process for calculating the expected power of a single unit is as follows: Based on the theoretical wind speed-power curve of the wind turbine generator set from the manufacturer, calculate the first... Typhoon turbine units at micro wind speed Ideal predicted power ; Introducing the dynamic cut-out probability calculated in step S300 The ideal predicted power of a single unit is converted into the expected power of a single unit, which reflects the actual power generation capacity of the unit during the typhoon. Its mathematical expression is: in, This represents the basic standby power loss value of the wind turbine unit in the feathering state when it is stopped.

[0015] Furthermore, combining the wake topology correction matrix under the current micro-wind direction, the expected power of each individual wind turbine in the entire field is aggregated, specifically including the following sub-steps: S401: Microscopic wind direction output by the computational fluid dynamics surrogate model. Determine the upstream and downstream spatial topological relationships between wind turbines within the target wind farm, and select the appropriate turbines for the first... The set of all upstream turbine units affected by the wake of the typhoon generator. ; S402: Calculate the dynamic probability wake attenuation coefficient; when the upstream unit When the unit is shut down, its wake effect is reduced, therefore the upstream unit... Operating probability As a penalty factor, its impact on downstream units is calculated. Expected speed loss : in, For thrust coefficient, The diameter of the wind turbine, The wake expansion coefficient is... For the unit to the unit Effective distance along the wind direction; S403: Calculate the first... using the square sum root method The combined wake velocity deficit rate of the typhoon turbine ; S404: The expected power of all units in the wind farm is aggregated and corrected using the comprehensive wake velocity deficit rate, and the result is output as the target wind farm at the time of typhoon passage. Final predicted total power : in, This represents the total number of wind turbines within the wind farm.

[0016] Beneficial effects This invention, by deeply integrating physically constrained neural networks and meteorological models, has for the first time established a second-level computational link from macroscopic typhoon spatiotemporal extrapolation to microscopic complex terrain flow field downscaling. Simultaneously, it overturns the traditional "one-size-fits-all" fixed wind speed shutdown decision logic, introducing a single-unit dynamic cut-out probability model driven by multi-dimensional physical characteristics such as wind speed, turbulence, and vibration. This model is innovatively integrated as a penalty factor into the station-level wake topology correction, perfectly reproducing the real physical process of "gradual shutdown" and "probabilistic disappearance of wake" of wind turbines during typhoon passage. This completely eliminates the "cliff-like" error in power prediction under extreme weather conditions, providing extremely reliable technical support for the safe defense and high-precision real-time dispatch of new power systems during typhoons and other severe weather events. Attached Figure Description

[0017] Figure 1 This is a flowchart of an extreme meteorological power prediction method based on typhoon path extrapolation according to the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but includes other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] The present invention will now be described in further detail with reference to the accompanying drawings: Example: As shown in the figure, an extreme meteorological power prediction method based on typhoon path extrapolation includes the following steps: S100: Obtain initial typhoon path and sea level pressure data, input the data into a pre-trained spatiotemporal graph convolutional neural network and parameterized typhoon wind field coupling model, and deduce and generate a macroscopic three-dimensional typhoon wind field spatiotemporal sequence covering the target wind farm area; S200: Construct a three-dimensional terrain and surface roughness model of the target wind farm, and use the macroscopic three-dimensional typhoon wind field spatiotemporal sequence as boundary conditions. Input the model into a computational fluid dynamics proxy model inspired by physical mechanisms, and perform downscaling calculations to obtain the time series data of microscopic wind speed, wind direction and turbulence intensity at the hub height of each wind turbine in the target wind farm. S300: Extract the real-time operating characteristic data of each wind turbine, combine the corresponding micro wind speed and turbulence intensity, and calculate the dynamic cut-out probability of each wind turbine triggering shutdown protection at a future predicted time through a pre-built dynamic cut-out probability model. The real-time operating characteristic data includes at least the nacelle vibration acceleration that reflects the load state of the turbine. S400: Based on the theoretical wind speed-power curve of a single unit and the corresponding micro wind speed, calculate the ideal predicted power of a single unit; use the dynamic cut-out probability to mathematically correct the ideal predicted power of a single unit to obtain the expected power of a single unit; and combine the wake topology correction matrix under the current micro wind direction to aggregate the expected power of a single unit of all the wind turbines in the field, and output the final predicted total power of the target wind farm during the typhoon.

