A method for predicting the two-stage power characteristics of wind field and generator unit under cold wave weather conditions

Abstract

This invention discloses a two-stage power characteristic prediction method for wind farms and wind turbines under cold wave meteorological conditions. The method includes: establishing a cold wave meteorological identification model to achieve accurate identification of cold wave weather; constructing wind speed-power characteristic curves with environmental parameter corrections and establishing a CFD simulation model for each wind turbine; then, establishing a turbine power prediction model by combining flow field characteristics and meteorological data, and using optimization algorithms to improve the model's accuracy and robustness; and constructing an overall wind farm power prediction model using the LightGBM machine learning algorithm to obtain the predicted power of the entire wind farm and individual turbines. This invention can significantly improve the accuracy and reliability of wind power prediction, reducing the operational risks of wind farms and the instability factors of the power system caused by prediction errors. Its functions include accurately predicting and evaluating the performance of wind turbines under different cold meteorological conditions, providing strong support for the optimized operation of wind farms and the stable operation of the power system.

Description

Technical Field

[0001] This invention discloses a two-stage power characteristic prediction method for wind farms and turbines under cold wave weather conditions, particularly relating to two-stage power prediction technology for wind farms and turbines under cold wave weather conditions in the wind power field. This method can effectively ensure stable power generation of wind power under cold wave weather conditions and maximize the reliability and economy of the power system. Background Technology

[0002] With the continuous growth of global demand for renewable energy, wind power has become one of the important clean energy sources. However, the power generation characteristics of wind power are highly uncertain, especially during cold weather. Changes in meteorological parameters such as wind speed, temperature, and humidity can have a profound impact on the power output of wind turbines. These changes can not only affect the operating status of wind turbines but also cause large fluctuations in power output, thus posing a serious challenge to the stable operation of the power grid. Therefore, accurately predicting wind power output has become a key factor in ensuring the efficient operation of wind farms.

[0003] During cold waves, meteorological parameters such as wind speed and temperature undergo drastic changes. Fluctuations in wind speed, a sharp drop in temperature, and the accumulation of ice and snow directly affect the operating efficiency and power output of wind turbines. For example, extreme low temperatures may cause changes in the viscosity of the lubricating oil in wind turbine units, thus affecting the normal operation of the turbines; strong winds may cause additional mechanical loads on the units, affecting the power generation performance of the turbines. These factors increase the instability of wind farm power generation under cold waves, making accurate prediction of wind power output particularly important for the rational scheduling and load distribution of wind farms.

[0004] Existing forecasting methods have several limitations under cold wave weather conditions. First, due to the rapid changes in meteorological conditions during cold waves, forecasting methods relying solely on conventional wind speed-power characteristic curves often fail to effectively address the rapid fluctuations in wind speed and temperature, resulting in insufficient forecast accuracy. Second, these methods struggle to capture the nonlinear and complex impacts of cold waves on wind turbine power output, especially under the combined effects of factors such as wind turbine lubricant viscosity and turbine layout; traditional models cannot adjust their forecasts in a timely manner. Furthermore, existing wind power forecasting methods typically exhibit poor adaptability to anomalous weather patterns, leading to significant errors in wind farm power forecasts under extreme weather conditions such as cold waves. By employing a novel two-stage power characteristic forecasting method for wind farms and turbines under cold wave weather conditions, the problems of insufficient accuracy, poor adaptability, and poor interpretability in traditional methods can be effectively addressed, providing more stable and accurate forecast results. Summary of the Invention

[0005] The purpose of this invention is to provide a two-stage power characteristic prediction method for wind farms and wind turbines under cold wave weather conditions. By combining, but not limited to, meteorological data, wind turbine operating status information, and wind turbine lubricating oil status information, and applying advanced machine learning methods such as LightGBM and Stacking, a two-stage power characteristic model for wind farms and wind turbines is established. This method can provide higher accuracy and more stable wind power prediction under cold wave weather conditions. Compared with existing technologies, this invention can effectively cope with meteorological changes under cold wave weather, improve the accuracy and robustness of wind power prediction, and provide strong technical support for the efficient operation of wind farms and the stable operation of the power grid.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a two-level power characteristic prediction method for wind farms and wind turbines under cold wave weather. This method performs power characteristic analysis at both the wind farm and wind turbine levels. By first predicting the power of a single wind turbine, the power of the entire wind farm is derived, forming a multi-level prediction structure from micro to macro. The specific steps are as follows:

