Wind power prediction method and system based on meteorological-icing coupling model
By constructing a meteorological-icing coupled model and combining CNN and Transformer models, the problem of accuracy in wind power prediction under icing conditions was solved, enabling accurate forecasting of icing disasters and wind power, and improving the operational stability and dispatch capability of the power grid under extreme weather conditions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SDIC HENAN NEW ENERGY CO LTD
- Filing Date
- 2025-12-23
- Publication Date
- 2026-05-19
AI Technical Summary
Existing wind power prediction models suffer from systematic biases under icing conditions, cannot effectively quantify power generation losses, and lack in-depth descriptions of the dynamic process of icing and its power impact mechanisms, leading to difficulties in grid dispatching and operational instability.
A method based on a meteorological-icing coupling model is adopted. By embedding a meteorological forecasting model and a rime and frost accumulation model, a hybrid architecture CNN and Transformer model is constructed to achieve accurate forecasting of icing disasters and wind power.
It significantly improves the accuracy and timeliness of wind power forecasting under icing weather, provides more reliable meteorological input, enhances the resilience and operational safety of the power grid under extreme weather conditions, and supports power grid dispatching departments in formulating power generation plans and optimizing dispatching strategies in advance.
Smart Images
Figure CN122068427A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a wind power prediction method and system based on a meteorological-icing coupling model, belonging to the field of new energy power prediction technology. Background Technology
[0002] Renewable energy, due to its cleanliness and sustainability, is now widely regarded as the most promising alternative energy source. Among all types of renewable energy, wind power dominates in installed capacity expansion and has become one of the most important sources of electricity in modern power grids. With its massive installed capacity, wind power has become a pillar power source for modern power grids. However, the output power of wind power generation is not stable but fluctuates constantly with seasonal variations, daily variations, and even minute-level turbulence in wind resources. This inherent randomness and uncertainty pose significant challenges to grid dispatching under high-proportion wind power integration. Therefore, accurate wind power forecasting is crucial for achieving rational grid dispatching, balancing supply and demand, and maintaining stable system operation.
[0003] Existing wind power prediction models can be broadly categorized into two types: physical methods and statistical methods. Physical methods utilize sub-models such as computational fluid dynamics, combining specific physical characteristics like terrain, surface roughness, obstacles, and atmospheric stability information to downscale numerical weather prediction data from grid points to the wind turbine level. Statistical models, on the other hand, rely on the statistical relationships between historical wind turbine power data and meteorological variables in numerical weather prediction (NWP). Currently, the most advanced statistical models are largely based on various machine learning algorithms. However, in most previous studies, both physical and statistical methods have typically only applied to "standard" scenarios, assuming that wind power output is unaffected by extreme weather events such as blizzards, sandstorms, and tornadoes. These models do not incorporate the physical mechanisms underlying the impact of extreme weather on wind turbine operation; most rely on machine learning-based statistical models, manually removing outlier data influenced by extreme conditions during data preprocessing to improve the overall model performance.
[0004] In recent years, frequent winter icing has posed a severe challenge to power supply security and the safe and stable operation of the power grid. Icing significantly deteriorates the aerodynamic performance of wind turbine blades, causing mass imbalances and abnormal loads. This not only reduces power generation efficiency but can also trigger protective shutdowns of the units, and in severe cases, even damage the blade structure. Against this backdrop, icing weather conditions have become one of the most important "non-standard" operating scenarios in the large-scale application of wind energy, urgently requiring high-accuracy wind power forecasting under icing conditions to support grid dispatching decisions and safe and stable operation.
[0005] Currently, most operational wind turbines in icing-prone areas lack effective anti-icing and de-icing systems. During winter operation, power generation losses due to blade icing are significant, and there are predictive blind spots. Although some studies have attempted to identify icing events using icing prediction models, these models are mostly limited to qualitative assessments of icing phenomena, lacking in-depth descriptions of the dynamic process of icing and its power impact mechanisms. They cannot effectively quantify power generation losses and are even less capable of supporting the power prediction needs of actual power dispatch. As mentioned earlier, existing wind power prediction models often exhibit systematic biases under icing conditions, severely limiting their applicability in engineering practice; and existing icing prediction models can only determine whether icing has occurred, failing to provide detailed information on potential wind power losses.
