A method, system, device and medium for predicting icing of a wind turbine

Abstract

The present application relates to the technical field of wind farm equipment monitoring, and discloses a wind turbine icing prediction method, system, device and medium, in response to an icing prediction request, wind turbine live data and numerical prediction data of a wind turbine in a wind farm are acquired and preprocessed respectively, the numerical prediction data after preprocessing is corrected by using a Bayesian model averaging method, corrected numerical prediction data is generated, a target icing type prediction model is constructed based on an LSSVM algorithm, the corrected numerical prediction data is input into the target icing type prediction model for prediction, a predicted icing type is output, an associated icing model is matched according to the predicted icing type for icing prediction, and wind turbine icing results are generated; the technical problem of the present application is that the data processing of the existing correction method for mode prediction data is relatively cumbersome, and there is a lack of a method for correcting multi-mode set prediction data, which further leads to insufficient accuracy of icing prediction.

Description

Technical Field

[0001] This invention relates to the field of wind farm equipment monitoring technology, and in particular to a method, system, equipment and medium for predicting wind turbine icing. Background Technology

[0002] Global climate change and the energy crisis are gradually becoming two major challenges facing contemporary society, and the development and utilization of renewable energy has become a key way to solve these problems. Wind energy, as a clean and efficient energy source, has attracted widespread attention and promotion due to its high power generation resource utilization rate, relatively mature technology, and decreasing costs with technological advancements, and is considered an important driving force for energy transition. However, wind turbine generators are quite sensitive to changes in weather conditions. For example, in humid climates during winter and spring, the blades are prone to icing. Icing alters the aerodynamic characteristics of the blades and increases the dynamic load on the unit, negatively impacting the safe operation and efficiency of wind farms. It may also damage the turbine structure and even cause safety accidents.

[0003] Current research on wind turbine icing prediction based on various machine learning algorithms is mostly based on artificial climate laboratories and computer simulation methods. There are few studies on using numerical forecasting to predict blade ice thickness, and the main focus is on predicting the start and end times of icing.

[0004] Current numerical weather prediction models primarily use deterministic forecasts, i.e., forecasts from a single model. Due to the uncertainties and chaotic characteristics of atmospheric motion, forecast accuracy is somewhat affected. In such cases, ensemble forecasts that incorporate consideration of uncertainties offer a viable solution. Furthermore, the influence of local micro-meteorological environments caused by wind turbines necessitates corrections to model forecast data. While numerous methods exist for correcting meteorological elements using statistical methods or machine learning / deep learning, these are mostly applied to correcting single-model data or to correcting multiple models individually before using ensemble learning. This data processing is cumbersome, and research on the correlation correction of multiple models (ensemble models) is limited. Summary of the Invention

[0005] This invention provides a method, system, device, and medium for predicting wind turbine icing. It solves the technical problem that most existing methods for correcting model forecast data focus on correcting single-model data or using ensemble learning to process multiple models after correcting them individually. This results in cumbersome data processing and a lack of methods for correcting multi-model ensemble forecast data, leading to insufficient accuracy in icing prediction.

[0006] The first aspect of this invention provides a method for predicting wind turbine icing, comprising: In response to a request for icing prediction of wind turbines in a wind farm, the system acquires real-time data and numerical forecast data of the wind turbines in the wind farm. The actual wind turbine data and the numerical forecast data are preprocessed respectively; The preprocessed numerical forecast data is corrected using the Bayesian model averaging method to generate corrected numerical forecast data. Based on the LSSVM algorithm, a target icing type prediction model is constructed using the preprocessed wind turbine real-time data; The modified numerical forecast data is used as input to the target icing type prediction model to predict the icing type, and the predicted icing type is output. Icing prediction is performed based on the icing model associated with the predicted icing type, and icing results for the wind turbine are generated.

[0007] Optionally, the wind turbine real-time data includes historical meteorological data and basic wind turbine data. The step of preprocessing the wind turbine real-time data and the numerical forecast data respectively includes: The historical meteorological data of the wind turbines are processed for missing values ​​and outliers respectively to generate the target wind turbine's real-time data; The numerical forecast data is interpolated using bilinear interpolation to generate initial numerical forecast data. Based on the preset wind speed correction model, the wind speed forecast data in the initial numerical forecast data is corrected using the basic wind turbine data to generate the target numerical forecast data.

[0008] Optionally, the preset wind speed correction model is specifically:

[0009] In the formula, Indicates corrected wind speed. This refers to the wind speed forecast data. This indicates the radius from the cross section to the blade root. Indicates the relative wind speed angle of attack. This represents the ratio of the relative velocity to the wind speed at the location of the cross-section. Indicates the tip speed ratio, Indicates the design speed. Indicates the fan speed. Indicates the fan's rotation radius. Indicates the linear velocity at the blade tip. Indicating the axial inducing factor of leaf pigment, Indicates the tangential inducing factor of leaf pigment. The leaf element angle represents the cross-section of the blade.

[0010] Optionally, the step of correcting the preprocessed numerical forecast data using the Bayesian model averaging method to generate corrected numerical forecast data includes: Based on the pre-set BMA prediction model, a normal distribution is used to fit the temperature to construct a temperature BMA model; Based on the pre-set BMA prediction model, the precipitation is fitted using the logistic regression model and the gamma regression model to construct the precipitation BMA model. Based on the pre-set BMA prediction model, the wind speed is fitted using the Weibull distribution to construct a wind speed BMA model. The posterior probabilities associated with the temperature BMA model, the precipitation BMA model, and the wind speed BMA model are obtained respectively using the expectation-maximization algorithm. The temperature forecast data and associated posterior probability in the preprocessed numerical forecast data are used as inputs to the temperature BMA model for correction, and the corrected temperature forecast data is output. The precipitation forecast data and the associated posterior probability in the preprocessed numerical forecast data are used as inputs to the precipitation BMA model for correction, and the corrected precipitation forecast data is output. The wind speed forecast data and the associated posterior probability in the preprocessed numerical forecast data are used to input the wind speed BMA model for correction, and the corrected wind speed forecast data is output.

[0011] Optionally, the step of constructing a target icing type prediction model based on the LSSVM algorithm using the preprocessed wind turbine real-time data includes: The preprocessed wind turbine real-time data is divided into test datasets and training datasets; Initialize the model parameters of the preset initial icing type prediction model; Based on the particle swarm optimization algorithm, the training dataset is used as input to the preset initial icing type prediction model to optimize the parameters and determine the optimal model parameters. The preset initial icing type prediction model is updated using the optimal model parameters to generate an intermediate icing type prediction model; The intermediate icing type prediction model is tested using the test dataset, and test indicators are generated. When the test indicators meet the preset test conditions, the target icing type prediction model is output.