[0007] S1 inputs the acquired meteorological data into a pre-trained cold wave weather identification model to obtain predictions of cold wave weather and its associated meteorological conditions. If the prediction indicates a cold wave, subsequent steps are performed. Based on historical meteorological data and wind turbine power data, S1 considers the different characteristics of cold wave weather and the temporal dynamics of meteorological changes. It uses a mean-shift clustering algorithm to achieve fine-grained classification of cold wave weather conditions, constructing a corresponding cold wave weather dataset containing meteorological features and their corresponding temporal dynamics. Combining the clustering results and cold wave weather data, machine learning is used to establish a cold wave weather identification model for accurate identification of cold waves.

[0008] Further detailing S1, S1 includes: S11: collecting historical meteorological data and normalizing or standardizing the collected data; S12: extracting cold wave characteristics from the meteorological data; labeling the time periods of cold waves and their corresponding meteorological characteristics based on historical meteorological data, classifying the cold wave weather in each time period according to the cold wave characteristics, and creating cold wave labels; S13: standardizing the cold wave characteristics selected from the meteorological data and then using Mean... The Shift clustering algorithm is used to classify different meteorological conditions under cold wave weather; S14: The annotation of cold wave events is combined with meteorological data to generate a cold wave weather dataset containing cold wave labels, and the dataset is divided into training set, validation set and test set for training and evaluation of the pre-built cold wave weather identification model; The cold wave weather identification model takes meteorological data as input and cold wave weather and its corresponding meteorological conditions as output; S15: The cold wave weather identification model is trained using the cold wave weather dataset, the hyperparameters are adjusted to improve the accuracy of the model, the model effect is evaluated using the validation set, and the cold wave weather identification model is obtained to achieve the identification of cold wave weather. S1 collects historical meteorological data, including wind speed, temperature, humidity, and wind turbine power output data, and normalizes or standardizes the data. It extracts temperature, humidity, wind speed, air pressure, and other cold wave meteorological characteristics. Based on the historical meteorological data, it labels the time periods of cold waves and their corresponding meteorological characteristics, classifying them according to characteristics such as extreme low temperatures and high wind speeds to create cold wave labels. After standardizing the cold wave-related feature data selected from the meteorological data, it uses the Mean Shift clustering algorithm to classify different cold wave meteorological conditions. It combines the cold wave event labels with the meteorological data to generate a dataset containing cold wave labels, dividing the dataset into training, validation, and test sets for subsequent model training and evaluation. Based on the cold wave dataset, it selects appropriate machine learning models, such as DNN or RF, and trains the model using the cold wave meteorological dataset. It adjusts hyperparameters to improve model accuracy and uses the validation set to evaluate the model's performance, resulting in a cold wave meteorological identification model for recognizing cold wave meteorological events.

[0009] In S2, temperature and humidity can jointly affect air density, and changes in air density directly affect the fan's power curve. The fan power and air density have the following relationship:

[0010]

[0011] Where P is the output power of the fan, and C P The power factor is an inherent characteristic of wind turbine generators and is closely related to the design and manufacturing of the generator set; S is the swept area of ​​the wind turbine; ρ is the air density; and V is the wind speed.

[0012] The air density ρ has the following relationship with temperature t:

[0013] ρ = 1.2761 + 0.0036t(p - 0.378p) w ) / 1000

[0014] Therefore, based on numerical weather prediction, wind field temperature, wind speed, humidity and related parameters are obtained, and combined with the actual working status of wind turbines, an environmental parameter-corrected wind speed-power characteristic curve is constructed. At the same time, the oil detection system is used to obtain the status information of the wind turbine lubricating oil in real time, which helps to build the subsequent CFD model.