[0006] Therefore, developing a novel wind power prediction model based on the icing impact mechanism to quantify power loss has become a key technical approach to improve the predictability of wind power and the resilience of power grid operation under icing weather, and has important theoretical significance and engineering application value. Summary of the Invention
[0007] The purpose of this invention is to provide a wind power prediction method and system based on a meteorological-icing coupling model. This method addresses the shortcomings of existing prediction models, such as their high dependence on meteorological forecast data and historical power generation data, the lack of input to the meteorological-icing coupling model, and the lag in the prediction of icing disasters by existing power prediction models. By embedding and coupling the meteorological forecast model with the rime and hoarfrost accumulation models, the method can predict icing disasters. It can generate more accurate wind farm power predictions under icing weather conditions, thereby providing key decision support for power grid dispatching departments.
[0008] To achieve the above objectives, the present invention employs the following technical solution: A wind power prediction method based on a meteorological-icing coupling model includes the following steps: Collect and process historical power generation data from power plants, historical meteorological data measured by wind towers, observation data from meteorological stations, and historical numerical weather forecast data. Perform data cleaning and standardization, and extract meteorological elements from the same period as the power generation data. Numerical weather forecasts are embedded and coupled with models of rime and hoarfrost accumulation, and topographic factors are introduced to construct a WRF-ICE icing forecast model, transforming historical numerical weather forecast data into high-precision historical numerical weather forecast data. A wind power prediction model is constructed, which adopts a hybrid architecture of CNN and Transformer model. The input includes cleaned historical power generation data of the wind farm, historical meteorological data measured by the wind tower, and high-precision historical numerical weather forecast data. Through multi-scale feature extraction and attention mechanism, the wind power prediction value is output. By comparing the predicted wind power with the actual power data, the mean absolute error, root mean square error and prediction accuracy are calculated, and the wind power prediction model is optimized accordingly. The optimized wind power prediction model is deployed to collect real-time power generation data from wind farms, real-time meteorological data measured by wind measurement towers, and current high-precision numerical weather forecast data. These data are then input into the optimized wind power prediction model to predict wind power output.
[0009] Preferably, the data cleaning and standardization includes: The historical power generation data of the power station and the historical meteorological data measured by the wind tower were aligned by time, and missing values were filled and outliers were removed to build a clean dataset. Based on ECWMF model forecast data, soil moisture, temperature, surface pressure, and u and v components of each pressure layer are extracted from historical numerical weather prediction data. Invalid values were removed and the meteorological station observation data were standardized.
[0010] Preferably, the construction of the WRF-ICE icing forecast model includes: Configure the WRF numerical weather prediction model, construct the initial field based on historical numerical weather prediction data, and optimize the parameterization scheme of physical processes; Meteorological data output from WRF were input into the Jones rime model and the Makkonen hoarfrost model, and topographic factors were introduced for localization correction to calculate the icing growth. The model parameters were corrected using icing observation data to establish a WRF-ICE icing thickness prediction model.
[0011] Preferably, the formula for calculating the ice accumulation growth of the Jones frost model is as follows: , , in, This refers to the revised standard for ice thickness increase due to rain-induced ice accumulation. The standard increase in ice thickness for rain-induced ice accumulation. The density of ice, The density of water, For precipitation intensity, For wind speed, For changes in the mass of accumulated ice, This represents the diameter of the ice buildup on the conductor in the previous time period. The weighting coefficient for ice content. This represents the standard ice thickness increase.
[0012] Preferably, the increase in standard ice thickness due to rain accumulation in the Jones rime model is further weighted by a weighting function. and The correction method is as follows: when hour, Rainfall-induced ice accumulation standard ice thickness increase =0; when hour, Rainfall-induced ice accumulation standard ice thickness increase =0; when and At that time, the increase in standard ice thickness due to rain-induced ice accumulation for: , in, The standard ice thickness increase for rain-induced icing calculated using the Jones simplified model; In other cases, the increase in standard ice thickness due to rain accumulation. for , , in, This is the lower limit of the ice content. The upper limits for ice content are set at 0.05 and 0.85 respectively. Ice content Wet-bulb temperature, This is the lower limit of the wet-bulb temperature. This represents the upper limit of the wet-bulb temperature. This is the weighting factor for the wet-bulb temperature at the ground. This is the weighting coefficient for ice content.
[0013] Preferably, the amount of ice accumulation in the rime is calculated using the Makkonen model: , in, For changes in the mass of accumulated ice, This refers to the liquid water content. For collision rate, The collection rate is 1. This represents the freeze rate, with a value of 1.
