Short-term wind power prediction method and system for fan icing scene

Abstract

The invention belongs to the technical field of wind power prediction, and particularly discloses a short-term wind power prediction method and system for a fan icing scene, and the method comprises the steps: constructing an icing model suitable for a fan blade, and solving key parameters in the icing model under a set meteorological condition in an auxiliary manner through icing simulation; acquiring meteorological data and icing data under a fan icing scene in a set time range of a to-be-measured area, inputting the meteorological data into the trained basic prediction model to obtain a wind power basic prediction result, and inputting the icing data and wind speed data into the trained power loss prediction model to obtain a power loss prediction result; and obtaining a final short-term wind power prediction result for the fan icing scene based on the wind power basic prediction result and the power loss prediction result. According to the method, the calculation accuracy of the icing data can be improved in the earlier stage of saving calculation resources, so that the reliability of a power prediction result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and in particular to a short-term wind power prediction method and system for wind turbine icing scenarios. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Wind energy is a clean, high-quality renewable energy source that plays an irreplaceable role in economic development, energy supply and reducing greenhouse gas emissions. The healthy development of wind power is of great significance to promoting the adjustment of energy structure.

[0004] Accurately predicting wind power is an effective measure to deal with its randomness and volatility, and is of great significance for supporting grid dispatching, improving the stability of power system operation, and increasing the consumption of new energy. However, the current meteorological environment is becoming increasingly complex, and extreme weather such as freezing rain and cold waves occur frequently, which can easily cause wind turbine blades to be covered with ice and affect the output fluctuation characteristics. In addition, affected by factors such as latitude or altitude, wind turbine blades are also often covered with ice at low temperatures, affecting wind turbine power output and the normal operation of wind farms.

[0005] In the above scenarios, traditional wind power prediction models are often built based on numerical weather forecasts and historical operating data, and do not take into account the impact of icing scenarios. Therefore, large prediction deviations may occur and threaten the safe and reliable operation of the power system. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a short-term wind power prediction method and system for wind turbine icing scenarios. By quantifying the icing data and considering the wind power data under the wind turbine blade icing scenario, the accuracy of wind power prediction can be improved, and the ability of the new power system to cope with wind power volatility can be improved.

[0007] In some embodiments, the following technical solutions are adopted: A short-term wind power prediction method for wind turbine icing scenarios, comprising: Considering the melting and shedding effects of ice, an icing model suitable for wind turbine blades is constructed, and the key parameters of the icing model under set meteorological conditions are solved through icing simulation. The discrete ice cover data are converted into continuous ice cover data corresponding to meteorological conditions at different times through data interpolation; Train the basic prediction model for wind power prediction and the power loss prediction model considering wind turbine icing scenarios respectively; Obtain meteorological data and icing data under the wind turbine icing scenario within a set time range in the test area, input the meteorological data into a trained basic prediction model to obtain a basic prediction result of wind power, and input the icing data and wind speed data into a trained power loss prediction model to obtain a power loss prediction result; Based on the basic wind power prediction results and power loss prediction results, the final short-term wind power prediction results for wind turbine icing scenarios are obtained.

[0008] Furthermore, considering the melting and shedding effects of ice, an ice model suitable for wind turbine blades is constructed, specifically: ; ; ; in, To consider the ice melting and shedding effect, the accumulated ice coverage of a single wind turbine is When the effects of ice melting and shedding are not considered i The increase in the maximum ice thickness at the moment, To quantify the proportionality coefficient of ice reduction between ice ablation and shedding effect, It means that after considering the effects of ice melting and shedding i The increase in the maximum ice thickness at the moment, express i The maximum ice coverage length of the airfoil at that moment; c represents the blade chord length, and n represents the set time; , are the collision coefficient and the freezing coefficient, is the mass fraction of water droplets in the air, is the velocity of the water drop, is the unit time, is the density of ice accumulation, This is the proportionality factor when solving for the volume of ice accumulation.

[0009] Furthermore, the mass fraction of water droplets in the air Specifically: ; Where Q and d represent the cloud water mixing ratio and air density, respectively.

[0010] Furthermore, the key parameters of the icing model under set meteorological conditions are solved through icing simulation, specifically: Construct a physical model of the fan blades, use Fluent software to calculate the air flow field around the fan blades, import the flow field calculation results into the icing simulation software, preset the median diameter of the water droplets and the liquid water content in the air, and calculate the collision coefficient through the water droplet motion trajectory simulation and icing simulation. , freezing coefficient , solve for the proportionality coefficient when solving for the ice accumulation volume and the proportionality factor of ice reduction .

[0011] Furthermore, the discrete ice cover data is converted into continuous ice cover data corresponding to meteorological conditions at different times through data interpolation, specifically: The values ​​of icing parameters α1 and α3 under any meteorological data are estimated by Kriging interpolation method, and then the icing data under meteorological data at different times are calculated.