[0015] Based on the wind speed-power characteristic curves corrected for environmental parameters, and combined with the wind turbine layout (location coordinates, spacing, number and arrangement of turbines in the wind farm, hub height, etc.), lubricating oil status information, and real-time operating parameters (input and output wind speeds, rated wind speed, number and diameter of blades), as well as environmental factors, a CFD simulation model of the turbines under cold wave weather is established. The CFD simulation model analyzes and simulates the flow field characteristics around the wind turbines during cold waves. By establishing the CFD model, the flow field changes during turbine operation are obtained, providing further data support for the establishment of a power prediction model for the turbines.

[0016] S4 is based on stacking ensemble learning. It establishes corresponding power characteristic prediction models for generator units based on flow field characteristics under different operating conditions and numerical weather forecasts. It uses existing data, including power, wind speed and direction, humidity, temperature, air pressure and meteorological data, and flow field characteristics including turbulence intensity, wind speed distribution and wake effect as input features. It also uses SFOA and / or HO optimization algorithms to optimize model parameters, thereby improving accuracy and robustness.

[0017] S5 performs preliminary power prediction based on the unit's power prediction characteristic model. Combining historical power data, wind speed, and humidity-related data, it uses a Transformer model or an MLP model to correct power prediction errors, and finally obtains the predicted power of the unit.

[0018] S6 acquires topographic and geomorphological data of the wind field using GIS or satellite remote sensing technology. Based on multi-source heterogeneous data, including the prediction results of the power characteristic prediction model of the turbine units, wind field numerical weather forecasts, wind field planning datasets from the National Satellite Meteorological Center, topographic and geomorphological data acquired through GIS or satellite remote sensing technology, and historical wind field data, it uses LightGBM machine learning to construct an overall power prediction model for the wind field. The wind field numerical weather forecasts include basic meteorological parameters, precipitation-related parameters, cloud visibility, radiation, and illumination parameters; the wind field planning dataset includes the geographical location information of the wind field (latitude, longitude, and altitude), meteorological parameters, wind energy resource assessment, and wind farm planning data.

[0019] S7 uses the Transformer model, which adaptively adjusts the model parameters during training to ultimately obtain the overall predicted power of the wind field.

[0020] Further steps S5 and S7 yield the predicted power of each wind turbine and the overall predicted power of the wind farm. This two-stage power prediction method, involving the wind farm and its turbines, aims to improve the accuracy and stability of power prediction for wind farms during cold weather, optimize wind farm operation and management, and is of great significance for the operation and management of wind farms and the stable operation of the power grid. Attached Figure Description

[0021] To more clearly illustrate the technical solutions implemented in this invention, the accompanying drawings required in the embodiments are briefly described below. It is obvious that the following drawings are only some embodiments of this invention, and those skilled in the art can derive other implementation schemes based on these drawings without creative effort.

[0022] Figure 1 This is the overall process architecture of a two-stage power characteristic prediction method for wind field and generator unit under cold wave weather conditions, as described in this invention.

[0023] Figure 2 This is the core architecture of the two-stage power characteristic prediction method for wind field and generator unit under cold wave meteorological line described in this invention, namely the two-stage power characteristic prediction model architecture for wind field and generator unit under cold wave meteorological conditions.

[0024] Figure 3 The process of classifying cold wave conditions and establishing a cold wave meteorological identification model was demonstrated. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. The embodiments described in this invention are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0026] like Figure 1 As shown, embodiments of the present invention propose an overall process architecture for a two-stage power prediction and analysis method for wind farms and turbines under cold wave conditions, including the following:

[0027] Input real-time numerical weather forecast data and relevant data on wind farms and turbines required for prediction into the two-level prediction model of wind farms and turbines to predict the power characteristics of wind farms and turbines.

[0028] like Figure 2As shown, the core of the two-stage power characteristic prediction method for wind farms and turbines under cold wave conditions proposed in this invention is to establish a two-stage power prediction model for wind farms and turbines, the specific content of which is as follows:

[0029] The process of classifying S1 cold wave meteorological conditions and establishing the cold wave meteorological model can be found in [link to relevant documentation]. Figure 3 The process involves collecting historical meteorological data and wind turbine power output data, normalizing or standardizing the data, extracting temperature, humidity, wind speed, air pressure, and other cold wave meteorological characteristics, labeling the time periods of cold waves and their corresponding characteristics based on historical meteorological data, and classifying cold wave data according to characteristics such as extreme low temperatures and high wind speeds to create cold wave labels. After standardizing the cold wave-related feature data selected from the meteorological data, the Mean Shift clustering algorithm is used to classify different meteorological conditions under cold wave conditions. The labeled cold wave events are combined with the meteorological data to generate a dataset containing cold wave labels, which is then divided into training, validation, and test sets for subsequent model training and evaluation. Based on the cold wave dataset, a suitable machine learning model, such as DNN or RF, is selected, and the model is trained using the cold wave meteorological dataset. Hyperparameters are adjusted to improve model accuracy, and the validation set is used to evaluate the model's performance, resulting in a cold wave meteorological identification model.

[0030] In S2, temperature and humidity can jointly affect air density, and changes in air density directly affect the fan's power curve. The fan power and air density have the following relationship:

[0031]

[0032] Where P is the output power of the fan, and C P The power factor is an inherent characteristic of wind turbine generators and is closely related to the design and manufacturing of the generator set; S is the swept area of ​​the wind turbine; ρ is the air density; and V is the wind speed.

[0033] The air density ρ has the following relationship with temperature t:

[0034] ρ = 1.2761 + 0.0036t(p - 0.378p) w ) / 1000

[0035] Therefore, by combining numerical weather forecasts to obtain parameters such as wind field temperature, wind speed, and humidity, and combining them with the actual operating status of wind turbines, a wind speed-power characteristic curve with environmental parameter correction is constructed. At the same time, the lubricating oil status information of wind turbines is obtained in real time using an oil detection system, which helps to build the subsequent CFD model.

[0036] S3 uses wind speed-power characteristic curves corrected for environmental parameters to establish CFD simulation models for each wind turbine under cold wave weather conditions, based on wind turbine layout, lubricating oil status information, real-time wind turbine operating parameters, and environmental factors. It simulates and analyzes the flow field characteristics around the wind turbine under cold wave weather conditions. By establishing CFD models, it obtains the flow field changes during wind turbine operation, providing further input data support for the establishment of power prediction models.

[0037] S4 is based on stacking ensemble learning. It establishes corresponding power characteristic prediction models for generator units based on flow field characteristics under different operating conditions and numerical weather forecasts. It uses existing data, including power, wind speed and direction, humidity, temperature, air pressure and meteorological data, and flow field characteristics including turbulence intensity, wind speed distribution and wake effect as input features. It also uses SFOA and / or HO optimization algorithms to optimize model parameters, thereby improving accuracy and robustness.

[0038] S5 performs preliminary power prediction based on the unit's power prediction characteristic model. Combining historical data, wind speed, humidity, lubricating oil status information of various unit components, and other relevant factors, it uses the Transformer model or MLP model to correct power prediction errors, and finally obtains the predicted power of the unit.

[0039] S6 acquires topographic and geomorphological feature data of the wind field using GIS or satellite remote sensing technology. Based on multi-source heterogeneous data, including the prediction results of the power characteristic prediction model of the turbine units, wind field numerical weather forecasts, wind field planning datasets from the National Satellite Meteorological Center, topographic factors acquired through GIS or satellite remote sensing technology, and historical wind field data, it uses LightGBM machine learning to construct an overall power prediction model for the wind field. The wind field numerical weather forecasts include basic meteorological parameters, precipitation-related parameters, cloud visibility, radiation, and illumination parameters; the wind field planning dataset includes the geographical location information of the wind field (latitude, longitude, and altitude), meteorological parameters, wind energy resource assessment, and wind farm planning data.

[0040] S7 uses the Transformer model, which adaptively adjusts the model parameters during training to ultimately obtain the overall predicted power of the wind field.

[0041] Further steps S5 and S7 yield the predicted power of each wind turbine and the overall predicted power of the wind farm. This two-stage power prediction method, involving the wind farm and turbines, aims to improve the accuracy and stability of power prediction for wind farms under cold weather conditions, optimize wind farm operation and management, and is of great significance for the operation and management of wind farms and the stable operation of the power grid.