[0012] Optionally, the corrected numerical forecast data further includes liquid water content data, collision rate, capture rate, freezing rate, and effective icing cross-section data. The step of predicting icing based on the icing model associated with the predicted icing type and generating wind turbine icing results includes: When the predicted icing type is rime ice, then the associated rime ice icing model is matched; The model is input with the liquid water content data, preset water density data, preset ice density data, the corrected precipitation forecast data, and the corrected wind speed forecast data, and outputs the rime icing data corresponding to the wind turbines of the wind farm. The specific model for ice formation and icing is as follows:

[0013] In the formula, The data on frost and ice accumulation specifically refers to the cumulative amount of freezing rain. This indicates the liquid water content data. This refers to the revised precipitation forecast data. Indicates the correction parameter. This represents the preset water density data. This refers to the corrected wind speed forecast data. This represents the preset ice density data; When the predicted icing type is rime and mixed rime, then the associated rime and mixed rime icing models are matched. The corrected wind speed forecast data, liquid water content data, collision rate, capture rate, freezing rate, and effective icing cross section data are used as inputs to the rime and mixed frost icing model, and the rime and mixed frost data corresponding to the wind turbines of the wind farm are output. The specific models for rime ice and mixed rime ice icing are as follows:

[0014] In the formula, This refers to the data on rime and mixed rime, specifically the icing rate. This represents the collision rate. Indicates the capture rate, This indicates the freezing rate. This represents the effective icing cross-section data.

[0015] Optionally, the corrected numerical forecast data further includes convection exchange coefficients, and also includes: When the corrected temperature forecast data is greater than the preset melting surface temperature, the corresponding melting model is matched; The modified air temperature forecast data, the preset ice melting surface temperature, the convection exchange coefficient, the preset Stefan-Boltzmann constant, the preset radiation constant, the preset ice specific heat data, and the preset water freezing latent heat data are used as inputs to the ice melting model, and the ice melting data corresponding to the wind turbines of the wind farm are output. The ice-melting model is specifically as follows:

[0016] In the formula, The aforementioned ice melt data specifically refers to the ice thickness growth rate. This represents the convection exchange coefficient. This refers to the revised temperature forecast data. This indicates the preset ice-melting surface temperature. This represents the preset Stefan-Boltzmann constant. This represents the preset radiation constant. This represents the preset ice specific heat data. This represents the preset latent heat of water freezing.

[0017] A second aspect of the present invention provides a wind turbine icing prediction system, comprising: The response module is used to respond to the icing prediction request of the wind turbines in the wind farm and to obtain the actual wind turbine data and numerical forecast data of the wind turbines in the wind farm. The data preprocessing module is used to preprocess the actual wind turbine data and the numerical forecast data respectively; The data correction module is used to correct the preprocessed numerical forecast data using the Bayesian model averaging method to generate corrected numerical forecast data. The model building module is used to build a target icing type prediction model based on the LSSVM algorithm and the preprocessed wind turbine real-time data. The icing type prediction module is used to input the modified numerical forecast data into the target icing type prediction model to predict the icing type and output the predicted icing type. The icing result module is used to predict icing based on the predicted icing type and the associated icing model, and generate wind turbine icing results.

[0018] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the wind turbine icing prediction method as described in any of the preceding claims.

[0019] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the wind turbine icing prediction method as described in any of the preceding claims.

[0020] As can be seen from the above technical solutions, the present invention has the following advantages: In response to a request for icing prediction of wind turbines in a wind farm, this paper acquires real-time wind turbine data and numerical weather prediction (NMR) data. Both data are preprocessed. A Bayesian model averaging method is used to correct the preprocessed NMR data, generating corrected NMR data. Based on the LSSVM algorithm, a target icing type prediction model is constructed using the preprocessed real-time data. The corrected NMR data is then input into this model to predict the icing type, outputting the predicted icing type. Icing prediction is then performed based on the predicted icing type and associated icing models, generating the wind turbine icing results. This approach addresses the shortcomings of existing methods for correcting model forecast data, which often focus on single-model data or correct multiple models individually before using ensemble learning. These methods are cumbersome and lack methods for correcting multi-model ensemble forecast data. This invention addresses the technical problem of insufficient accuracy in icing prediction. Firstly, it considers not only single forecast values ​​but also ensemble forecasts or single-model forecasts from different start times, increasing the data sources for icing prediction. Furthermore, multi-member (ensemble) forecasts also contribute to increased accuracy. Secondly, it uses Bayesian model averaging (BMA) to correct for temperature, precipitation, and wind speed, improving the availability of numerical forecast data and expanding the prediction methods for traditional icing models on wind turbines. Thirdly, it uses machine learning algorithms to fit historical data to determine icing types, taking into account the specific characteristics of different environments, resulting in strong local applicability and generalizability. Finally, it improves the accuracy of icing prediction by using an improved Jones simple model that considers melting conditions and the Makkonen icing prediction model. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of the steps of a wind turbine icing prediction method provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of the steps of a wind turbine icing prediction method provided in Embodiment 2 of the present invention; Figure 3 This is an overall technical roadmap for wind turbine icing prediction provided in Embodiment 2 of the present invention; Figure 4 This is a BMA technology roadmap provided in Embodiment 2 of the present invention; Figure 5 This is a flowchart of the EM algorithm provided in Embodiment 2 of the present invention; Figure 6 This is a flowchart of the modeling process for the target icing type prediction model provided in Embodiment 2 of the present invention; Figure 7 This is a structural block diagram of a wind turbine icing prediction system provided in Embodiment 3 of the present invention; Figure 8 A structural block diagram of a computer device according to an embodiment of the present invention is shown. Detailed Implementation

[0023] This invention provides a wind turbine icing prediction method, system, device, and medium to address the technical problem that existing methods for correcting model forecast data mostly focus on correcting single-model data or using ensemble learning to process multiple models after correcting them individually. This results in cumbersome data processing and a lack of methods for correcting multi-model ensemble forecast data, leading to insufficient accuracy in icing prediction.

[0024] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only 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 scope of protection of this invention.