[0014] Preferably, the wind power prediction model includes an input layer, a CNN feature extraction module, a feature transformation module, a Transformer module, and an output layer; The CNN feature extraction module includes parallel Residual7_7 residual blocks and two Residual residual blocks. The output features of the Residual7_7 residual blocks are processed by a spatial attention module, and the output features of the Residual residual blocks are processed by a channel attention module and a hybrid attention module, respectively. The feature transformation module converts the features output by the CNN feature extraction module into sequence data through adaptive pooling and dimensionality rearrangement, adds learnable positional encoding, and then inputs it into the Transformer module. The Transformer module uses a multi-head self-attention mechanism to model long-range dependencies, and finally inputs and outputs the layers through a feedforward neural network. The output layer outputs wind power prediction values through nonlinear mapping. The output layer includes global average pooling, a fully connected layer, and a Softmax function in sequence.
[0015] A wind power prediction system based on a meteorological-icing coupling model includes: The data acquisition and processing module is used to collect and process historical power generation data from the power station, historical meteorological data measured by the wind tower, observation data from meteorological stations, and historical numerical weather forecast data. It also performs data cleaning and standardization, and extracts meteorological elements from the same period as the power generation data. The meteorological-icing coupled forecast module is used to run the WRF-ICE icing forecast model and convert historical numerical weather forecast data into high-precision historical numerical weather forecast data. The power prediction module constructs a wind power prediction model, which adopts a hybrid architecture of CNN and Transformer models. The inputs include cleaned historical power generation data of the wind farm, historical meteorological data measured by the wind tower, and high-precision historical numerical weather forecast data. Through multi-scale feature extraction and attention mechanism, the output is the wind power prediction value. The training module compares the predicted wind power with the actual power data, calculates the mean absolute error, root mean square error, and prediction accuracy, and optimizes the wind power prediction model accordingly.
[0016] Preferably, the meteorological-icing coupled forecasting module further includes a WRF model configuration unit, a Jones-Makkonen icing calculation unit, and a terrain correction unit.
[0017] The advantages of this invention are: The WRF-ICE icing forecast model embeds the physical formation mechanism of icing into the core of the forecast, directly generating accurate icing thickness forecasts. This effectively improves the accuracy of icing forecasts, overcomes the shortcomings of large-scale global models in this type of weather forecasting, and provides more reliable meteorological input for power prediction.
[0018] This invention fills a crucial physical gap between weather forecasting and power prediction, innovatively embedding the WRF model with icing models such as Jones and Makkonen, transforming conventional meteorological elements into quantitative indicators of icing hazards specific to wind farms. It provides power prediction models with previously completely missing key physical inputs describing icing hazards.
[0019] This significantly improves the accuracy and timeliness of predictions during icing disasters, elevating post-disaster fitting to pre-disaster early warning. Traditional power models can only passively respond after icing causes a power drop. The wind power prediction method based on the meteorological-icing coupled model can predict the impact of icing on the aerodynamic characteristics and operating status of wind turbines in advance, thus making early predictions before a sharp drop in power occurs, greatly reducing power prediction errors under icing disasters. This shift from perceiving losses to predicting risks buys valuable decision-making time for grid dispatch.
[0020] This technology enhances the resilience of the power grid in the face of extreme weather and ensures operational safety. By providing more accurate and earlier predictions of wind power output drops under extreme weather conditions, this patented technology enables power grid dispatching departments to formulate power generation plans in advance, activate backup resources, and optimize dispatching strategies. This effectively balances the power shortage caused by sudden drops in wind power, thereby significantly improving the power supply security and operational stability of the power system under extreme winter weather conditions and promoting the high-level consumption of new energy sources. Attached Figure Description
[0021] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.
[0022] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 Schematic diagram of the WRF coupled icing model architecture; Figure 3 Schematic diagram of wind power prediction model architecture; Figure 4 This embodiment illustrates the regression analysis between predicted and actual values from the icing model. Figure 5 This embodiment presents a schematic diagram of regional icing forecasting. Figure 6 This embodiment presents a schematic diagram of regional icing forecasting. Detailed Implementation
[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0024] Example 1 like Figure 1 As shown, a wind power prediction method based on a meteorological-icing coupled model is proposed. This method addresses the shortcomings of existing prediction models, such as high dependence on meteorological forecast data and historical power generation data, lack of input for the meteorological-icing coupled model, and the lag in icing disaster forecasting of existing power prediction models. By embedding and coupling a meteorological forecasting model (WRF) with Jones and Makkonen icing models, the method enables the prediction of icing disasters, constructs a wind power prediction model based on the icing model, and verifies the effectiveness of the model. Specifically, it includes four core steps: data collection and standardization, meteorological-icing model coupling, wind power prediction model construction, and icing process power prediction model verification. This method can generate more accurate wind farm power prediction curves under icing weather conditions, thus providing crucial decision support for grid dispatching departments. Specifically, it enables the power grid to detect the risk of a sudden drop in wind power output in advance, gaining valuable early warning time for starting backup power sources, adjusting power generation plans, and even initiating orderly power consumption schemes in extreme situations. Ultimately, it effectively improves the power supply security and operational resilience of high-proportion renewable energy power grids under extreme weather conditions.