[0021] Furthermore, the specific implementation process of step S100 is as follows: In the specific implementation process, addressing the primary challenge of wind power prediction during typhoon passage—the drastic variability of macroscopic meteorological systems—this invention first executes step S100, which aims to deduce and generate a macroscopic three-dimensional spatiotemporal sequence of typhoon wind fields covering the target wind farm area. Since a typhoon is a massive weather system with a scale reaching hundreds of kilometers, its movement path and intensity directly determine the initial meteorological boundary conditions surrounding the wind farm. Therefore, this step is the physical foundation for all subsequent microscopic downscaling calculations.

[0022] First, the system acquires initial typhoon data in real time through a meteorological interface, including historical typhoon observation data and real-time sea level pressure field data of the target sea area. During the data fusion processing stage, the system aligns and correlates the typhoon center latitude and longitude, minimum central pressure, and maximum wind speed radius at multiple historical moments with meteorological grid data distributed across the target sea area, constructing a graph network input matrix with a spatiotemporal topology.

[0023] After acquiring and processing the preceding data, the system feeds the graph network input matrix into a pre-trained spatiotemporal graph convolutional neural network (STGCN) for forward inference. During the network's computation, the spatial graph convolutional layer first captures the spatial topological relationships between the typhoon's center and its surrounding meteorological grid points, accurately quantifying the spatial distribution of pressure gradient forces. Next, the extracted spatial features are passed to the temporal convolutional layer, which uses gating mechanisms (such as GRU or LSTM structures) to deeply mine the temporal evolution of the typhoon's path and intensity. Through the alternating extraction and fusion of spatiotemporal features, the neural network ultimately outputs macroscopic typhoon state parameters for each moment within the future prediction period, specifically including the predicted typhoon center latitude and longitude, the predicted minimum central pressure, and the predicted maximum wind speed radius.

[0024] However, the network output alone only provides discrete state parameters, which cannot be directly used for wind farm power prediction. Therefore, after obtaining the aforementioned macroscopic typhoon state parameters, this embodiment further inputs them into a parameterized typhoon wind field model to reconstruct the macroscopic three-dimensional flow field in a physically driven manner. Specifically, the system, based on the classic Holland parameterized wind pressure model, calculates the wind pressure at a distance of [missing information - likely a radius] from the predicted typhoon center. Free atmospheric gradient wind speed Its core operational logic follows the physical equations below: During the above calculation process, the system will output the predicted maximum wind speed radius from the neural network. With the lowest air pressure at the forecast center Substituting into the equation, and combining it with the real-time acquired ambient air pressure... air density Typhoon shape coefficient and Coriolis parameters The wind speed in the free flow field at high altitudes, unaffected by surface friction, was calculated.

[0025] Finally, to transition the upper-level free-flow field to the near-surface region where the wind turbines are located, the system incorporates surface roughness parameters from the sea or land where the target wind farm is situated, and uses the logarithmic wind profile equation to correct the calculated gradient wind speed for boundary layer friction attenuation. Through this correction process, the system can smoothly calculate the macroscopic wind speed distribution patterns at different vertical height levels from sea level up to the hub height of the wind turbines. At this point, a series of three-dimensional mesh data with timestamps and vertical distribution information are combined and stitched together, successfully generating a macroscopic three-dimensional typhoon wind field spatiotemporal sequence covering the target wind farm area, providing solid boundary conditions for subsequent microscopic downscaling calculations.