[0025] Since current numerical weather prediction models primarily use deterministic forecasts, i.e., single-model forecasts, the accuracy is affected by the uncertainties and chaotic characteristics of atmospheric motion. In such cases, ensemble forecasts that incorporate consideration of uncertainties offer a viable solution. Furthermore, the influence of local micro-meteorological environments caused by wind turbines necessitates corrections to model forecast data. While many methods exist for correcting meteorological elements, they are mostly applied to single-model data or involve correcting multiple models individually and then processing the data using ensemble learning methods, resulting in cumbersome data processing. Moreover, research on the correlation correction of multiple models (ensemble models) is limited. To address these issues, this paper proposes a wind turbine icing prediction method based on multi-member forecasts and Bayesian methods for correcting meteorological elements. By using corrected multi-member forecast data for icing prediction, the accuracy and applicability of icing predictions are improved. This provides early warning support for anti-icing and de-icing systems, promoting the efficient utilization of wind energy.

[0026] Example 1

[0027] Please see Figure 1 , Figure 1This is a flowchart illustrating the steps of a wind turbine icing prediction method provided in Embodiment 1 of the present invention.

[0028] The present invention provides a method for predicting wind turbine icing, comprising: Step 101: Respond to the icing prediction request for wind turbines in the wind farm and obtain the actual wind turbine data and numerical forecast data.

[0029] An icing prediction request refers to a request command for predicting icing on wind turbines in a wind farm.

[0030] The wind turbine's real-time data includes historical meteorological data and basic wind turbine data. The historical meteorological data consists of historical meteorological observation data of the wind turbine, while the basic wind turbine data consists of design specification data, such as basic wind turbine parameters like wind turbine speed, wind turbine rotation radius, and blade tip linear velocity.

[0031] Numerical weather forecast data refers to forecast data from a set of numerical models.

[0032] It is worth mentioning that hourly forecast accuracy is fundamental for icing prediction, so hourly numerical model ensemble forecast data can be used. However, considering the limitations of some ensemble forecast data, such as excessively coarse resolution and difficulty in obtaining them, hourly forecast data from high-precision single numerical models at different start times can be used. These single numerical models with different start times can be treated as a single ensemble member, with priority given to data with shorter start time intervals. A shorter start time represents an extension of the usable ensemble forecast period.

[0033] In this embodiment of the invention, in response to a request command to predict icing of wind turbines in a wind farm, the real-time data and numerical forecast data of the wind turbines in the wind farm are obtained.

[0034] Step 102: Preprocess the actual wind turbine data and numerical forecast data respectively.

[0035] Preprocessing involves handling missing and outlier values ​​in the historical meteorological data of the wind turbines in the actual data, interpolating the numerical forecast data, and then correcting the wind speed forecast data in the numerical forecast data using a preset wind speed correction model.

[0036] In this embodiment of the invention, missing value processing and outlier processing are performed on the historical meteorological data of the wind turbine in the actual wind turbine data to generate preprocessed actual wind turbine data. Missing value processing and outlier processing are performed on the historical meteorological data of the wind turbine in the actual wind turbine data. Interpolation operation is performed on the numerical forecast data. Then, the wind speed forecast data in the numerical forecast data is corrected by a preset wind speed correction model to generate preprocessed numerical forecast data.

[0037] Step 103: Use the Bayesian model averaging method to correct the preprocessed numerical forecast data and generate corrected numerical forecast data.

[0038] Bayesian Model Averaging (BMA) is a method used to calculate the weights of multiple models, primarily applied in data analysis and estimation. Based on Bayes' theorem, BMA uses the probability of falling into different models as the standard for their weights, thus obtaining a comprehensive evaluation of multiple models.

[0039] In this embodiment of the invention, the Bayesian model averaging method is used to construct forecast data corresponding to various meteorological elements in the preprocessed numerical forecast data, and then the data is corrected to generate corrected numerical forecast data.

[0040] Step 104: Based on the LSSVM algorithm, construct a target icing type prediction model using preprocessed wind turbine real-time data.

[0041] In this embodiment of the invention, a least squares support vector machine (LSSVM) is used as the initial prediction model, and the model parameters are optimized by particle swarm optimization to determine the optimal model parameters. Then, preprocessed wind turbine real-time data is input into the model for training to construct a target icing type prediction model.

[0042] Step 105: Input the modified numerical weather prediction data into the target icing type prediction model to predict the icing type and output the predicted icing type.

[0043] In this embodiment of the invention, the modified numerical forecast data is input into the target icing type prediction model to predict the icing type, and then the predicted icing type of the wind turbines in the wind farm is output.

[0044] Step 106: Perform icing prediction based on the icing model associated with the predicted icing type and generate wind turbine icing results.

[0045] In this embodiment of the invention, based on the predicted icing type of the wind turbines in the wind farm, an icing model associated with the icing type is matched, and modified numerical forecast data is input into the icing model to output the icing data of the wind turbines in the wind farm.

[0046] In this invention, in response to a request for icing prediction of wind turbines in a wind farm, real-time wind turbine data and numerical weather prediction data are acquired. These data are preprocessed, and the preprocessed numerical weather prediction data is corrected using a Bayesian model averaging method to generate corrected numerical weather prediction data. Based on the LSSVM algorithm, a target icing type prediction model is constructed using the preprocessed real-time wind turbine data. The corrected numerical weather prediction data is then input into the target icing type prediction model to predict the icing type, outputting the predicted icing type. Icing prediction is then performed based on the icing model associated with the predicted icing type, generating the wind turbine icing result. This invention addresses the problem that existing methods for correcting model forecast data mostly focus on correcting single-model data or correcting multiple models individually before using ensemble learning, resulting in cumbersome data processing and a lack of methods for correcting multi-model ensemble forecast data. The present invention addresses the technical problem of insufficient accuracy in icing prediction by addressing several issues. Firstly, it considers not only single forecast values ​​but also ensemble forecasts or single-model forecasts from different start times, increasing the data sources for icing prediction. Furthermore, multi-member (ensemble) forecasts also contribute to increased accuracy. Secondly, it uses Bayesian model averaging (BMA) to correct for temperature, precipitation, and wind speed, improving the availability of numerical forecast data and expanding the prediction methods for traditional icing models on wind turbines. Thirdly, it uses machine learning algorithms to fit historical data to determine icing types, taking into account the specific characteristics of different environments, resulting in strong local applicability and generalizability. Finally, it improves the accuracy of icing prediction by combining an improved Jones simple model that considers melting conditions with the Makkonen icing prediction model.

[0047] Example 2

[0048] Please see Figure 2 , Figure 2 This is a flowchart illustrating the steps of a wind turbine icing prediction method provided in Embodiment 2 of the present invention. Please refer to... Figure 3 , Figure 3 This is the overall technical roadmap for wind turbine icing prediction provided in Embodiment 2 of the present invention.

[0049] The present invention provides a method for predicting wind turbine icing, comprising: Step 201: Respond to the icing prediction request for wind turbines in the wind farm and obtain the actual wind turbine data and numerical forecast data.