[0042] The above scheme will be explained in detail below:

[0043] In S1, meteorological data is collected from historical meteorological data and wind turbine power output data to normalize or standardize the data. Temperature, humidity, wind speed, air pressure, and other cold wave meteorological characteristics are extracted, and the time periods of cold waves and their corresponding meteorological characteristics are labeled based on historical meteorological data. Cold wave labels are created by classifying cold wave characteristics such as extreme low temperatures and high wind speeds. After standardizing the cold wave-related feature data selected from the meteorological data, the Mean Shift clustering algorithm is used to classify different meteorological conditions under cold wave conditions. First, a window radius R is selected, and a data point is chosen as the window center. The weight α of each data point is calculated using a Gaussian kernel function. i The weight calculation formula is as follows:

[0044]

[0045] The parameters are represented as follows: α i : is the weight of the i-th data point G(·); : is the Gaussian kernel function x i : is the position of the i-th data point, c: is the position of the current window center, σ: is the standard deviation of the Gaussian kernel function. σ =R, ||x i -c|| 2 : is data point x i The square of the distance between the current window center c and the center c.

[0046] After obtaining the weight value of each data point, substitute it into the position update formula.

[0047]

[0048] The parameters represent the following meanings: c new : New window center position, ω i : The weight of the i-th data point, x i : is the position of the i-th data point, n: is the total number of data points.

[0049] The center position of the window is updated by calculating the weighted average position of all data points. The weights are determined by a Gaussian kernel function based on the distance between the data points and the current window center. This weight calculation and position update process is repeated iteratively until all data points are clustered into different clusters. These clusters are then associated with different operating conditions, specifically normal operating conditions, low-temperature operating conditions, blade icing conditions, and extreme cold freezing conditions. The annotations of cold wave events are combined with meteorological data to generate a dataset containing cold wave labels. This dataset is divided into training, validation, and test sets for subsequent model training and evaluation. Based on the cold wave dataset, a suitable machine learning model, such as DNN or RF, is selected. The model is trained using the cold wave meteorological dataset, and hyperparameters are adjusted to improve model accuracy. The validation set is used to evaluate the model's performance, resulting in a cold wave meteorological identification model.

[0050] In S2, temperature and humidity can jointly affect air density, and changes in air density directly affect the fan's power curve. The fan power and air density have the following relationship:

[0051]

[0052] Where P is the output power of the fan, and C P The power factor is an inherent characteristic of wind turbine generators and is closely related to the design and manufacturing of the generator set; S is the swept area of ​​the wind turbine; ρ is the air density; and V is the wind speed.

[0053] The air density ρ has the following relationship with temperature t:

[0054] ρ = 1.2761 + 0.0036t(p - 0.378p) w ) / 1000

[0055] Where p is atmospheric pressure, p w It is the water vapor pressure.

[0056] Therefore, by combining numerical weather forecasts to obtain parameters such as wind field temperature, wind speed, and humidity, and combining them with the actual operating status of wind turbines, a wind speed-power characteristic curve with environmental parameter correction is constructed. At the same time, the lubricating oil status information of wind turbines is obtained in real time using an oil detection system, which helps to build the subsequent CFD model.

[0057] Based on the wind speed-power characteristic curves corrected for environmental parameters, and combined with wind turbine layout, lubricating oil status information, real-time turbine operating parameters, and environmental factors, S3 establishes a CFD simulation model of the turbine under cold wave weather conditions. The CFD simulation model is built using ANSYS Fluent or OpenFOAM to simulate and analyze the flow field characteristics around the wind turbine under cold wave weather conditions. Establishing a CFD model helps us understand the flow field changes during wind turbine operation under cold weather conditions, especially phenomena such as airflow turbulence, eddies, and wind speed changes that may occur in low-temperature environments. Through CFD simulation, we can also obtain detailed flow field parameters, such as wind speed distribution, airflow direction, pressure changes, and temperature gradients. These data provide important references for the real-time operating status of the wind turbine. Simultaneously, these simulation results can further support the establishment of power prediction models. During the CFD simulation, either the K-ε or K-ω model can be selected to simulate the turbulence characteristics of the wind field. The standard K-ε equation for the turbulence model is as follows:

[0058]

[0059] The parameters are represented as follows: k: turbulent kinetic energy, t: time, U: velocity vector. Gradient operator, P k : Turbulence generation term, ε: Turbulence dissipation rate, μ t : Turbulent viscosity, σ k σ ε C ε1 C ε2 : Turbulence model constants. When setting boundary conditions, input wind speed and temperature under cold wave weather conditions, consider the impact of low temperature on lubricating oil performance, and set the fan speed and power output as real-time parameters. For wind speed under cold wave weather conditions, the boundary conditions are:

[0060] U = U wind

[0061] Among them, U wind This refers to wind speed.