[0026] Furthermore, the specific implementation process of S200 is as follows: After obtaining the macroscopic three-dimensional typhoon wind field spatiotemporal sequence output in step S100, this embodiment aims to bridge the scale gap between the macroscopic meteorological background field and the complex microscopic topography of the wind farm, achieving high-precision downscaling calculation of the microscopic wind field. Because the underlying wind field is strongly disturbed by topographic undulations and vegetation cover when a typhoon makes landfall or approaches coastal or complex mountain wind farms, resulting in severe wind shear and local acceleration effects, a high-resolution computational fluid dynamics surrogate model must be introduced for flow field reconstruction.

[0027] First, the system constructs a high-precision 3D terrain and surface roughness model for the target wind farm. In practice, the system analyzes elevation data (DEM) and land cover data from a Geographic Information System (GIS) to transform the ridges, valleys, vegetation, and sea surface conditions within the wind farm into no-slip physical boundary conditions at the bottom of the computational domain. Subsequently, the system extracts the aforementioned macroscopic 3D typhoon wind field spatiotemporal sequence and maps it to the lateral inflow boundary conditions and initial top boundary conditions of the computational domain using a spatial interpolation algorithm. This precise mapping of boundary conditions allows subsequent fluid calculations to be constrained within the realistic large-scale typhoon circulation context.

[0028] After completing the physical space modeling and boundary setting, the system constructs a computational fluid dynamics surrogate model (PINN) inspired by physical mechanisms to replace the traditional, extremely time-consuming CFD numerical iterative solver. This surrogate model is essentially a deep feedforward neural network, whose input layer receives arbitrary spatial coordinates within the target wind farm. and discrete time parameters The output layer directly outputs the three-dimensional microscopic wind speed vector corresponding to that spatiotemporal node. and micro pressure To ensure that the flow field output by the neural network conforms to the real physical laws, this embodiment abandons the purely data-driven mean square error loss when training the surrogate model, and instead innovatively constructs a joint loss function that includes the data fitting residual and the fluid dynamics physical constraint residual.

[0029] Specifically, the system utilizes automatic differentiation techniques to accurately calculate the partial derivatives of the neural network output layer results with respect to the input spatial coordinates and time, and directly substitutes them into the core mass conservation equations and incompressible Navier-Stokes equations in fluid mechanics. This results in a physical constraint loss. The mathematical expression is defined as: in, The first term represents the divergence of the wind speed vector, and the second term represents the residual of mass conservation. air density, The kinematic viscosity coefficient, and An adaptive penalty weighting coefficient to balance the strength of physical constraints.

[0030] In the aforementioned forward constraint and backpropagation optimization process, the first term penalizes the wind speed vector divergence. The non-zero region ensures mass conservation, while the second term constrains the momentum conservation of fluid particles under typhoon forcing. air density, The kinematic viscosity coefficient, and This refers to the adaptive penalty weights used to balance the strength of physical constraints. By directly embedding partial differential equations into the neural network computation graph, the surrogate model can learn the three-dimensional wind field mapping relationship that conforms to strict fluid dynamics laws in a very short inference time.

[0031] Once the surrogate model is trained and converged, the system can enter the efficient downscaling inference stage. The system extracts the precise latitude and longitude coordinates and hub center height coordinates of each wind turbine in the target wind farm, and feeds them directly into the model as input parameters. The surrogate model instantly outputs the real-time three-dimensional microscopic wind speed vector at the hub height of each turbine. .