[0012] Furthermore, the basic prediction model for wind power prediction and the power loss prediction model considering wind turbine icing scenario are trained separately, specifically: Constructing a dataset D1 containing conventional meteorological data and corresponding wind power data, and a dataset D2 and a dataset D3 containing meteorological data and corresponding wind power data under icing scenarios respectively; Use the data set D1 to train the basic prediction model to obtain a trained basic prediction model; The wind power of the data set D2 is predicted using the trained basic prediction model, and the difference between the predicted wind power value and the actual value is recorded as power loss; the power loss and the corresponding wind speed and icing data constitute the power loss data set D4; The power loss prediction model is trained using the data set D4 to obtain a trained power loss prediction model; The trained basic prediction model and power loss prediction model are tested using the dataset D3.

[0013] Furthermore, the basic prediction model uses the LightGBM model, and the power loss prediction model uses the Few-shot model.

[0014] In other embodiments, the following technical solutions are adopted: A short-term wind power prediction system for wind turbine icing scenarios, comprising: The icing model simulation and solution module is used to consider the melting and shedding effects of ice, build an icing model suitable for wind turbine blades, and solve the key parameters of the icing model under set meteorological conditions through icing simulation; An ice cover data interpolation module is used to convert discrete ice cover data into continuous ice cover data corresponding to meteorological conditions at different times through data interpolation; A model training module is used to train the basic prediction model for wind power prediction and the power loss prediction model considering wind turbine icing scenarios; The power prediction module is used to obtain meteorological data and icing data under the wind turbine icing scenario within a set time range in the test area. The meteorological data is input into a trained basic prediction model to obtain a basic wind power prediction result. The icing data and wind speed data are input into a trained power loss prediction model to obtain a power loss prediction result. Based on the basic wind power prediction result and the power loss prediction result, the final short-term wind power prediction result for the wind turbine icing scenario is obtained.

[0015] In other embodiments, the following technical solutions are adopted: A terminal device comprises a processor and a memory, wherein the processor is used to implement instructions; the memory is used to store a plurality of instructions, wherein the instructions are suitable for being loaded by the processor and executing the above-mentioned short-term wind power prediction method for wind turbine icing scenarios.

[0016] In other embodiments, the following technical solutions are adopted: A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the above-mentioned short-term wind power prediction method for wind turbine icing scenarios.

[0017] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention fully considers the processes of ice melting and shedding, constructs a calculation formula for the degree of ice coverage on wind turbine blades, and assists in obtaining key parameters of ice coverage through ice coverage simulation. The ice coverage parameters are expanded using Kriging interpolation, and a mapping relationship between meteorological data and ice coverage parameters is established, thereby realizing the acquisition of ice coverage data under meteorological conditions at any time. The calculation accuracy of ice coverage data can be improved while saving computing resources in the early stage, thereby improving the reliability of power prediction results.

[0018] (2) The present invention constructs a basic wind power prediction model and a power loss prediction model that take into account the impact of icing, respectively. The power loss of wind power output under icing scenarios is calculated based on the icing degree data, and the conventional wind power prediction power is corrected to improve the prediction accuracy, thereby improving the ability of the new energy system to cope with wind power uncertainty under extreme icing scenarios and ensuring stable operation of the system.

[0019] Other features and advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1This is a flow chart of a short-term wind power prediction method for wind turbine icing scenarios in an embodiment of the present invention; Figure 2 A schematic diagram of the definition of ice coverage in an embodiment of the present invention; Figure 3 This is a flow chart of the simulation of ice coating on wind turbine blades in an embodiment of the present invention; Figure 4 It is a collision coefficient simulation result diagram under different meteorological conditions in an embodiment of the present invention; Figure 5 It is a freezing coefficient simulation result diagram under different meteorological conditions in an embodiment of the present invention; Figure 6 Schematic diagram of ice coverage degree change in an ice coverage event in an embodiment of the present invention; Figure 7 This is the wind power prediction result of icing event A in the embodiment of the present invention; Figure 8 This is the wind power prediction result of icing event B in the embodiment of the present invention; Fig. 9 This is the wind power prediction result of icing event C in the embodiment of the present invention. DETAILED DESCRIPTION

[0021] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0022] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0023] Embodiment 1 In one or more embodiments, a short-term wind power forecasting method for wind turbine icing scenarios is disclosed, combining Figure 1 , specifically including the following process: S101: Considering the melting and shedding effects of ice, an icing model suitable for wind turbine blades is constructed, and the key parameters of the icing model under set meteorological conditions are solved by icing simulation.

[0024] In this embodiment, the ratio of the maximum ice thickness of the wind turbine airfoil to the chord length is first defined as the degree of ice coverage of the wind turbine blades, and a specific expression form suitable for ice coverage of the wind turbine blades is derived based on the Makkonen ice coverage model.

[0025] In a low temperature and humid environment, the air will contain more supercooled water droplets. Supercooled water droplets refer to water droplets with a temperature below 0°C but not yet frozen. Supercooled water droplets move with the wind. When they collide with the fan blades, some of the supercooled water droplets are captured by the blades. The captured water droplets release heat and freeze, thereby forming an ice layer on the surface of the fan blades. The icing of fan blades is affected by meteorological conditions. The meteorological factors that have a greater impact include temperature, air humidity (in this embodiment, it is reflected in the median diameter MVD of the supercooled water droplet particles and the liquid water content LWC in the air) and wind speed and direction. Existing literature has concluded that the necessary conditions for blades to be covered with ice are: the ambient temperature is below 0°C and the air humidity is above 85%.