[0050] In this embodiment of the invention, the specific implementation process of step 201 is similar to that of step 101, and will not be repeated here.

[0051] Furthermore, the wind turbine real-time data includes historical meteorological data and basic wind turbine data.

[0052] Step 202: Perform missing value processing and outlier processing on the historical meteorological data of the wind turbine to generate the target wind turbine's real-time data.

[0053] In this embodiment of the invention, missing value processing methods include deletion and interpolation, and outlier processing methods include data cleaning and data transformation. By performing missing value processing and outlier processing on the historical meteorological data of the wind turbine in the wind turbine real-time data, preprocessed wind turbine real-time data is generated.

[0054] Step 203: Use bilinear interpolation to interpolate the numerical forecast data to generate initial numerical forecast data.

[0055] In this embodiment of the invention, since the obtained numerical forecast data has a coarse resolution, it is necessary to downscale the forecast data. In order to reduce the amount of data calculation, bilinear interpolation is used to interpolate the meteorological data in the numerical forecast data.

[0056] Step 204: Based on the preset wind speed correction model, the wind speed forecast data in the initial numerical forecast data is corrected using the basic wind turbine data to generate the target numerical forecast data.

[0057] Furthermore, the basic data of the wind turbine includes the radius from the cross section to the blade root, design speed, wind turbine rotation speed, wind turbine rotation radius, blade tip linear velocity, axial induction factor of the blade element, tangential induction factor of the blade element, and blade element angle of the blade cross section. The preset wind speed correction model is as follows:

[0058] In the formula, Indicates corrected wind speed. This indicates wind speed forecast data. This indicates the radius from the cross section to the blade root. Indicates the relative wind speed angle of attack. This represents the ratio of the relative velocity to the wind speed at the location of the cross-section. Indicates the tip speed ratio, Indicates the design speed. Indicates the fan speed. Indicates the fan's rotation radius. Indicates the linear velocity at the blade tip. Indicating the axial inducing factor of leaf pigment, Indicates the tangential inducing factor of leaf pigment. The leaf element angle represents the cross-section of the blade.

[0059] In this embodiment of the invention, it is worth mentioning that the wind speed in the forecast value needs to be corrected due to the influence of the local micro-meteorological environment caused by the rotation of the wind blades. Therefore, a preset wind speed correction model is used to correct the wind speed forecast data in the initial numerical forecast data to generate preprocessed numerical forecast data. This is the first correction of the wind speed, which can reduce data errors and improve the accuracy of the numerical forecast data.

[0060] Step 205: Correct the preprocessed numerical forecast data using the Bayesian model averaging method to generate corrected numerical forecast data.

[0061] Furthermore, the revised numerical forecast data includes revised temperature forecast data, revised precipitation forecast data, and revised wind speed forecast data. Step 205 may include the following sub-steps: It should be noted that Bayesian model averaging (BMA) corrections were applied to the meteorological elements. Considering that the meteorological elements required for icing prediction models are mainly temperature, precipitation, and wind speed, please refer to [the relevant documentation / reference]. Figure 4 The BMA (Browser-Mechanical Analysis) method is used to correct the three factors separately. The BMA prediction model is as follows:

[0062] In the formula, Indicates the variable that needs to be predicted. Indicates the first Among the meteorological parameters, the first one is... The predicted values ​​of each model, Indicates the first Model of meteorological parameters The posterior probability, or weight, of the best predictive model reflects the contribution of each model to the forecasting skill during the training period.

[0063] S11. Based on the pre-set BMA prediction model, the temperature is fitted using a normal distribution to construct a temperature BMA model.

[0064] In this embodiment of the invention, it is assumed that the temperature follows a normal distribution; therefore, the temperature BMA model is as follows:

[0065] In the formula, The probability density function representing temperature. Represents variance. This represents the posterior probability of the optimal temperature prediction model. Indicates the first The predicted values ​​of a temperature prediction model, and All represent model parameters, determined through linear regression: .

[0066] S12. Based on the pre-set BMA prediction model, the precipitation is fitted using the logistic regression model and the gamma regression model to construct the precipitation BMA model.

[0067] Logistic regression, also known as logistic regression or logi regression, is a linear model used for classification problems. It transforms continuous predicted values ​​into probability values ​​by mapping the output of the linear regression model through a sigmoid function, thus achieving classification. Gamma regression, on the other hand, is a regression model based on the Gamma distribution. The Gamma distribution is a continuous probability distribution, often used to describe the right skewness of normally distributed data. Gamma regression models establish the relationship between independent and dependent variables by estimating the parameters of the Gamma distribution.

[0068] In this embodiment of the invention, the logistic regression model ( ), gamma regression model ( The precipitation data is fitted, and the generated parameters are used to determine the precipitation using linear regression. Therefore, the precipitation BMA model is as follows:

[0069] In the formula, Indicates the first The predicted values ​​of a precipitation prediction model, This represents the posterior probability of the optimal precipitation prediction model.

[0070] S13. Based on the pre-set BMA prediction model, the wind speed is fitted using the Weibull distribution to construct the wind speed BMA model.

[0071] The Weibull distribution, also known as the Weibull distribution, is a continuous probability distribution.

[0072] In this embodiment of the invention, the Weibull distribution is used to fit the wind speed; therefore, the wind speed BMA model is as follows:

[0073] In the formula, Indicates the first The predicted values ​​of a wind speed prediction model, This represents the posterior probability of the optimal precipitation prediction model. Indicates the actual wind speed. Indicates shape parameters, The scale parameter is represented by the two parameters. The two parameters can be obtained by directly fitting the historical meteorological data of the wind turbine using the built-in fitdist function of MATLAB.

[0074] S14. Obtain the posterior probabilities associated with the temperature BMA model, precipitation BMA model, and wind speed BMA model respectively using the expectation-maximization algorithm.

[0075] The Expectation-Maximization (EM) algorithm is an iterative optimization method that combines different unsupervised machine learning algorithms to find the maximum likelihood or maximum a posteriori estimate of parameters in a statistical model involving unobserved latent variables. The EM algorithm is commonly used for latent variable models and can handle missing data. It consists of an estimation step and a maximization step, forming an iterative process to improve model fit.