[0032] Furthermore, the specific implementation process of S300 is as follows: After obtaining the high-precision output of the microscopic wind speed and turbulence intensity time-series data at the hub height of each wind turbine in step S200, step S300 aims to accurately quantify the dynamic probability of the turbine triggering its self-protection mechanism under extreme weather conditions. Traditional wind power prediction systems typically rely solely on a rigid wind speed threshold (e.g., setting a microscopic wind speed of 25 m / s as the threshold for shutdown and grid disconnection). This idealized "one-size-fits-all" logic completely ignores the fact that in the real physical world, when wind turbines face extreme weather such as typhoons, their control systems implement progressive, nonlinear cut-off protection based on multiple factors such as turbine vibration, yaw hysteresis, and sudden turbulence changes. It is precisely this traditional rigid judgment that leads to frequent, precipitous errors in the power prediction curve during typhoons.

[0033] The system first establishes high-frequency real-time communication with the target wind farm's Supervisory Control and Data Acquisition (SCADA) system, and then extracts key feature data that accurately reflects the current load and attitude status of the turbines from massive amounts of turbine operation records. Specifically, for the first... Typhoon turbine units in the future forecast time The system not only simultaneously acquires the micro wind speed derived from the forward deduction in step S200, but also... With turbulence intensity It also extracts cabin vibration acceleration in real time from the SCADA data stream. and the yaw error rate of the crew .

[0034] Subsequently, the system inputs the fused multidimensional feature parameters into a pre-built dynamic cut-out probability model based on multidimensional logistic regression to calculate the cut-out probability of the specific wind turbine at time [time value missing]. Dynamic handover probability of triggering shutdown protection mechanism This calculation process is not a simple linear superposition, but rather a continuous probability mapping through a highly nonlinear Sigmoid activation function. Its core mathematical expression is rigorously defined as follows: in, Basic bias term, These are the weight coefficients of the corresponding feature variables.

[0035] In this probability calculation equation, the left side outputs a continuous probability value between 0 and 1, profoundly reflecting the fuzzy transitional state of the unit between "normal grid connection" and "feather shutdown". The right side of the equation... The basic bias term, and These represent the contribution of four characteristic variables—microscopic wind speed, turbulence intensity, nacelle vibration acceleration, and yaw error rate—to the shutdown decision weight, respectively.

[0036] To ensure that the dynamic cut-off probability model possesses strong adaptability and generalization capabilities for specific turbine models in the target wind farm, this embodiment further details the offline iterative training process of the aforementioned core weight coefficients. Before model deployment, the system first retrieves long-sequence SCADA historical operation records of the target wind farm during multiple typhoons in its history. Events that actually triggered the typhoon self-protection mechanism due to wind shear or excessive vibration, resulting in turbine shutdown and grid disconnection, are extracted and their status labels are strictly set to 1. Conversely, the status labels for periods when the turbines are normally connected to the grid are set to 0. Based on this binary-labeled historical sample set, the system uses maximum likelihood estimation for backward iterative optimization. In the optimization calculation, the system uses minimizing the cross-entropy loss function as the global optimization objective, continuously adjusting the parameters of each dimension through the gradient descent algorithm until the optimal set of weight coefficients that maximizes the probability of historical shutdown events is fitted and solved. It is through this rigorous closed-loop training mechanism that the model possesses the "intelligence" to predict complex shutdown behaviors of real turbines in advance, thereby completely eliminating the cliff-like prediction defects of traditional methods.

[0037] Furthermore, the specific implementation process of S400 is as follows: After accurately quantifying the dynamic cut-out probability of each wind turbine during a typhoon in step S300, this embodiment of the invention proceeds to the final site-level aggregation and correction step S400. This step aims to map the probabilistic state of a single turbine to the predicted total power available for grid connection across the entire site. First, the system needs to assess the baseline power generation capacity of a single turbine.