[0026] The adverse effects of ice on wind turbine blades on wind power prediction are mainly reflected in the following aspects: ① Ice changes the shape of the blades, affecting their aerodynamic performance, and the output power of the wind turbine decreases accordingly. Therefore, the actual power under ice condition is lower than the power predicted by the conventional prediction model; ② Ice increases the load on the wind turbine blades, but due to the irregularity of ice, the increased load is unbalanced, resulting in changes in the amplitude and frequency of the blades, increasing the risk of wind turbine operation. In severe cases, it causes the wind turbine to shut down, further expanding the prediction error.

[0027] The prior art proposes an icing model for the surface of an object, which decomposes the icing process into three stages: supercooled water droplets in the air collide with the object, supercooled water droplets are captured by the surface of the object, and liquid water droplets freeze. The Makkonen icing model is as follows: (1) In the formula, α1, α2, and α3 are coefficients that characterize the behavior of supercooled water droplets in the three stages of the icing model. Specifically, the collision coefficient α1 is the ratio of the amount of water droplets that actually collide with the fan blade to the amount of all water droplets that can collide. This coefficient is related to the wind speed and the liquid water content; the capture rate α2: water droplets that collide with the fan blade may bounce or splash off the blade surface, so it is necessary to introduce a parameter to characterize the amount of water droplets remaining on the blade surface, namely the capture rate, which represents the ratio of the amount of water droplets remaining on the blade surface to the amount of all water droplets that collide. This coefficient is related to the material of the object; the freezing coefficient α3 is the ratio of the amount of water droplets frozen on the fan blade surface to the amount of water droplets captured. This coefficient is directly affected by temperature and is related to the liquid water content. is the mass fraction of water droplets in the air, that is, the liquid water content LWC in the air; is the velocity of the water drop, and A is the cross-sectional area of ​​the object that collides.

[0028] The Makkonen icing model can be applied to wire icing scenarios and aircraft wing icing scenarios, but the calculation formula for wind turbine blade icing scenarios is still blank.

[0029] This embodiment will start from the Makkonen model and derive its specific expression in the scenario of wind turbine blade icing.

[0030] Combined with Numerical Weather Prediction (NWP) data =15min resolution, this time is defined as a unit time, and it is approximately assumed that the values ​​of meteorological factors and model parameters remain unchanged within a unit time, then In the area of ​​, the increased ice mass per unit time is: (2) During the icing process, supercooled water droplets generally do not rebound, so α2 is usually considered to be equal to 1. The numerical solution method for α1 and α3 is a complex process. This embodiment uses simulation to assist in obtaining the values ​​of the above parameters under different meteorological conditions. In the application scenario of this embodiment, v is approximately equal to the wind speed. Since the liquid water content in the air is It is not within the forecast range of NWP and needs to be calculated and solved with the help of the output physical quantities of the WRF forecast model. The relevant formula is as follows: (3) Where Q and d represent the cloud water mixing ratio and air density, respectively, both of which can be obtained from WRF output.

[0031] like Figure 2 As shown in the figure, the ratio of the maximum ice thickness of the wind turbine airfoil to the chord length is defined as the ice degree, that is: (4) Where I represents the degree of blade icing, represents the maximum ice coverage length of the airfoil, and c represents the blade chord length.

[0032] Existing research shows that the most serious ice coverage occurs at the windward point of the blade tip. When processing, it can be approximately assumed that the ice thickness is greatest at 95% of the blade span, and only the ice coverage at this position is analyzed.

[0033] Assuming that in its vicinity If the ice thickness in the area is approximately equal, then the following relationship holds: (5) In the formula, is the density of ice accumulation, for = the increase in ice thickness at the studied location within 15 minutes, To solve the proportional coefficient of ice accumulation volume, it is determined by the shape of ice accumulation. Due to the irregular shape of ice accumulation, this embodiment The value of is obtained from the statistical experience of a large number of ice accumulation simulations. Combining equations (2) and (5), the expression for the increase in ice thickness per unit time can be obtained as: (6) The accumulated ice coverage is: (7) in, is the increase in ice thickness at the maximum ice position per unit time, and n is the total time.

[0034] The above formula can be used to obtain the moment-by-moment numerical results of the ice coverage of a single wind turbine. In this embodiment, the following simplification is made: it is assumed that in a wind farm, each wind turbine faces the same meteorological conditions. Therefore, the calculation result of the ice coverage of a single wind turbine can be used as the average ice coverage of all wind turbines in the wind farm.

[0035] It can be seen from the above analysis that the key to calculating the degree of icing is to obtain the key parameters α1 and α3. This embodiment uses an icing simulation method to assist in solving the key icing parameters.

[0036] The ice coating process is a complex physical process that requires consideration of fluid movement and heat exchange. Two stages are considered during the simulation: (1) Supercooled water droplets move under the action of wind, hit the surface of fan blades and are captured, exchanging heat with the blade surface and surrounding air.