[0076] In the embodiments of the present invention, please refer to Figure 5 First, the parameters are initialized, and the initial log-likelihood function is calculated. In the Experimentation step, the unobserved variables for each group (k, s, t) are calculated, where k refers to the k-th prediction model ("forecast member"), and s and t represent spatial and temporal indices. In the Maximization step, the corresponding parameters and functions are updated, and an algorithm convergence test is performed. The algorithm stops when the decision condition is less than the threshold ε, and the weights (posterior probabilities) and variances are output. The posterior probabilities of the temperature BMA model, precipitation BMA model, and wind speed BMA model, and the variance of the temperature BMA model are obtained respectively using the Expectation-Maximization (EM) algorithm. Parameter estimation here is performed using the conventional Expectation-Maximization (EM) algorithm, which will not be elaborated further.

[0077] S15. The temperature forecast data and associated posterior probability from the preprocessed numerical forecast data are used to input the temperature BMA model for correction, and the corrected temperature forecast data is output.

[0078] In this embodiment of the invention, temperature forecast data and associated posterior probabilities are input into the temperature BMA model for correction, and corrected temperature forecast data is output.

[0079] S16. The precipitation forecast data and associated posterior probability from the preprocessed numerical forecast data are used to input the precipitation BMA model for correction, and the corrected precipitation forecast data is output.

[0080] In this embodiment of the invention, precipitation forecast data and associated posterior probabilities are input into the precipitation BMA model for correction, and corrected precipitation forecast data is output.

[0081] S17. The wind speed forecast data and associated posterior probability from the preprocessed numerical forecast data are used to input the wind speed BMA model for correction, and the corrected wind speed forecast data is output.

[0082] In this embodiment of the invention, wind speed forecast data and associated posterior probabilities are input into the wind speed BMA model for correction, and corrected wind speed forecast data is output.

[0083] The average value of the initially acquired numerical forecast dataset and the forecast values ​​corrected by BMA were compared with historical meteorological monitoring data, and the BMA method was evaluated using MAE and RMSE scores (formulas).

[0084] The following are the scoring formulas for MAE and RMSE respectively:

[0085] In the formula, Indicates the mean absolute error. Indicates the number of predicted values. Indicates the first One predicted value, Indicates the first One measured value, This represents the root mean square error.

[0086] Step 206: Based on the LSSVM algorithm, construct a target icing type prediction model using preprocessed wind turbine real-time data.

[0087] Furthermore, step 206 may include the following sub-steps: S21. Divide the preprocessed wind turbine real-time data into data segments to generate test datasets and training datasets.

[0088] S22. Initialize the model parameters of the preset initial icing type prediction model.

[0089] S23. Based on the particle swarm optimization algorithm, the training dataset is used as input to the preset initial icing type prediction model to optimize the parameters and determine the optimal model parameters.

[0090] S24. Update the preset initial icing type prediction model using the optimal model parameters to generate an intermediate icing type prediction model.

[0091] S25. Use the test dataset to input the intermediate icing type prediction model for testing and generate test indicators.

[0092] S26. When the test indicators meet the preset test conditions, the target icing type prediction model is output.

[0093] The test metrics refer to the relative error (RE) and the root mean square logarithmic error (RMSLE).

[0094] In the embodiments of the present invention, please refer to Figure 6 Within the constraints of preset model parameters, a first data particle swarm corresponding to all model parameters is generated, and the position and velocity of each first data particle in the first data particle swarm are initialized. The fitness of each first data particle is evaluated, and the first data fitness corresponding to each first data particle is output. The position and velocity of each first data particle are optimized using a particle swarm optimization algorithm to generate a second data particle swarm. Based on a preset data model, the fitness of each second data particle is evaluated, and the second data fitness corresponding to each second data particle is output. Each first data fitness is compared with its corresponding second data fitness, and the position corresponding to the minimum value of the data fitness is taken as the historical optimal position of the data individual, and an optimized data particle swarm is constructed. The individual data fitness corresponding to the historical optimal position of each data individual in the optimized data particle swarm is traversed to determine the global optimal position of the data. When the fitness of all individual data meets the preset iteration termination condition, the global optimal position of the data is taken as the optimal parameter of the model and output.

[0095] In this embodiment of the invention, the initial icing type prediction model is preset to be a Least Squares Support Vector Machine (LSSVM) model. The input is the predicted index data such as temperature, precipitation, and relative humidity from the modified numerical forecast data. The SSV is initialized and the POS parameters are optimized. The training dataset and the test dataset are set according to the historical monitoring data ratio of 7:3. Based on the PSO optimization results, the test dataset is imported to verify the optimal LSSVM model. The relative error (RE) and root mean square logarithmic error (RMSLE) are used for testing and evaluation. When the preset test conditions are met, the target icing type prediction model is output.

[0096] Step 207: Input the modified numerical weather prediction data into the target icing type prediction model to predict the icing type and output the predicted icing type.

[0097] In this embodiment of the invention, modified numerical weather prediction data is input into the target icing type prediction model to predict the icing type and output the predicted icing type.

[0098] Step 208: Perform icing prediction based on the icing model associated with the predicted icing type and generate wind turbine icing results.

[0099] Furthermore, the revised numerical forecast data also includes liquid water content data, collision rate, capture rate, freezing rate, and effective icing cross-section data. Step 208 may include the following sub-steps: It should be noted that the types of data included in the corrected numerical weather prediction data are not limited to temperature, precipitation, and wind speed forecasts. They also include data such as liquid water content, water density, and ice density. In the BMA correction step, only the temperature, precipitation, and wind speed forecasts are corrected. The corrected meteorological data is then integrated with the uncorrected liquid water content, water density, and ice density data to obtain the corrected numerical weather prediction data.

[0100] S31. When the predicted icing type is rime ice, then match the associated rime ice icing model.

[0101] In this embodiment of the invention, when the target icing type prediction model outputs a predicted icing type of rime ice, the associated rime ice icing model is matched.

[0102] S32. Input the rime ice model with liquid water content data, preset water density data, preset ice density data, corrected precipitation forecast data and corrected wind speed forecast data, and output the rime ice data corresponding to the wind turbines of the wind farm. It should be noted that the rime ice model uses the Jones simplified model with parameter correction; the rime ice icing model is as follows:

[0103] In the formula, This indicates data on frost and ice accumulation, specifically the cumulative freezing rainfall. This indicates the liquid water content data. This indicates a correction to the precipitation forecast data. Indicates the correction parameter. This indicates the preset water density data. This indicates a correction to the wind speed forecast data. This indicates the preset ice density data.

[0104] In this embodiment of the invention, liquid water content data, preset water density data, preset ice density data, corrected precipitation forecast data, and corrected wind speed forecast data are input into the rime ice model for calculation, and the rime ice data corresponding to the wind turbines of the wind farm is output.

[0105] S33. When the predicted icing type is rime and mixed frost, then match the associated rime and mixed frost icing models.