[0025] Specifically, it includes the following steps: S1: Collect and process historical power generation data from power plants, historical meteorological data measured by wind towers, observation data from meteorological stations, and historical numerical weather forecast data. Perform data cleaning and standardization, and extract meteorological elements from the same period as the power generation data. S2: Embedded coupling of numerical weather forecasts with rime and hoarfrost icing models, and introduction of topographic factors, to construct the WRF-ICE icing forecast model, and to transform historical numerical weather forecast data into high-precision historical numerical weather forecast data. S3: Construct a wind power prediction model, using a hybrid architecture of CNN and Transformer models. The inputs include cleaned historical power generation data from wind farms, historical meteorological data measured by wind towers, and high-precision historical numerical weather forecast data. Through multi-scale feature extraction and attention mechanisms, the output is the predicted wind power value. S4: Compare the predicted wind power with the actual power data, calculate the mean absolute error, root mean square error and prediction accuracy, and optimize the wind power prediction model accordingly. S5: Deploy the optimized wind power prediction model, collect real-time power generation data from the wind farm, real-time meteorological data measured by the wind tower, and current high-precision numerical weather forecast data, and input them into the optimized wind power prediction model to predict wind power.
[0026] As a refinement of the above embodiment, step S1 involves standardizing and cleaning the historical power generation data of the power station, the historical meteorological data measured by the wind tower, and the observation data from meteorological department stations; and extracting meteorological elements from the historical numerical weather forecast data. The specific implementation steps are as follows: S101: Comprehensively collect historical power generation data of the power station, historical meteorological data measured by the wind measurement tower, and historical numerical weather forecast data for the same period.
[0027] S102: Historical power generation data from wind farms and historical meteorological data measured by wind towers are aligned and integrated according to a unified time base to construct an initial dataset. Data cleaning is then performed on the dataset. First, missing values and invalid negative values in the power generation data are filled with zeros to ensure the completeness and continuity of the dataset and avoid affecting the accuracy of subsequent analysis. Second, extreme weather, communication failures, damaged measuring instruments, wind turbine failures, and wind curtailment can all cause a large amount of abnormal data. After identifying abnormal data that does not conform to theoretical relationships based on wind speed-power scatter plots, an outlier is automatically removed using a machine learning method based on an autoregressive model, achieving effective cleaning. Finally, manual review is supplemented by automatic cleaning to manually remove hidden anomalies that are difficult for the model to fully identify in special scenarios such as power curtailment operation, obtaining a reliable dataset that lays a solid foundation for subsequent modeling.
[0028] S103: Extraction of historical numerical weather prediction data elements. Based on ECWMF model forecast data, extract elements from historical periods of the model's input variables, such as soil moisture, temperature, surface air pressure, and u and v components of each pressure layer.
[0029] S104: Perform data cleaning and standardization on meteorological station observation data, mainly including invalid value removal.
[0030] Wind tower data (providing wind speed, wind direction, temperature, and air pressure data at different altitudes) is the most direct and reliable field-level meteorological observation data. In addition, station observation data from meteorological departments are also integrated to obtain real-time information on temperature, pressure, humidity, and wind over a wider area. This data, after undergoing a rigorous quality control process, is used in real-time for data assimilation, dynamically correcting the initial field of the numerical model and significantly reducing prediction bias caused by complex terrain.
[0031] Numerical weather prediction data primarily utilizes EC-HRES data from the European Centre for Medium-Range Weather Prediction (ECMWF). Through data assimilation techniques, global historical forecast data is integrated into the numerical model, providing spatiotemporally continuous "best estimates" of dozens of vertical levels, from the surface to the upper atmosphere, including temperature, humidity, air pressure, wind speed, and wind direction. Its horizontal resolution is typically 0.25° × 0.25° (approximately 30 km), and its temporal resolution is 1 hour, providing physically consistent and reliable initial and boundary conditions for mesoscale numerical models.