[0037] (2) Some of the captured water droplets are frozen on the blade surface to form ice, while the unfrozen water droplets move along the flow field on the blade surface or evaporate. As an alternative to experimental methods, modeling and simulation analysis can provide a data basis for ice analysis by reproducing the ice accumulation process on wind turbine blades.

[0038] FENSAP-ICE is a commonly used icing simulation software based on the Navier-Stokes (RANS) equations. It can obtain relatively accurate results. However, since it is more suitable for analyzing the final icing results and ignoring the icing process, this embodiment uses this software for icing simulation only to implement the icing parameters in the Makkonen icing model. α 1 and α 3, the main factors affecting icing, such as temperature, wind speed, liquid water content and median diameter of water droplets, were studied to determine the effects of icing parameters and the calculation results of icing parameters under different combinations of meteorological elements. Figure 3 Follow the steps in the simulation to simulate ice coverage.

[0039] In order to simulate the icing process, it is necessary to first calculate the air flow field around the fan blades, and then calculate the motion trajectory of the water droplets in the flow field to determine the amount of supercooled water droplets collected on the blade surface. Since the flow field calculation module of FENSAP-ICE itself has a relatively simple function and a slow calculation speed, the Fluent software is first used to calculate the flow field. After the calculation is completed, the results are imported into FENSAP-ICE for subsequent calculation and analysis of the motion trajectory of water droplets and icing. When simulating icing, considering that the liquid water droplets in the air are not distributed in a single particle size in reality, the median diameter MVD of the water droplets and the liquid water content LWC in the air are set in advance, and the Custom distribution model provided in FENSAP-ICE is used to customize the median diameter of the liquid water droplets and its proportion so that they obey the normal distribution. After setting the values ​​of other remaining meteorological factors and the icing time, run the software to obtain the corresponding icing parameter simulation results. It should be noted that the airfoils of different wind turbines may be different. When performing icing simulation, the airfoil should be selected according to the specific wind turbine. The target wind farm wind turbine model selected in this embodiment is GW115 / 2.0MW, and the airfoil used in the icing simulation is NACA4412.

[0040] In addition, a complete ice covering process should also consider the melting and shedding of ice. However, the Makkonen model does not consider the above processes. The ice shedding rate is proportional to the amount of ice already accumulated. Take it as a fixed value.

[0041] Furthermore, when considering the melting and shedding effects of ice, the increase in ice cover per unit time can be expressed by the following formula: (8) Accordingly, when considering the melting and shedding effects of ice, the formula for calculating the ice coverage at each moment is: (9) Using ice simulation to assist in solving and , a series of simulation results of ice thickness are obtained, an overdetermined set of equations is formed, and the least squares method is used to find the equation that minimizes the sum of square errors of all equations. and Approximate value of .

[0042] S102: Converting discrete ice cover data into continuous ice cover data corresponding to meteorological conditions at different times through data interpolation.

[0043] As mentioned above, by setting the specific values ​​of the icing meteorology, the FENSAP-ICE software can be used to obtain the icing model parameters under the corresponding meteorological environment, and then calculate the icing thickness and icing degree. However, only the icing results under discrete conditions can be obtained, while the actual meteorological factors are continuously changing. When predicting, it takes a lot of time and computing resources to obtain the icing results under different predicted meteorological conditions through simulation, and the real-time and ease of use are difficult to meet the prediction requirements. Therefore, this embodiment uses the Kriging interpolation algorithm to realize the transformation of discrete results into continuous space, and further combines the meteorological data to calculate the time series of icing degree at different times in the icing event.

[0044] The Kriging interpolation algorithm is a spatial interpolation method based on the theory of variogram, which is used to achieve the optimal error estimation of variables in a limited area. This interpolation method is not only based on the spatial position of the interpolation point and the adjacent sampling points, but also takes into account the mutual position relationship of these sampling points, making full use of the distribution characteristics of the existing observations to achieve a more accurate and more realistic estimation than the traditional interpolation method. Extending the concept of "space" to "numerical space", Kriging interpolation can be used to achieve interpolation estimation of high-dimensional data.

[0045] The Kriging interpolation regression model formula is as follows: (10) In the formula, x0 is the point to be interpolated, Z * (x0) is the interpolation estimation result, x i (i =1,2,……,n) are known sampling points in the study area, Z(x i ) is its sampling value, λ i is the weight coefficient to be determined. This formula shows that the Kriging interpolation result is a weighted combination of the sampling values ​​of other known sampling points in the study area.

[0046] The weight coefficient can be calculated by the following formula: (11) In the formula, C ( x i ,x j ) is the sampling point x i 、x j is the covariance function of the sampled values ​​at , and μ is the Lagrange multiplier introduced when seeking the minimum value of the estimated variance.

[0047] In this embodiment, a high-dimensional numerical space is formed by meteorological factors such as wind speed, temperature, and air humidity, and Kriging interpolation is used to estimate the values ​​of icing parameters α1 and α3 in this data space.

[0048] After the above-mentioned Kriging interpolation, the icing parameters α1 and α3 under any meteorological values ​​can be obtained, and then the corresponding icing data can be calculated according to the meteorological data at different times, providing data support for the subsequent wind power prediction under wind turbine icing scenarios.