[0106] In this embodiment of the invention, when the target icing type prediction model outputs the predicted icing type as rime and mixed frost, the associated rime and mixed frost icing models are matched.

[0107] S34. Input the rime and mixed frost icing models with modified wind speed forecast data, liquid water content data, collision rate, capture rate, freezing rate and effective icing cross section data, and output the rime and mixed frost data corresponding to the wind turbines of the wind farm. It should be noted that the Makkonen icing model is used for rime and mixed rime. The specific icing model for rime and mixed rime is as follows:

[0108] In the formula, This represents data on rime and mixed rime, specifically the icing rate. Indicates the collision rate. Indicates capture rate, Indicates the freeze rate. This represents the effective icing cross-section data.

[0109] In this embodiment of the invention, modified wind speed forecast data, liquid water content data, collision rate, capture rate, freezing rate, and effective icing cross section data are input into the rime and mixed frost icing model for calculation, and the rime and mixed frost data corresponding to the wind turbines of the wind farm are output.

[0110] Furthermore, the revised numerical forecast data also includes the convection exchange coefficient, and step 208 may also include the following sub-steps: S35. When the corrected temperature forecast data is greater than the preset melting surface temperature, the corresponding melting model is matched.

[0111] In this embodiment of the invention, the melting conditions are considered simultaneously. That is, when the air temperature is greater than the melting surface temperature, melting occurs. The temperature forecast data is monitored and corrected in real time. When the corrected temperature forecast data is greater than the preset melting surface temperature, the corresponding melting model is matched.

[0112] S36. Input the modified air temperature forecast data, preset melting surface temperature, convection exchange coefficient, preset Stefan Boltzmann constant, preset radiation constant, preset ice specific heat data and preset water freezing latent heat data into the melting model, and output the melting data corresponding to the wind turbines of the wind farm. The specific ice-melting model is as follows:

[0113] In the formula, This represents ice melt data, specifically the ice thickness growth rate. Indicates the convection exchange coefficient. This indicates a correction to the temperature forecast data. This indicates the preset surface temperature for melting ice. This indicates the presupposed Stefan Boltzmann constant. This represents the preset radiation constant. This indicates the preset ice specific heat data. This indicates the preset latent heat of water freezing.

[0114] In this embodiment of the invention, modified air temperature forecast data, preset ice melting surface temperature, convection exchange coefficient, preset Stefan Boltzmann constant, preset radiation constant, preset ice specific heat data and preset water freezing latent heat data are input into the ice melting model for calculation, and the ice melting data corresponding to the wind turbines of the wind farm is output.

[0115] It should be noted that this invention uses a physical model for prediction, which requires distinguishing between rime and hoarfrost types. Although the Makkonen model has wide applicability, the Jones model is simpler and faster to calculate for rime icing. Since traditional icing type diagnosis requires multi-level atmospheric forecast data, but for small-scale wind turbine meteorological monitoring and numerical model forecast data, this data is often lacking or inaccurate, this invention uses an LSSVM-based method to predict the icing type and then calculates the corresponding ice thickness and other icing data. Furthermore, BMA only corrects for factors that significantly affect icing, such as temperature, precipitation, and wind speed; for other variables required in the icing model, the ensemble average of the corresponding variables from the numerical model is used.

[0116] Furthermore, the effectiveness of wind turbine icing prediction is evaluated using the TS score (formula). The higher the TS score, the better the prediction effect.

[0117] The following is the TS scoring formula:

[0118] In the formula, Indicates the forecast success index. Indicates the number of times the forecast was correct. Indicates the number of false alarms. Indicates the number of false alarms.

[0119] In this invention, in response to a request for icing prediction of wind turbines in a wind farm, real-time wind turbine data and numerical weather prediction data are acquired. These data are preprocessed, and the preprocessed numerical weather prediction data is corrected using a Bayesian model averaging method to generate corrected numerical weather prediction data. Based on the LSSVM algorithm, a target icing type prediction model is constructed using the preprocessed real-time wind turbine data. The corrected numerical weather prediction data is then input into the target icing type prediction model to predict the icing type, outputting the predicted icing type. Icing prediction is then performed based on the icing model associated with the predicted icing type, generating the wind turbine icing result. This invention addresses the problem that existing methods for correcting model forecast data mostly focus on correcting single-model data or correcting multiple models individually before using ensemble learning, resulting in cumbersome data processing and a lack of methods for correcting multi-model ensemble forecast data. The present invention addresses the technical problem of insufficient accuracy in icing prediction by addressing several issues. Firstly, it considers not only single forecast values ​​but also ensemble forecasts or single-model forecasts from different start times, increasing the data sources for icing prediction. Furthermore, multi-member (ensemble) forecasts also contribute to increased accuracy. Secondly, it uses Bayesian model averaging (BMA) to correct for temperature, precipitation, and wind speed, improving the availability of numerical forecast data and expanding the prediction methods for traditional icing models on wind turbines. Thirdly, it uses machine learning algorithms to fit historical data to determine icing types, taking into account the specific characteristics of different environments, resulting in strong local applicability and generalizability. Finally, it improves the accuracy of icing prediction by combining an improved Jones simple model that considers melting conditions with the Makkonen icing prediction model.

[0120] Example 3

[0121] Please see Figure 7 , Figure 7 This is a structural block diagram of a wind turbine icing prediction system provided in Embodiment 3 of the present invention.

[0122] The present invention provides a wind turbine icing prediction system, comprising: The response module 301 is used to respond to the icing prediction request for wind turbines in the wind farm and to obtain the actual wind turbine data and numerical forecast data. The data preprocessing module 302 is used to preprocess the actual wind turbine data and numerical forecast data respectively. The data correction module 303 is used to correct the preprocessed numerical forecast data using the Bayesian model averaging method to generate corrected numerical forecast data. The model building module 304 is used to build a target icing type prediction model based on the LSSVM algorithm and the preprocessed actual wind turbine data. The icing type prediction module 305 is used to input the corrected numerical forecast data into the target icing type prediction model to predict the icing type and output the predicted icing type. The icing result module 306 is used to perform icing prediction based on the icing model matched with the predicted icing type and generate the wind turbine icing result.

[0123] Furthermore, the wind turbine real-time data includes historical meteorological data and basic wind turbine data. The data preprocessing module 302 includes: a preprocessed wind turbine real-time data submodule, used to process missing values ​​and outliers in the historical meteorological data of the wind turbine to generate target wind turbine real-time data; an initial numerical forecast data submodule, used to perform interpolation operations on the numerical forecast data using bilinear interpolation to generate initial numerical forecast data; and a preprocessed numerical forecast data submodule, used to correct the wind speed forecast data in the initial numerical forecast data based on a preset wind speed correction model and using the basic wind turbine data to generate target numerical forecast data.