[0032] Historical power generation data of new energy power plants includes active power, reactive power, cumulative power generation and theoretical power generation, power curtailment data, operation and maintenance data, wind turbine fault records, etc., at the power plant and individual unit levels. These data are aligned and stored using a unified timestamp, and together they form the basic data set for training and validating wind power prediction models and analyzing the operation status of power plants.
[0033] As a refinement of the above embodiments, such as Figure 2 As shown, step S2 constructs a WRF-ICE icing forecast model based on the Weather Forecasting Model (WRF) and the Jones and Makkonen icing models. The specific implementation steps are as follows: S201: Configure the WRF model, a mesoscale weather numerical forecasting model, to construct a land-atmosphere coupled initial field containing rich scale information based on multi-layer data such as soil moisture and temperature with high spatiotemporal resolution, and optimize the parameterization scheme of physical processes to improve the simulation accuracy of weather processes such as freezing rain and rime. S202: Input meteorological forecast model (WRF) data into physical icing models such as Jones (ice frost) and Makkonen (rime), introduce regional topographic factors, and perform localized corrections to the models to achieve a refined simulation of icing.
[0034] The increase in rime ice accumulation is calculated using the revised Jones simplified model formula as follows: , , in, This refers to the revised standard for ice thickness increase due to rain-induced ice accumulation. The standard increase in ice thickness for rain-induced ice accumulation. The density of ice, The density of water, For precipitation intensity, For wind speed, For changes in the mass of accumulated ice, This represents the diameter of the ice buildup on the conductor in the previous time period. The weighting coefficient for ice content. This represents the standard ice thickness increase.
[0035] The Jones simple model is only applicable to simulating rime formation during freezing rain. However, other types of mixed precipitation (such as freezing rain mixed with rain, freezing rain mixed with ice pellets, and freezing rain mixed with snow) can also form rime. Therefore, the standard ice thickness increase for rime formation in the Jones rime model is further adjusted using a weighting function. and The correction method is as follows: when hour, , indicates that the precipitation type is snow or (and) ice pellets, and the precipitation is unlikely to adhere to the conductor. The standard ice thickness increase for rain-induced icing is... =0; when hour, This indicates that the precipitation type is rain, and the standard for ice thickness increase due to rain accumulation is... =0; when and At that time, the precipitation type was freezing rain, the precipitation completely froze, and the increase in ice thickness was measured according to the standard for accumulated ice. for: , in, The standard ice thickness increase for rain-induced icing calculated using the Jones simplified model; In other cases, the precipitation is a mixture of freezing rain, freezing rain with snow, or (and) ice pellets, with only a portion of the precipitation freezing on the conductor. The standard ice thickness increase for rain-induced icing is... for , , in, This is the lower limit of the ice content. The upper limits for ice content are set at 0.05 and 0.85 respectively. Ice content Wet-bulb temperature, This is the lower limit of the wet-bulb temperature. This represents the upper limit of the wet-bulb temperature. This is the weighting factor for the wet-bulb temperature at the ground. This is the weighting coefficient for ice content.
[0036] The amount of ice accumulation in the rime was calculated using the Makkonen model: , in, For changes in the mass of accumulated ice, This refers to the liquid water content. For collision rate, The collection rate is 1. This represents the freeze rate, with a value of 1. According to Finstad et al., parameters are obtained by using the median volume diameter of the droplet, the diameter of the ice accumulation on the wire, the density of air, and the absolute viscosity of air.
[0037] The model parameters were corrected using icing observation results, and an optimized WRF coupled icing thickness prediction model was finally established.
[0038] As a refinement of the above embodiments, such as Figure 3 As shown, step S3 constructs a wind power prediction model: Wind power prediction is achieved by combining a hybrid architecture of CNN and Transformer. The input multi-channel data includes cleaned historical power generation data, meteorological data from the wind tower, and extracted historical numerical weather forecast data for the same period. First, multi-scale spatial features are extracted through multiple residual blocks Residual_7_7 and Residual. The residual connection alleviates gradient vanishing, and the spatial and channel attention module dynamically enhances the feature response of key positions and channels. Subsequently, the feature map is converted into a sequence form through adaptive pooling and dimensional rearrangement, and spatial location information is injected by combining learnable positional encoding. Then, a multi-head self-attention mechanism is used to model the long-range temporal and spatial dependencies in the sequence, capturing the coordinated changes of wind turbine clusters in the wind farm and the impact of historical states on the current power. Finally, the global features output by attention are input into the MLP prediction head, and the univariate wind power prediction value is output through nonlinear mapping.