[0038] The system uses the factory-provided theoretical wind speed-power curve from the wind turbine manufacturer, combined with the micro wind speed output in step S200. Calculate the first value by looking up a table or interpolation. Predicted power output of typhoon generators under ideal conditions (i.e., assuming no faults or protective shutdowns). However, due to the extreme weather conditions caused by typhoons, the units are actually in a superposition of "operation" and "shutdown" states. Therefore, the system incorporates the dynamic cut-out probability calculated above. This ideal power is then converted into a statistically significant single-unit expected power. To closely approximate actual engineering conditions, the calculation of this expected power not only reduced the generated electricity but also specifically considered the basic standby power loss value of the unit when it is shut down and disconnected from the grid. For example, the minimal power consumption required to maintain the yaw system and hydraulic control system. Its mathematical expression is rigorously constructed as follows: in, This represents the basic standby power loss value of the wind turbine unit in the feathering state when it is stopped.

[0039] After calculating the independent expected power of each turbine, the system must further address the complex aerodynamic coupling problem within the wind farm, namely the wake effect. In traditional wind power forecasting, wake attenuation models (such as the Jensen model) typically assume that all upstream turbines are continuously operating and absorbing wind energy, thus causing a fixed wind speed loss for downstream turbines. However, under the extreme condition of a typhoon landfall, this assumption is extremely flawed: once the upstream turbines trigger their self-protection mechanism and shut down, their blades will perform feathering, and their thrust and blocking effect on the wind will decrease sharply. The downstream turbines that were originally blocked will then experience a "wind speed recovery" and a "power surge".

[0040] To accurately capture this dynamic physical process, this embodiment innovatively constructs a dynamic probabilistic wake attenuation mechanism. The system first uses the real-time micro-wind direction output in step S200... Dynamically determine the upstream and downstream spatial topological relationships within the wind farm, and filter out the current time-dependent relationships. The combined effect of wake shielding from typhoon turbines on all upstream turbine units. Subsequently, the system broke with convention and connected the upstream units The actual operating probability, As a dynamic penalty factor, it is deeply integrated into the derivation equations of the classic wake model. Therefore, the upstream units... For downstream units Expected speed loss Redefined as: in, The thrust coefficient at the current wind speed. The diameter of the wind turbine, The expansion coefficient, which reflects the wake diffusion rate, This represents the effective projected distance between the two units along the current wind direction. Through this interdisciplinary formula reconstruction, the downtime probability of the upstream unit is perfectly mapped to the attenuation coefficient of the downstream wake intensity.

[0041] Finally, the system uses the sum of squares method to superimpose the multiple wakes and calculate the first... The combined wake velocity deficit rate of the typhoon turbine The system utilizes this comprehensive loss rate to perform a cubic-level aggregation correction on the expected power of individual units across all wind farms (since power is proportional to the cube of wind speed). Finally, the system outputs the target wind farm's performance at the predicted typhoon passage time. Final predicted total power Its aggregation formula is ,in This represents the total number of generating units in the entire facility. Thus, this invention has completely established a closed-loop prediction system encompassing macroscopic meteorological typhoon projection, microscopic three-dimensional flow field downscaling, dynamic intelligent switching of individual units, and station-level probabilistic wake correction, providing high-precision decision-making support for new power system dispatching under extreme weather conditions.

[0042] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting extreme meteorological power based on typhoon path extrapolation, characterized in that, Includes the following steps: S100: Obtain initial typhoon path and sea level pressure data, input the data into a pre-trained spatiotemporal graph convolutional neural network and parameterized typhoon wind field coupling model, and deduce and generate a macroscopic three-dimensional typhoon wind field spatiotemporal sequence covering the target wind farm area; S200: Construct a three-dimensional terrain and surface roughness model of the target wind farm, and use the macroscopic three-dimensional typhoon wind field spatiotemporal sequence as boundary conditions. Input the model into a computational fluid dynamics proxy model inspired by physical mechanisms, and perform downscaling calculations to obtain the time series data of microscopic wind speed, wind direction and turbulence intensity at the hub height of each wind turbine in the target wind farm. S300: Extract the real-time operating characteristic data of each wind turbine, combine the corresponding micro wind speed and turbulence intensity, and calculate the dynamic cut-out probability of each wind turbine triggering shutdown protection at a future predicted time through a pre-built dynamic cut-out probability model. The real-time operating characteristic data includes at least the nacelle vibration acceleration that reflects the load state of the turbine. S400: Based on the theoretical wind speed-power curve of a single unit and the corresponding micro wind speed, calculate the ideal predicted power of a single unit; use the dynamic cut-out probability to mathematically correct the ideal predicted power of a single unit to obtain the expected power of a single unit; and combine the wake topology correction matrix under the current micro wind direction to aggregate the expected power of a single unit of all the wind turbines in the field, and output the final predicted total power of the target wind farm during the typhoon.