[0124] Furthermore, the preset wind speed correction model is as follows:

[0125] In the formula, Indicates corrected wind speed. This indicates wind speed forecast data. This indicates the radius from the cross section to the blade root. Indicates the relative wind speed angle of attack. This represents the ratio of the relative velocity to the wind speed at the location of the cross-section. Indicates the tip speed ratio, Indicates the design speed. Indicates the fan speed. Indicates the fan's rotation radius. Indicates the linear velocity at the blade tip. Indicating the axial inducing factor of leaf pigment, Indicates the tangential inducing factor of leaf pigment. The leaf element angle represents the cross-section of the blade.

[0126] Furthermore, the data correction module 303 includes: a temperature BMA model submodule, used to construct a temperature BMA model by fitting a normal distribution to the temperature based on a pre-set BMA prediction model; a precipitation BMA model submodule, used to construct a precipitation BMA model by fitting a logistic regression model and a gamma regression model to the precipitation based on a pre-set BMA prediction model; a wind speed BMA model submodule, used to construct a wind speed BMA model by fitting a Weibull distribution to the wind based on a pre-set BMA prediction model; and a posterior probability submodule, used to obtain the temperature BMA model and the precipitation BMA model respectively using the expectation-maximization algorithm. The system includes: a posterior probability associated with the wind speed BMA model; a revised temperature forecast data submodule, used to input the temperature forecast data and associated posterior probability from the preprocessed numerical weather prediction data into the temperature BMA model for correction, and output revised temperature forecast data; a revised precipitation forecast data submodule, used to input the precipitation forecast data and associated posterior probability from the preprocessed numerical weather prediction data into the precipitation BMA model for correction, and output revised precipitation forecast data; and a revised wind speed forecast data submodule, used to input the wind speed forecast data and associated posterior probability from the preprocessed numerical weather prediction data into the wind speed BMA model for correction, and output revised wind speed forecast data.

[0127] Furthermore, the model building module 304 includes: a data partitioning submodule, used to partition the preprocessed wind turbine real-time data to generate a test dataset and a training dataset; a model initialization submodule, used to initialize the model parameters of the preset initial icing type prediction model; a parameter optimization submodule, used to optimize the parameters of the preset initial icing type prediction model based on the particle swarm optimization algorithm, using the training dataset as input, to determine the optimal model parameters; an update submodule, used to update the preset initial icing type prediction model using the optimal model parameters, generating an intermediate icing type prediction model; a testing submodule, used to test the intermediate icing type prediction model using the test dataset as input, generating test indicators; and a target icing type prediction model submodule, used to output the target icing type prediction model when the test indicators meet preset test conditions.

[0128] Furthermore, the revised numerical forecast data also includes liquid water content data, collision rate, capture rate, freezing rate, and effective icing cross section data. The icing result module 306 includes: a rime icing model submodule, used to match the associated rime icing model when the predicted icing type is rime; and a rime icing data submodule, used to input liquid water content data, preset water density data, preset ice density data, revised precipitation forecast data, and revised wind speed forecast data into the rime icing model, and output rime icing data corresponding to the wind turbines of the wind farm. The specific model of rime ice accumulation is as follows:

[0129] In the formula, This indicates data on frost and ice accumulation, specifically the cumulative freezing rainfall. This indicates the liquid water content data. This indicates a correction to the precipitation forecast data. Indicates the correction parameter. This indicates the preset water density data. This indicates a correction to the wind speed forecast data. This indicates the preset ice density data; The rime and mixed frost icing model submodule is used to match the associated rime and mixed frost icing models when the predicted icing type is rime and mixed frost. The rime and mixed frost data submodule is used to input the rime and mixed frost icing models with corrected wind speed forecast data, liquid water content data, collision rate, capture rate, freezing rate and effective icing cross section data, and output the rime and mixed frost data corresponding to the wind turbines of the wind farm. The specific models for rime ice and mixed rime ice icing are as follows:

[0130] In the formula, This represents data on rime and mixed rime, specifically the icing rate. Indicates the collision rate. Indicates capture rate, Indicates the freeze rate. This represents the effective icing cross-section data.

[0131] Furthermore, the corrected numerical forecast data also includes the convection exchange coefficient. The icing result module 306 also includes: an ice melting model submodule, used to match the corresponding ice melting model when the corrected temperature forecast data is greater than the preset ice melting surface temperature; and an ice melting data submodule, used to input the corrected temperature forecast data, preset ice melting surface temperature, convection exchange coefficient, preset Stefan-Boltzmann constant, preset radiation constant, preset ice specific heat data and preset water freezing latent heat data into the ice melting model, and output the ice melting data corresponding to the wind turbines of the wind farm. The specific ice-melting model is as follows:

[0132] In the formula, This represents ice melt data, specifically the ice thickness growth rate. Indicates the convection exchange coefficient. This indicates a correction to the temperature forecast data. This indicates the preset surface temperature for melting ice. This indicates the presupposed Stefan Boltzmann constant. This represents the preset radiation constant. This indicates the preset ice specific heat data. This indicates the preset latent heat of water freezing.

[0133] Please see Figure 8 , Figure 8 A structural block diagram of a computer device according to an embodiment of the present invention is shown.

[0134] An electronic device according to an embodiment of the present invention includes: a memory 401 and a processor 402. The memory 401 stores a computer program. When the computer program is executed by the processor 402, the processor 402 executes the wind turbine icing prediction method as described in any of the above embodiments.

[0135] Memory 401 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Memory 401 has storage space 403 for program code 413 for performing any of the method steps described above. For example, storage space 403 for program code may include individual program codes 413 for implementing the various steps in the methods described above. This program code may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When run by a computing processing device, this code causes the computing processing device to perform the various steps in the methods described above. This program code may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When this code is run by a computing device, it causes the computing device to perform the various steps in the wind turbine icing prediction method described above.

[0136] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the wind turbine icing prediction method as described in any of the above embodiments.