[0039] In a preferred but non-limiting embodiment of the present invention, the overall design of the model begins with the input structure. The input adopts the form of a five-dimensional tensor, namely (batch_size, channels, time_steps, height, width), which includes various meteorological elements, time series, and spatial grid information. This structure ensures the integrity of the data across the "element-time-space" dimensions, allowing the model to be built directly onto the original spatiotemporally coupled data, avoiding the information loss caused by the separation of space and time in traditional methods. After the data enters the model, the primary issue to be addressed is the representation of the temporal order. Therefore, a location encoding mechanism is introduced during the modeling process, enabling the model to more accurately understand the temporal mapping relationship between wind speed and power.
[0040] Based on the reasonable representation of temporal features, the model further enhances its ability to capture key information through an attention mechanism, avoiding interference from redundant data. After being weighted by the attention mechanism, the data is fed into residual blocks based on 3D convolution to achieve multi-scale spatiotemporal feature extraction. However, although 3D convolution performs excellently in local and medium-term feature extraction, its temporal receptive field is still limited by the size of the convolution kernel, making it difficult to fully capture dependencies spanning tens of hours or even longer. In wind power forecasting tasks, long-term meteorological processes (such as cold air accumulation or the evolution of large-scale weather systems) have a significant impact on future power output. Therefore, the model introduces a Transformer module to compensate for the shortcomings of 3D convolution in long-term dependency modeling.
[0041] As a refinement of the above embodiments, step S4 specifically includes: The power prediction data sequence output by the wind power prediction model is compared with the historical actual power data sequence of the wind farm. Multiple preset quantitative indicators are used to statistically evaluate the model's prediction accuracy and event capture capability under icing disaster conditions. The evaluation indicators used include, but are not limited to: Accuracy: The accuracy of short-term next-day power prediction is calculated based on the icing thickness predicted by the WRF-ICE coupled model, combined with the actual power data of the wind farm and the assessment rules of the "two detailed rules" of Henan Province. Mean absolute error: The average absolute error between the predicted power data series and the historical actual power data series is calculated to assess the overall bias level of the prediction model during the icing process. Root mean square error: The mean square root of the error between the predicted power data sequence and the historical actual power data sequence is calculated. This indicator is more sensitive to large error terms in the prediction and is used to comprehensively evaluate the prediction stability and accuracy of the model in the nonlinear strong disturbance process of icing.
[0042] Through the above-mentioned multi-indicator joint verification system, the accuracy and reliability of the power prediction model in the special meteorological disaster process of icing are comprehensively evaluated from multiple dimensions such as event early warning, overall deviation and extreme error.
[0043] To more clearly illustrate the key features of the wind power prediction method based on the meteorological-icing coupling model and its system, as well as the significant advancements it brings to existing technologies, a verification example is provided below to further verify the feasibility of the aforementioned meteorological-icing coupling power prediction model process. This verification example demonstrates the establishment and operationalization of an optimized meteorological-icing coupling power prediction model at a wind farm in Henan Province.
[0044] First, the accuracy of the icing forecast in the Henan region was verified. The dataset consists of meteorological elements at 15-minute intervals and a resolution of 2km×2km from March 3 to March 5, 2025, as predicted by WRF daily, and meteorological observations during the same period. Based on the dataset, the verification results from 08:00 on March 3 to 08:00 on March 4 are given here. Figure 4 The graph showing the predictive capability evaluation of the icing model based on WRF mode coupling is displayed. Figure 5 The spatial distribution of ice cover forecasts in Henan Province based on WRF model coupling is shown. As can be seen from the figure, the ice cover model has good predictive ability, with a correlation coefficient of 0.74, a root mean square error of 3.17, and a mean absolute error of 1.73.
[0045] Based on the meteorological-icing coupled wind power prediction model, the wind power prediction results for renewable energy power plants under icing conditions are obtained (partial results are shown in...). Figure 6 As shown in the figure, the weather-icing coupled wind power prediction model effectively improves the accuracy of wind power prediction, making the prediction results more consistent with actual operating scenarios. From March 3 to March 5, 2025, wind turbines experienced large-scale shutdowns due to blade icing, resulting in an overall zero measured power data. If only the original prediction model is used, the impact of extreme weather is not fully considered, and the prediction results deviate significantly from the actual output on the 3rd and 4th, only gradually responding to the power changes caused by icing after the 4th. After introducing the extreme weather correction algorithm, the model can dynamically adjust according to the impact of disaster events on unit operation, making the predicted power curve closer to the actual generated power and responding more promptly to power changes caused by icing. This result shows that incorporating extreme weather correction can effectively improve the accuracy of wind power prediction, making the prediction results more consistent with actual operating scenarios.