2. The extreme meteorological power prediction method based on typhoon path extrapolation according to claim 1, characterized in that, In step S100, a macroscopic three-dimensional typhoon wind field spatiotemporal sequence covering the target wind farm area is generated, which specifically includes the following sub-steps: S101: Obtain historical typhoon observation data, which includes the latitude and longitude of the typhoon center, the lowest central pressure, and the radius of the maximum wind speed at multiple historical moments; S102: Construct the spatiotemporal graph convolutional neural network, taking the historical observation data and meteorological grid data of the target sea area as input; extract the spatial topological correlation features between the typhoon center and its surrounding meteorological grids through the spatial graph convolutional layer in the network, and extract the temporal evolution features of the typhoon's movement path and intensity through the temporal convolutional layer. S103: Output macroscopic typhoon state parameters for each moment within the future forecast period. The macroscopic typhoon state parameters include: the latitude and longitude of the predicted typhoon center, the predicted minimum air pressure at the center, and the radius of the predicted maximum wind speed.

3. The extreme meteorological power prediction method based on typhoon path extrapolation according to claim 2, characterized in that, In step S100, after obtaining the macroscopic typhoon state parameters, they are input into the parameterized typhoon wind field model to generate a macroscopic three-dimensional typhoon wind field spatiotemporal sequence. The specific calculation logic is as follows: Based on the Holland parametric wind pressure model formula, the radius of the predicted typhoon center is calculated. Free atmospheric gradient wind speed : in, This refers to the typhoon shape coefficient. air density, The radius of the predicted maximum wind speed. The ambient air pressure, The lowest air pressure at the prediction center, For Coriolis parameters.

4. The extreme meteorological power prediction method based on typhoon path extrapolation according to claim 3, characterized in that, In step S200, the process of constructing and training the physics-inspired computational fluid dynamics surrogate model includes: A deep neural network is established, comprising an input layer, hidden layers, and an output layer, wherein the input layer receives arbitrary spatial coordinates within the target wind farm. and time parameters The output layer outputs the three-dimensional microscopic wind speed vector corresponding to that point. and micro pressure ; Construct a joint loss function that includes data fitting loss and physical constraint loss; The partial derivatives of the output layer results with respect to the input spatial coordinates and time are calculated using automatic differentiation techniques, and the weights and biases of the deep neural network are iteratively optimized by minimizing the joint loss function.

5. The extreme meteorological power prediction method based on typhoon path extrapolation according to claim 4, characterized in that, Physical constraint loss in the joint loss function It is constructed based on the residuals of the mass conservation equation and the incompressible Navier-Stokes equations in fluid mechanics, and its mathematical expression is: in, The first term represents the divergence of the wind speed vector, and the second term represents the residual of mass conservation. air density, The kinematic viscosity coefficient, and An adaptive penalty weighting coefficient to balance the strength of physical constraints.

6. The extreme meteorological power prediction method based on typhoon path extrapolation according to claim 5, characterized in that, In step S200, the specific steps for performing downscaling calculations are as follows: The elevation data and surface roughness data of the three-dimensional terrain model are mapped to the bottom no-slip boundary condition, and the spatiotemporal sequence of the macroscopic three-dimensional typhoon wind field is spatially interpolated and used as the lateral boundary condition and top initial boundary condition of the computational domain. The longitude, latitude, and hub height coordinates of each wind turbine are input into the trained computational fluid dynamics surrogate model, which directly outputs the corresponding three-dimensional microscopic wind speed vector through forward inference. ; For the three-dimensional microscopic wind speed vector The system is decomposed to obtain the microscopic wind speed and direction on the horizontal plane; and the ratio of the fluctuation standard deviation of the microscopic wind speed to the average wind speed within a time step is extracted to calculate the time series data of the turbulence intensity of the corresponding unit.