[0062] For temperature, use temperature boundary conditions:

[0063] T wind =T ambient

[0064] Among them, T wind The temperature is caused by wind speed, T ambient It refers to the ambient temperature.

[0065] Furthermore, the effect of low temperature on lubricating oil viscosity is usually characterized using a temperature-dependent viscosity model. The relationship between temperature and viscosity can be expressed by the Arrhenius equation:

[0066]

[0067] The parameters are represented as follows: μ(T): viscosity at temperature T, μ0: viscosity at the reference temperature, E a Activation energy, R: gas constant, T: temperature, T0: reference temperature

[0068] S4, based on stacking ensemble learning, establishes corresponding unit power characteristic prediction models according to flow field characteristics under different operating conditions and numerical weather forecasts. It utilizes existing data, including power, wind speed, wind direction, humidity, temperature, and atmospheric pressure meteorological data, as well as flow field characteristics including turbulence intensity, wind speed distribution, and wake effects, as input features. Since SFOA and HO optimization algorithms can effectively explore the hyperparameter space and improve the model's accuracy and robustness, they are used to optimize the model parameters to enhance accuracy and robustness.

[0069] S5 performs preliminary power prediction based on the unit's power prediction characteristic model. Combining historical data, wind speed, humidity, and other relevant factors, it uses either a Transformer model or an MLP model to correct power prediction errors. This is just one example. The Transformer model uses time-series data as its foundation and learns the temporal patterns of errors using a self-attention mechanism. Through training, the Transformer model can capture the patterns of errors in power prediction, thereby adjusting the original prediction results. During the Transformer model training process, power prediction errors from historical data are used for training to optimize the model's hyperparameters (such as the number of attention layers and the dimension of hidden layers). Through continuous adjustment and optimization, it is ensured that the Transformer model can accurately correct deviations in power prediction. Finally, the corrected results output by the Transformer model are applied to the preliminary predicted power of the unit's power prediction characteristic model to obtain the final predicted power of the unit.

[0070] To construct a comprehensive power prediction model for the wind farm, we first acquired topographic and geomorphological data using Geographic Information System (GIS) or satellite remote sensing technology. This data includes information on the wind farm's topography, landforms, elevation, and slope, helping us understand the wind speed distribution in different areas of the wind farm and its impact on turbine power output. Next, based on multi-source heterogeneous data, including turbine power characteristic prediction model results, wind farm numerical weather prediction, wind farm planning datasets, historical wind farm data, and topographic factors, we used the machine learning algorithm LightGBM to construct a comprehensive power prediction model for the wind farm. Specifically, we first collected and preprocessed all relevant data, including historical turbine power data, numerical weather prediction data, and topographic data. Then, because LightGBM can handle large-scale, multi-dimensional data and improve the model's predictive ability through ensemble learning, we input this multi-source heterogeneous data into the LightGBM model for training, thereby constructing the wind farm power prediction model.

[0071] S7 employs a Transformer model, adaptively adjusting model parameters during training to ultimately obtain the overall predicted power of the wind farm. This process effectively improves the accuracy and robustness of wind farm power prediction, providing more reliable data support for wind farm operation optimization.