[0046] Example 2 A wind power prediction system based on a meteorological-icing coupling model includes: The data acquisition and processing module is used to collect and process historical power generation data from the power station, historical meteorological data measured by the wind tower, observation data from meteorological stations, and historical numerical weather forecast data. It also performs data cleaning and standardization, and extracts meteorological elements from the same period as the power generation data. The meteorological-icing coupled forecast module is used to run the WRF-ICE icing forecast model and convert historical numerical weather forecast data into high-precision historical numerical weather forecast data. The power prediction module constructs a wind power prediction model, which adopts a hybrid architecture of CNN and Transformer models. The inputs include cleaned historical power generation data of the wind farm, historical meteorological data measured by the wind tower, and high-precision historical numerical weather forecast data. Through multi-scale feature extraction and attention mechanism, the output is the wind power prediction value. The training module compares the predicted wind power with the actual power data, calculates the mean absolute error, root mean square error, and prediction accuracy, and optimizes the wind power prediction model accordingly.
[0047] As a refinement of the above embodiments, the meteorological-icing coupled forecasting module further includes a WRF model configuration unit, a Jones-Makkonen icing calculation unit, and a terrain correction unit.
[0048] This disclosure also provides a wind power prediction device based on a meteorological-icing coupling model, including a processor and a memory. Optionally, the device may further include a communication interface and a bus. The processor, communication interface, and memory can communicate with each other via the bus. The communication interface can be used for information transmission. The processor can call logical instructions in the memory to execute the wind power prediction method based on the meteorological-icing coupling model described above.
[0049] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.
[0050] Memory, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as the program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor executes the program instructions / modules stored in the memory to perform functional applications and data processing, that is, to implement the wind power prediction method based on the meteorological-icing coupling model in the above embodiments.
[0051] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory may include high-speed random access memory and may also include non-volatile memory.
[0052] This disclosure provides a computer-readable storage medium storing computer-executable instructions configured to execute the above-described wind power prediction method based on a meteorological-icing coupling model.
[0053] The aforementioned computer-readable storage medium may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.
[0054] The technical solutions of this disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more 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 method described in this disclosure. The aforementioned storage medium can be a non-transitory storage medium, including: a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and other media capable of storing program code. It can also be a transient storage medium.
[0055] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A wind power prediction method based on a meteorological-icing coupling model, characterized in that, Includes the following steps: Collect and process historical power generation data from power plants, historical meteorological data measured by wind towers, observation data from meteorological stations, and historical numerical weather forecast data. Perform data cleaning and standardization, and extract meteorological elements from the same period as the power generation data. Numerical weather forecasts are embedded and coupled with models of rime and hoarfrost accumulation, and topographic factors are introduced to construct a WRF-ICE icing forecast model, transforming historical numerical weather forecast data into high-precision historical numerical weather forecast data. A wind power prediction model is constructed, which adopts a hybrid architecture of CNN and Transformer model. The input includes cleaned historical power generation data of the wind farm, historical meteorological data measured by the wind tower, and high-precision historical numerical weather forecast data. Through multi-scale feature extraction and attention mechanism, the wind power prediction value is output. By comparing the predicted wind power with the actual power data, the mean absolute error, root mean square error and prediction accuracy are calculated, and the wind power prediction model is optimized accordingly. The optimized wind power prediction model is deployed to collect real-time power generation data from wind farms, real-time meteorological data measured by wind measurement towers, and current high-precision numerical weather forecast data. These data are then input into the optimized wind power prediction model to predict wind power output.
2. The wind power prediction method based on the meteorological-icing coupling model according to claim 1, characterized in that, The data cleaning and standardization include: The historical power generation data of the power station and the historical meteorological data measured by the wind tower were aligned by time, and missing values were filled and outliers were removed to build a clean dataset. Based on ECWMF model forecast data, soil moisture, temperature, surface pressure, and u and v components of each pressure layer are extracted from historical numerical weather prediction data. Invalid values were removed and the meteorological station observation data were standardized.