7. The extreme meteorological power prediction method based on typhoon path extrapolation according to claim 6, characterized in that, In step S3, specifically in step S300, the specific process of constructing and calculating the dynamic cut-out probability includes: Obtain the wind turbine at the predicted future time. micro wind speed With turbulence intensity ; Extract cabin vibration acceleration from the real-time operational feature data. and the yaw error rate of the crew ; The above features are input into a pre-built dynamic cut-out probability model based on multidimensional logistic regression to calculate the first... Typhoon turbine units at all times Dynamic handover probability of triggering shutdown protection The calculation formula is as follows: in, Basic bias term, These are the weight coefficients of the corresponding feature variables.

8. The extreme meteorological power prediction method based on typhoon path extrapolation according to claim 7, characterized in that, In step S4, the weight coefficients of each feature variable in the dynamic cut-out probability model are obtained through the following training steps: Collect SCADA system operation records of the target wind farm during historical typhoons, extract historical shutdown events and their corresponding micro wind speed, turbulence intensity, nacelle vibration acceleration and yaw error rate, and construct a historical sample set; The historical sample set is processed by binary labeling: the status label of the wind turbine being normally connected to the grid and generating electricity is set to 0, and the status label of the wind turbine being shut down and disconnected from the grid due to the triggering of the anti-typhoon self-protection mechanism is set to 1. The maximum likelihood estimation method is used to iteratively optimize the labeled historical sample set, aiming to minimize the cross-entropy loss function, and to fit and solve for the weight coefficients that maximize the probability of historical shutdown events. .

9. The extreme meteorological power prediction method based on typhoon path extrapolation according to claim 8, characterized in that, In step S400, the specific process for calculating the expected power of a single unit is as follows: Based on the theoretical wind speed-power curve of the wind turbine generator set from the manufacturer, calculate the first... Typhoon turbine units at micro wind speed Ideal predicted power ; Introducing the dynamic cut-out probability calculated in step S300 The ideal predicted power of a single unit is converted into the expected power of a single unit, which reflects the actual power generation capacity of the unit during the typhoon. Its mathematical expression is: in, This represents the basic standby power loss value of the wind turbine unit in the feathering state when it is stopped.

10. The extreme meteorological power prediction method based on typhoon path extrapolation according to claim 8, characterized in that, Based on the wake topology correction matrix under the current micro wind direction, the expected power of each individual wind turbine in the entire field is aggregated, specifically including the following sub-steps: S401: Microscopic wind direction output by the computational fluid dynamics surrogate model. Determine the upstream and downstream spatial topological relationships between wind turbines within the target wind farm, and select the appropriate turbines for the first... The set of all upstream turbine units affected by the wake of the typhoon generator. ; S402: Calculate the dynamic probability wake attenuation coefficient; when the upstream unit When the unit is shut down, its wake effect is reduced, therefore the upstream unit... Operating probability As a penalty factor, its impact on downstream units is calculated. Expected speed loss : in, For thrust coefficient, The diameter of the wind turbine, The wake expansion coefficient is... For the unit to the unit Effective distance along the wind direction; S403: Calculate the first... using the square sum root method The combined wake velocity deficit rate of the typhoon turbine ; S404: The expected power of all units in the wind farm is aggregated and corrected using the comprehensive wake velocity deficit rate, and the result is output as the target wind farm at the time of typhoon passage. Final predicted total power : in, This represents the total number of wind turbines within the wind farm.

Images