[0049] S103: respectively training a basic prediction model for wind power prediction and a power loss prediction model considering wind turbine icing scenarios.

[0050] In this embodiment, historical real data are collected and three data sets, namely training set D1, validation set D2 and prediction set D3, are constructed respectively, wherein the training set D1 contains conventional meteorological data and wind power data; the validation set D2 and prediction set D3 both contain wind power data under icing scenarios and their corresponding meteorological data.

[0051] Basic prediction model training: With meteorological data as input and wind power data as output, the LightGBM model is trained on the dataset D1 as the basic prediction model for wind power prediction.

[0052] The trained LightGBM model is used to predict the wind power of the validation set D2. The difference between the predicted power and the actual power is recorded as power loss, which together with the corresponding wind speed data and icing data constitutes the power loss sample set D4.

[0053] With icing data and wind speed as input and power loss as output, the Few-shot model is trained on the power loss sample set D4 to construct a mapping relationship between icing degree, wind speed and power loss; and a trained power loss prediction model is obtained.

[0054] First, the trained LightGBM model is used to predict the wind power in the prediction set D3 to obtain the basic wind power prediction result; at the same time, the trained Few-shot model is used to predict the power loss of the prediction set D3; finally, the basic wind power prediction result is subtracted from the power loss prediction result to obtain the final wind power prediction result considering the impact of icing.

[0055] After the above training and prediction process, the trained basic prediction model and power loss prediction model are obtained.

[0056] In this embodiment, LightGBM (Light Gradient Boosting Machine) is an efficient ensemble learning algorithm based on the gradient boosting framework. It is an improved version of XGBoost, and its design goal is to improve training speed and efficiency while maintaining high accuracy. The principle of LightGBM in regression tasks is mainly based on the gradient boosting decision tree (GBDT) algorithm, which approximates the minimization process of the objective function by gradually adding decision trees. Each newly added tree aims to reduce the residual left by the previous tree, thereby improving the accuracy of the model.

[0057] The total loss of LightGBM consists of two parts: loss function and regularization term. Assume that the training dataset is ,in, y i refers to wind power, x i Refers to meteorological factors related to wind power (such as wind speed, temperature, humidity, etc.); the model's prediction value in round t is .

[0058] The total loss can be expressed as: (12) (13) In the formula, loss is the loss function, which can be squared loss , It is the part that performs regularization on each tree, mainly used to prevent overfitting and ensure the generalization ability of the model. L is the number of leaf nodes in the tree, w j is the predicted value of the jth leaf node, and are regularization parameters that control the complexity of the tree and the sum of squared leaf weights, respectively.

[0059] In each iteration, the model adds a new tree f t To reduce the residual error. The new tree is fitted according to the negative gradient, which is similar to updating the parameters in the gradient descent method: (14) In the formula, is the learning rate, is the new tree for sample x i The predicted value of .

[0060] Tree t The goal is to fit the residual between the previous round of model prediction and the true value. For the squared loss function, the negative gradient is: (15) Use negative gradient r it As the target value, f is learned on the training data through the decision tree algorithm. t The construction of each tree focuses on maximizing the gain after data splitting, while taking into account the regularization term to avoid overfitting. The above process is repeated until the maximum number of trees is reached.

[0061] In this way, each tree tries to correct the error left by the previous tree, and the accuracy of the entire model gradually increases with the number of iterations. The optimization and efficient implementation of LightGBM make this process run quickly even on large data sets.

[0062] In this embodiment, the Few-shot learning algorithm is implemented based on meta-learning, which is an important branch of machine learning. It aims to improve the generalization ability and adaptability of the algorithm by learning how to learn. The core idea is to use existing task experience to improve the learning efficiency and effect of new tasks. The goal of meta-learning is to enable the model to quickly adapt to new tasks and reduce dependence on large-scale data and long-term training. Due to the small amount of ice-covered data, the error data set samples constructed are small, and it is difficult for traditional artificial intelligence algorithms to play their advantages. Therefore, it is considered to use meta-learning to construct a few-sample learning model to achieve more accurate error prediction based on a small amount of ice-covered samples.

[0063] Meta-learning can be divided into model-based methods and optimization-based methods. Model-Agnostic Meta-Learning (MAML) is an optimization-based meta-learning method that can optimize the initial parameters of the model so that when the model encounters a new task, it can quickly adapt and achieve good performance with only a small amount of training data and very few gradient updates. Since MAML does not rely on a specific model architecture, it is applicable to a wide range of task types and can effectively share knowledge between multiple tasks, thereby greatly improving the model's generalization ability on new tasks, especially when data is scarce.

[0064] Specifically, the data set is first divided into multiple tasks, each task contains k groups of samples, and the task data is recorded as .in, Indicates power loss, where i =1, because it is one-dimensional, j represents the group number of the sample; Indicates wind speed and icing degree data (affecting power loss), where i =2, because it is two-dimensional, j represents the group number of the sample.

[0065] Define an LSTM model to implement nonlinear regression, and use the MAML algorithm to train the model to achieve few-sample learning.

[0066] The training process is divided into two parts: inner loop and outer loop.