[0137] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0139] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0140] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0141] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0142] 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 do 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 wind turbine icing, characterized in that, include: In response to a request for icing prediction of wind turbines in a wind farm, the system acquires real-time data and numerical forecast data of the wind turbines in the wind farm. The actual wind turbine data and the numerical forecast data are preprocessed respectively; The preprocessed numerical forecast data is corrected using the Bayesian model averaging method to generate corrected numerical forecast data. Based on the LSSVM algorithm, a target icing type prediction model is constructed using the preprocessed wind turbine real-time data; The modified numerical forecast data is used as input to the target icing type prediction model to predict the icing type, and the predicted icing type is output. Icing prediction is performed based on the icing model associated with the predicted icing type, and icing results for the wind turbine are generated. The wind turbine real-time data includes historical meteorological data and basic wind turbine data. The steps of preprocessing the wind turbine real-time data and the numerical forecast data respectively include: The historical meteorological data of the wind turbines are processed for missing values ​​and outliers respectively to generate the target wind turbine's real-time data; The numerical forecast data is interpolated using bilinear interpolation to generate initial numerical forecast data. Based on the preset wind speed correction model, the wind speed forecast data in the initial numerical forecast data is corrected using the basic wind turbine data to generate target numerical forecast data. The preset wind speed correction model is specifically as follows: In the formula, Indicates corrected wind speed. This refers to the wind speed forecast data. This indicates the radius from the cross section to the blade root. Indicates the relative wind speed angle of attack. This represents the ratio of the relative velocity to the wind speed at the location of the cross-section. Indicates the tip speed ratio. Indicates the design speed. Indicates the fan speed. Indicates the fan's rotation radius. Indicates the linear velocity at the blade tip. Indicating the axial inducing factor of leaf pigment, Indicates the tangential inducing factor of leaf pigment. The leaf element angle represents the cross-section of the blade.

2. The wind turbine icing prediction method according to claim 1, characterized in that, The step of correcting the preprocessed numerical forecast data using the Bayesian model averaging method to generate corrected numerical forecast data includes: Based on the pre-set BMA prediction model, a normal distribution is used to fit the temperature to construct a temperature BMA model; Based on the pre-set BMA prediction model, the precipitation is fitted using the logistic regression model and the gamma regression model to construct the precipitation BMA model. Based on the pre-set BMA prediction model, the wind speed is fitted using the Weibull distribution to construct a wind speed BMA model. The posterior probabilities associated with the temperature BMA model, the precipitation BMA model, and the wind speed BMA model are obtained respectively using the expectation-maximization algorithm. The temperature forecast data and the associated posterior probability in the preprocessed numerical forecast data are used as inputs to the temperature BMA model for correction, and the corrected temperature forecast data is output. The precipitation forecast data and the associated posterior probability in the preprocessed numerical forecast data are used as inputs to the precipitation BMA model for correction, and the corrected precipitation forecast data is output. The wind speed forecast data and the associated posterior probability in the preprocessed numerical forecast data are used to input the wind speed BMA model for correction, and the corrected wind speed forecast data is output.

3. The wind turbine icing prediction method according to claim 1, characterized in that, The step of constructing a target icing type prediction model based on the LSSVM algorithm using the preprocessed wind turbine real-time data includes: The preprocessed wind turbine real-time data is divided into test datasets and training datasets; Initialize the model parameters of the preset initial icing type prediction model; Based on the particle swarm optimization algorithm, the training dataset is used as input to the preset initial icing type prediction model to optimize the parameters and determine the optimal model parameters. The preset initial icing type prediction model is updated using the optimal model parameters to generate an intermediate icing type prediction model; The intermediate icing type prediction model is tested using the test dataset, and test indicators are generated. When the test indicators meet the preset test conditions, the target icing type prediction model is output.

4. The wind turbine icing prediction method according to claim 2, characterized in that, The corrected numerical forecast data also includes liquid water content data, collision rate, capture rate, freezing rate, and effective icing cross-section data. The step of generating wind turbine icing results by performing icing prediction based on the icing model matched with the predicted icing type includes: When the predicted icing type is rime ice, then the associated rime ice icing model is matched; The model is input with the liquid water content data, preset water density data, preset ice density data, the corrected precipitation forecast data, and the corrected wind speed forecast data, and outputs the rime icing data corresponding to the wind turbines of the wind farm. The specific model for ice formation and icing is as follows: In the formula, The data on frost and ice accumulation refers specifically to the cumulative amount of freezing rain. This indicates the liquid water content data. This refers to the revised precipitation forecast data. Indicates the correction parameter. This represents the preset water density data. This refers to the corrected wind speed forecast data. This represents the preset ice density data; When the predicted icing type is rime and mixed rime, then the associated rime and mixed rime icing models are matched. The corrected wind speed forecast data, liquid water content data, collision rate, capture rate, freezing rate, and effective icing cross section data are used as inputs to the rime and mixed frost icing model, and the rime and mixed frost data corresponding to the wind turbines of the wind farm are output. The specific models for rime ice and mixed rime ice icing are as follows: In the formula, This refers to the data on rime and mixed rime, specifically the icing rate. This represents the collision rate. Indicates the capture rate, This indicates the freezing rate. This represents the effective icing cross-section data.

5. The wind turbine icing prediction method according to claim 4, characterized in that, The revised numerical forecast data also includes convection exchange coefficients, and further includes: When the corrected temperature forecast data is greater than the preset melting surface temperature, the corresponding melting model is matched; The modified air temperature forecast data, the preset ice melting surface temperature, the convection exchange coefficient, the preset Stefan-Boltzmann constant, the preset radiation constant, the preset ice specific heat data, and the preset water freezing latent heat data are used as inputs to the ice melting model, and the ice melting data corresponding to the wind turbines of the wind farm are output. The ice-melting model is specifically as follows: In the formula, The aforementioned ice melt data specifically refers to the ice thickness growth rate. This represents the convection exchange coefficient. This refers to the revised temperature forecast data. This indicates the preset ice-melting surface temperature. This represents the preset Stefan-Boltzmann constant. This represents the preset radiation constant. This represents the preset ice specific heat data. This indicates the preset latent heat of water freezing.

6. A wind turbine icing prediction system, characterized in that, The wind turbine icing prediction system is used to implement the wind turbine icing prediction method as described in any one of claims 1-5, and the wind turbine icing prediction system includes: The response module is used to respond to the icing prediction request of the wind turbines in the wind farm and to obtain the actual wind turbine data and numerical forecast data of the wind turbines in the wind farm. The data preprocessing module is used to preprocess the actual wind turbine data and the numerical forecast data respectively; The data correction module is used to correct the preprocessed numerical forecast data using the Bayesian model averaging method to generate corrected numerical forecast data. The model building module is used to build a target icing type prediction model based on the LSSVM algorithm and the preprocessed wind turbine real-time data. The icing type prediction module is used to input the modified numerical forecast data into the target icing type prediction model to predict the icing type and output the predicted icing type. The icing result module is used to predict icing based on the predicted icing type and the associated icing model, and generate wind turbine icing results.

7. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the wind turbine icing prediction method as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the wind turbine icing prediction method as described in any one of claims 1-5.

Images