Claims

1. A method for predicting the two-stage power characteristics of wind field and turbine unit under cold wave weather, characterized in that, Includes the following steps: S1: Input the acquired meteorological data into the pre-trained cold wave meteorological identification model to obtain the prediction results of cold wave meteorology and the meteorological conditions to which it belongs. If the prediction result is cold wave meteorology, proceed to the next step. S2: Based on numerical weather forecasting, environmental parameters are obtained, and a wind speed-power characteristic curve after environmental parameter correction is constructed under the actual working conditions of the wind turbine. At the same time, the lubricating oil status information of the wind turbine is obtained in real time using an oil detection system. S3: Based on the wind speed-power characteristic curve corrected by environmental parameters, a CFD simulation model of each wind turbine under cold wave weather is established using wind turbine layout information, lubricating oil status information, real-time wind turbine operating parameters and environmental parameters to simulate the flow field characteristics around the wind turbine under cold wave weather and obtain the flow field changes during wind turbine operation through CFD simulation. S4: Construct a unit power characteristic prediction model, which takes meteorological data and flow field characteristics under different meteorological conditions as input and unit power as output; S5: Based on the unit power characteristic prediction model, a preliminary predicted power is obtained; combined with historical power data, environmental parameters and lubricating oil status information, the preliminary predicted power is corrected using a Transformer model or an MLP model to obtain the unit predicted power of the wind turbine. S6: Use LightGBM machine learning to construct a power prediction model for the overall wind farm. The power prediction model for the overall wind farm takes the predicted power of the units obtained in S5, as well as the pre-acquired numerical weather forecast data, wind farm planning data, topographic feature data and wind farm SCADA data as inputs, and the overall power of the wind farm as output. S7: Using the Transformer model, the model parameters of the overall wind field power prediction model are adaptively adjusted during the training process to obtain the corrected overall wind field predicted power.

2. The method for predicting the two-stage power characteristics of wind field and generator unit under cold wave meteorological conditions as described in claim 1 is characterized in that, The establishment of the cold wave weather identification model described in S1 includes the following steps: S11: Collect historical meteorological data and normalize or standardize the collected data. S12: Extract cold wave characteristics from meteorological data; label the time periods of cold waves and their corresponding cold wave characteristics based on historical meteorological data; classify the cold wave weather in each time period based on the cold wave characteristics; and create cold wave labels. S13: After standardizing the cold wave characteristics selected from the meteorological data, the Mean Shift clustering algorithm is used to classify different meteorological conditions under the cold wave weather. S14: Combine the annotation of cold wave events with meteorological data to generate a cold wave meteorological dataset containing cold wave labels, and divide the dataset into a training set, a validation set, and a test set for training and evaluation of a pre-built cold wave meteorological identification model; the cold wave meteorological identification model takes meteorological data as input and outputs cold wave meteorology and the corresponding meteorological conditions. S15: Train the cold wave weather identification model using the cold wave weather dataset, adjust the hyperparameters to improve the model's accuracy, evaluate the model's performance using the validation set, and obtain the cold wave weather identification model to achieve the identification of cold wave weather.

3. The method for predicting the two-stage power characteristics of wind field and generator unit under cold wave meteorological conditions according to claim 1, characterized in that: In S2, the environmental parameters include the temperature, wind speed, and humidity of the wind field.

4. The method for predicting the two-stage power characteristics of wind field and generator unit under cold wave meteorological conditions according to claim 3, characterized in that: In S3, the wind turbine layout information includes the wind turbine's location coordinates, spacing, number and arrangement of wind turbines in the wind farm, and hub height; the lubricating oil status information includes the lubricating oil viscosity, temperature, and moisture content; and the wind turbine's real-time operating parameters include cut-in and cut-out wind speeds, rated wind speed, and number and diameter of blades.

5. The method for predicting the two-stage power characteristics of wind field and generator unit under cold wave meteorological conditions according to claim 2, characterized in that: In S4, the unit power characteristic prediction model adopts the Stacking algorithm in the ensemble learning method and is established based on the flow field characteristics and meteorological data under different meteorological conditions; the meteorological data includes wind speed, wind direction, humidity, temperature and air pressure; the flow field characteristics include turbulence intensity, wind speed distribution and wake effect.

6. The method for predicting the two-stage power characteristics of wind field and generator unit under cold wave meteorological conditions according to claim 1, characterized in that: In S6, the terrain and landform feature data are obtained through GIS or satellite remote sensing technology, and the terrain and landform feature data include the topography, landform, elevation, and slope of the wind field; The wind field numerical weather forecast data includes basic meteorological parameters, precipitation-related parameters, cloud visibility, radiation, and illumination. The wind farm planning data includes geographical location information such as latitude, longitude, and altitude of the wind farm, meteorological parameters, wind energy resource assessment, and wind farm planning data.

Images