[0067] The inner loop can be understood as performing a small-scale training on each task, so that the performance of the model on the task can be improved. Specifically, a task T is selected from the task set. i , whose training data is , calculate the loss function of the model on the training data of this task.

[0068] (16) Where θ is the initial model parameter.

[0069] According to the gradient of the loss function, the model parameters are updated to optimize the performance of the task: (17) In the formula, is the inner loop learning rate, It is task T i The updated model parameters.

[0070] The goal of the inner loop is to enable the model to quickly adapt to the training data for each specific task. After the inner loop is completed, the model parameters θ i ′ has undergone a task-specific update and has improved performance on that task.

[0071] The purpose of the outer loop is to aggregate information on multiple tasks, thereby adjusting the initial model parameters θ so that the model can achieve good performance when faced with new tasks with a small number of inner loop updates. This is equivalent to finding a good starting point that allows rapid learning on different tasks.

[0072] Specifically, for each task T i , using the updated parameters θ in the inner loop i ′ Calculate the validation loss on the validation set of this task: (18) The validation losses of all tasks are summed up to get the global meta-loss, based on which the initial model parameters θ are updated: (19) In the formula, is the outer loop learning rate.

[0073] Due to the different sizes of the processed data samples, the number of iterations of the inner loop and the outer loop also varies. Generally speaking, the number of iterations of the inner loop can be set to 1-5 times, while the number of iterations of the outer loop can be set to dozens to hundreds of times.

[0074] S104: when performing actual prediction, obtaining meteorological data and icing data under the wind turbine icing scenario within a set time range in the test area, inputting the meteorological data into a trained basic prediction model to obtain a basic prediction result of wind power, and inputting the icing data and wind speed data into a trained power loss prediction model to obtain a power loss prediction result; S105: Based on the basic wind power prediction results and the power loss prediction results, a final short-term wind power prediction result for the wind turbine icing scenario is obtained.

[0075] This example takes a wind farm in Northeast China as the research object, and sets up a calculation example to verify the effectiveness of the proposed method. The data used include NWP data from October 1, 2020 to February 28, 2021, the measured data of wind speed and wind power at the site, and the ice cover records provided by the site. The data set is divided according to Table 1.

[0076] Table 1 Dataset division

[0077] Firstly, the key parameters of the icing model are solved by icing simulation. The relevant variable settings in the icing simulation are shown in Table 2.

[0078] Table 2 Input meteorological factor values

[0079] The collision coefficient is affected by wind speed and liquid water content, while the freezing coefficient is mainly affected by temperature and liquid water content. The simulation of ice coverage is carried out according to the meteorological factors set in Table 2. The simulation results of ice coverage parameters are as follows: Figure 4 and Figure 5 As shown. Figure 4 It can be seen that the collision coefficient is roughly proportional to the wind speed and liquid water content. Under the same liquid water content, the higher the wind speed, the more water droplets will collide with the wind turbine blades, and the larger the collision coefficient. Figure 5 It can be seen that the freezing coefficient is not a simple linear relationship with meteorological variables. When the temperature is low enough, such as -10℃ and below, almost all water droplets left on the leaf surface can freeze, so the freezing coefficient can reach a level close to 1; when the temperature is high, because the liquid water droplets on the leaves have a "saturation" of being frozen, not all water droplets can be frozen. As the liquid water content in the air increases, the proportion of water droplets that can be frozen decreases, so the freezing coefficient decreases accordingly.

[0080] In order to reduce time resources and computing resources, the present invention proposes to use Kriging interpolation to obtain ice cover parameter results in a continuous meteorological numerical space. Taking the ice cover parameters of 96 time points in a day as an example, the time required for the calculation of the interpolation method and the traditional simulation method is shown in Table 3.

[0081] Table 3 Comparison of operation time

[0082] The comparison between the predicted value obtained by interpolation and the “actual value” obtained by simulation in high-dimensional data space is given in Table 4. The results show that the ice cover parameters obtained by the interpolation method have smaller errors than the simulation results, that is, the Kriging interpolation method can better realize the expansion of ice cover parameters from discrete values ​​to continuous numerical space, thereby replacing the simulation work and effectively saving time and computing resources.

[0083] Table 4 Comparison of interpolation results and actual values

[0084] The ice coverage at each moment is calculated using the method proposed by the present invention. Figure 6 The changes in the degree of icing during several icing events that occurred in February 2021 are shown. According to the characteristics of the curve, it can be further divided into three types: rapid icing growth, slow icing growth, and icing first and then melting.

[0085] It can be seen from the figure that the ice cover change process on different dates has great differences in growth rate, change trend, etc., and the method proposed in the present invention can reflect the phenomenon of ice melting and shedding. The ice cover change process is mainly affected by meteorological factors. Combined with the above ice cover parameter interpolation results, after calculating the ice cover on different dates and comparing the results, the following influencing rules can be summarized: (1) Effect of temperature: As the temperature decreases, the blade freezing coefficient increases, the amount of ice formed by the water droplets after impact increases, and the amount of ice coverage increases in the same period of time. However, as the temperature decreases, the lower temperature will cause the water droplets with smaller median volume in the air to turn into ice crystals, resulting in fewer water droplets impacting the blade surface, and the amount of ice coverage in the same period of time will decrease. Combined with the calculation results of the ice coverage parameters, the turning point appears at around -10°C.