3. The wind power prediction method based on the meteorological-icing coupling model according to claim 1, characterized in that, The construction of the WRF-ICE icing forecast model includes: Configure the WRF numerical weather prediction model, construct the initial field based on historical numerical weather prediction data, and optimize the parameterization scheme of physical processes; Meteorological data output from WRF were input into the Jones rime model and the Makkonen hoarfrost model, and topographic factors were introduced for localization correction to calculate the icing growth. The model parameters were corrected using icing observation data to establish a WRF-ICE icing thickness prediction model.
4. The wind power prediction method based on the meteorological-icing coupling model according to claim 3, characterized in that, The formula for calculating the ice accumulation growth in the Jones rime model is as follows: ， ， in, This represents the revised standard for ice thickness increase due to rain-induced ice accumulation. The standard increase in ice thickness for rain-induced ice accumulation. The density of ice, The density of water, For precipitation intensity, For wind speed, For changes in the mass of accumulated ice, This refers to the diameter of the ice buildup on the conductor in the previous time period. The weighting coefficient for ice content. This represents the standard ice thickness increase.
5. The wind power prediction method based on the meteorological-icing coupling model according to claim 4, characterized in that, The standard ice thickness increase of the Jones rime model is also calculated using a weighting function. and The correction method is as follows: when hour, Rainfall-induced ice accumulation standard ice thickness increase =0; when hour, Rainfall-induced ice accumulation standard ice thickness increase =0; when and At that time, the increase in standard ice thickness due to rain-induced ice accumulation for: ， in, The standard ice thickness increase for rain-induced icing calculated using the Jones simplified model; In other cases, the increase in standard ice thickness due to rain accumulation. for ， ， in, This is the lower limit of the ice content. The upper limits for ice content are set at 0.05 and 0.85 respectively. Ice content Wet-bulb temperature, This is the lower limit of the wet-bulb temperature. This represents the upper limit of the wet-bulb temperature. This is the weighting factor for the wet-bulb temperature at the ground. This is the weighting coefficient for ice content.
6. The wind power prediction method based on the meteorological-icing coupling model according to claim 5, characterized in that, The amount of ice accumulation in the rime was calculated using the Makkonen model: ， in, For changes in the mass of accumulated ice, This refers to the liquid water content. For collision rate, The collection rate is 1. This represents the freeze rate, with a value of 1.
7. The wind power prediction method based on the meteorological-icing coupling model according to claim 1, characterized in that, The wind power prediction model includes an input layer, a CNN feature extraction module, a feature transformation module, a Transformer module, and an output layer. The CNN feature extraction module includes parallel Residual7_7 residual blocks and two Residual residual blocks. The output features of the Residual7_7 residual blocks are processed by a spatial attention module, and the output features of the Residual residual blocks are processed by a channel attention module and a hybrid attention module, respectively. The feature transformation module converts the features output by the CNN feature extraction module into sequence data through adaptive pooling and dimensionality rearrangement, adds learnable positional encoding, and then inputs it into the Transformer module. The Transformer module uses a multi-head self-attention mechanism to model long-range dependencies, and finally inputs and outputs the layers through a feedforward neural network. The output layer outputs wind power prediction values through nonlinear mapping. The output layer includes global average pooling, a fully connected layer, and a Softmax function in sequence.
8. A wind power prediction system based on a meteorological-icing coupling model, characterized in that, The wind power prediction method based on the meteorological-icing coupling model as described in any one of claims 1-7 includes: The data acquisition and processing module is used to collect and process historical power generation data from the power station, historical meteorological data measured by the wind tower, observation data from meteorological stations, and historical numerical weather forecast data. It also performs data cleaning and standardization, and extracts meteorological elements from the same period as the power generation data. The meteorological-icing coupled forecast module is used to run the WRF-ICE icing forecast model and convert historical numerical weather forecast data into high-precision historical numerical weather forecast data. The power prediction module constructs a wind power prediction model, which adopts a hybrid architecture of CNN and Transformer models. The inputs include cleaned historical power generation data of the wind farm, historical meteorological data measured by the wind tower, and high-precision historical numerical weather forecast data. Through multi-scale feature extraction and attention mechanism, the output is the wind power prediction value. The training module compares the predicted wind power with the actual power data, calculates the mean absolute error, root mean square error, and prediction accuracy, and optimizes the wind power prediction model accordingly.
9. A wind power prediction device based on a meteorological-icing coupling model, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to execute the wind power prediction method based on the meteorological-icing coupling model as described in any one of claims 1-7 when running the program instructions.
10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the wind power prediction method based on the meteorological-icing coupling model as described in any one of claims 1-7 above.