[0086] (2) Influence of wind speed: The mass of ice on the blade surface increases with the increase of wind speed. The main reason is that wind speed directly affects the amount of water droplets collected on the blade surface: when the wind speed is low, the inertial force of the water droplets is relatively small compared to the viscous force they are subjected to, so fewer water droplets hit the blade surface; as the wind speed increases, the inertial force gradually increases, and the number of water droplets hitting the blade surface increases. That is, the greater the wind speed, the more supercooled water droplets blown onto the blade, and the more ice is covered on the blade when the temperature is low enough.

[0087] (3) Influence of liquid water content: When other icing conditions are the same, the larger the LWC, the more water droplets will hit the blade surface per unit time. If the ambient temperature is low enough, the amount of ice on the blade surface will increase faster.

[0088] In order to further verify the effectiveness of the prediction method proposed in this example, Figure 6 The wind power prediction results for the three icing events A, B, and C are shown in Figure 1. Figure 7-Figure 9 As shown in the figure, it can be seen that after error correction, the predicted power value decreases, but it is closer to the real power in most time periods, that is, the method proposed in the present invention can reflect the influence of power loss caused by ice on wind turbine blades when predicting wind power.

[0089] Combined with the icing calculation results, the icing degree calculated in icing event A generally shows a trend of increasing first and then decreasing, and the corresponding power loss value has a similar trend. When the icing degree is low, the power loss is low, and when the icing degree is high, the power loss also reaches a higher level. Although the icing degree calculated in icing event B shows a trend of continuous increase, its power loss does not strictly follow the trend of gradual increase. This is because the magnitude of the power loss value is not only determined by the icing degree, but also affected by factors such as wind speed. In this icing event, when the icing degree was low in the early stage, the magnitude of the power loss was mainly determined by the wind speed; and in the later stage, although the icing degree increased to a higher level, the wind speed was low at this time, the wind power itself was low, and the power loss was not large. The icing degree calculated in icing event C has an obvious rapid increase trend. At this time, the icing degree plays a greater role in the power loss, so the power loss also roughly conforms to the trend of gradual increase.

[0090] Based on the above analysis, we can summarize the impact of wind turbine blade icing on wind power: ① Blade icing will cause the wind turbine output power to decrease, resulting in "power loss", and under the same wind speed conditions, the greater the degree of icing, the greater the power loss value; ② The size of the power loss value is affected by both the degree of icing and the wind speed conditions. When the wind speed is high, a lower degree of icing will also cause a larger power loss. Under the same degree of icing, the higher the wind speed, the greater the percentage of power loss.

[0091] The normalized root mean square error NRMSE and the normalized mean absolute error NMAE are selected as prediction evaluation indicators, and the prediction errors are given in Table 5.

[0092] Table 5 Comparison of prediction errors

[0093] It can be seen from Table 5 that after the influence of icing is added to the prediction model using the method proposed in the present invention, the prediction accuracy of wind power is improved, and the NRMSE and NMAE are reduced by 0.0397 and 0.0439 respectively compared with when the influence of icing is not considered.

[0094] The method of this embodiment improves the prediction accuracy, thereby enhancing the ability of the new energy system to cope with wind power uncertainty in extreme icing scenarios and ensuring stable operation of the system.

[0095] Embodiment 2 In one or more embodiments, a short-term wind power prediction system for wind turbine icing scenarios is disclosed, comprising: The icing model simulation and solution module is used to consider the melting and shedding effects of ice, build an icing model suitable for wind turbine blades, and solve the key parameters of the icing model under set meteorological conditions through icing simulation; An ice cover data interpolation module is used to convert discrete ice cover data into continuous ice cover data corresponding to meteorological conditions at different times through data interpolation; A model training module is used to train the basic prediction model for wind power prediction and the power loss prediction model considering wind turbine icing scenarios; The power prediction module is used to obtain meteorological data and icing data under the wind turbine icing scenario within a set time range in the test area. The meteorological data is input into a trained basic prediction model to obtain a basic wind power prediction result. The icing data and wind speed data are input into a trained power loss prediction model to obtain a power loss prediction result. Based on the basic wind power prediction result and the power loss prediction result, the final short-term wind power prediction result for the wind turbine icing scenario is obtained.

[0096] It should be noted that the specific implementation method of the above modules is exactly the same as that in Example 1 and will not be described in detail.

[0097] Embodiment 3 In one or more embodiments, a terminal device is disclosed, which includes a processor and a memory, the processor is used to implement instructions; the memory is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the short-term wind power prediction method for wind turbine icing scenarios described in Example 1.

[0098] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0099] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0100] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software.

[0101] Embodiment 4 In one or more embodiments, a computer-readable storage medium is disclosed, in which a plurality of instructions are stored, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the short-term wind power prediction method for wind turbine icing scenarios described in Example 1.

[0102] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A short-term wind power prediction method for wind turbine icing scenarios, characterized in that: include: Considering the melting and shedding effects of ice, an icing model suitable for wind turbine blades is constructed, and the key parameters of the icing model under set meteorological conditions are solved through icing simulation. The discrete ice cover data are converted into continuous ice cover data corresponding to meteorological conditions at different times through data interpolation; Train the basic prediction model for wind power prediction and the power loss prediction model considering wind turbine icing scenarios respectively; Obtain meteorological data and icing data under the wind turbine icing scenario within a set time range in the test area, input the meteorological data into a trained basic prediction model to obtain a basic prediction result of wind power, and input the icing data and wind speed data into a trained power loss prediction model to obtain a power loss prediction result; Based on the basic wind power prediction results and power loss prediction results, the final short-term wind power prediction results for wind turbine icing scenarios are obtained.

2. A short-term wind power prediction method for wind turbine icing scenarios according to claim 1, characterized in that: Considering the melting and shedding effects of ice, an ice model suitable for wind turbine blades is constructed. Specifically: ； ； ； in, To consider the ice melting and shedding effect, the accumulated ice coverage of a single wind turbine is When the effects of ice melting and shedding are not considered i The increase in the maximum ice thickness at the moment, To quantify the proportionality coefficient of ice reduction between ice ablation and shedding effect, It means that after considering the effects of ice melting and shedding i The increase in the maximum ice thickness at the moment, express i The maximum ice coverage length of the airfoil at that moment; c represents the blade chord length, and n represents the set time; , are the collision coefficient and the freezing coefficient, is the mass fraction of water droplets in the air, is the velocity of the water drop, is the unit time, is the density of ice accumulation, This is the proportionality factor when solving for the volume of ice accumulation.

3. A short-term wind power prediction method for wind turbine icing scenarios as claimed in claim 2, characterized in that: Mass fraction of water droplets in air Specifically: ； Where Q and d represent the cloud water mixing ratio and air density, respectively.

4. A short-term wind power prediction method for wind turbine icing scenarios as claimed in claim 2, characterized in that: The key parameters of the icing model under set meteorological conditions are solved through icing simulation, specifically: Construct a physical model of the fan blades, use Fluent software to calculate the air flow field around the fan blades, import the flow field calculation results into the icing simulation software, preset the median diameter of the water droplets and the liquid water content in the air, and calculate the collision coefficient through the water droplet motion trajectory simulation and icing simulation. , freezing coefficient , solve for the proportionality coefficient when solving for the ice accumulation volume and the proportionality factor of ice reduction .

5. The short-term wind power prediction method for wind turbine icing scenarios according to claim 2, characterized in that: The discrete ice cover data is converted into continuous ice cover data corresponding to meteorological conditions at different times through data interpolation, specifically: The values ​​of icing parameters α1 and α3 under any meteorological data are estimated by Kriging interpolation method, and then the icing data under meteorological data at different times are calculated.

6. The short-term wind power prediction method for wind turbine icing scenarios according to claim 1, characterized in that: The basic prediction model for wind power prediction and the power loss prediction model considering wind turbine icing scenarios are trained separately. Specifically: Constructing a dataset D1 containing conventional meteorological data and corresponding wind power data, and a dataset D2 and a dataset D3 containing meteorological data and corresponding wind power data under icing scenarios respectively; Use the data set D1 to train the basic prediction model to obtain a trained basic prediction model; The wind power of the data set D2 is predicted using the trained basic prediction model, and the difference between the predicted wind power value and the actual value is recorded as power loss; the power loss and the corresponding wind speed and icing data constitute the power loss data set D4; The power loss prediction model is trained using the data set D4 to obtain a trained power loss prediction model; The trained basic prediction model and power loss prediction model are tested using the dataset D3.

7. The short-term wind power prediction method for wind turbine icing scenarios according to claim 1, characterized in that: The basic prediction model uses the LightGBM model, and the power loss prediction model uses the Few-shot model.

8. A short-term wind power prediction system for wind turbine icing scenarios, characterized in that: include: The icing model simulation and solution module is used to consider the melting and shedding effects of ice, build an icing model suitable for wind turbine blades, and solve the key parameters of the icing model under set meteorological conditions through icing simulation; An ice cover data interpolation module is used to convert discrete ice cover data into continuous ice cover data corresponding to meteorological conditions at different times through data interpolation; A model training module is used to train the basic prediction model for wind power prediction and the power loss prediction model considering wind turbine icing scenarios; The power prediction module is used to obtain meteorological data and icing data under the wind turbine icing scenario within a set time range in the test area. The meteorological data is input into a trained basic prediction model to obtain a basic wind power prediction result. The icing data and wind speed data are input into a trained power loss prediction model to obtain a power loss prediction result. Based on the basic wind power prediction result and the power loss prediction result, the final short-term wind power prediction result for the wind turbine icing scenario is obtained.

9. A terminal device, comprising a processor and a memory, wherein the processor is used to implement instructions; and the memory is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the short-term wind power prediction method for wind turbine icing scenarios as described in any one of claims 1-7.

10. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the short-term wind power prediction method for wind turbine icing scenarios as described in any one of claims 1-7.

Images