Icing area prediction method and device, electronic equipment and storage medium

Abstract

This application belongs to the field of power system disaster prevention technology, and provides a method, device, electronic equipment, and storage medium for predicting icing areas. The method includes: acquiring atmospheric reanalysis datasets, site observation datasets, and terrain elevation datasets; preprocessing and filtering icing elements in the above data to obtain data based on freezing rain and climate fusion datasets, terrain feature tensors, and icing element data; inputting the obtained data into a pre-trained terrain adaptive convolutional network Monte Carlo random deactivation model, outputting the average probability of freezing rain and the standard deviation of uncertainty; inputting the average probability of freezing rain, the standard deviation of uncertainty, 10-meter high wind speed, wet-bulb temperature, supercooled precipitation, preset time step, icing density, and conductor equivalent radius into a Makkonen calibration model, outputting the predicted icing thickness; determining the predicted icing area based on a preset region, the predicted icing thickness, and preset conditions; thus improving the prediction accuracy of icing thickness and icing area.

Description

Technical Field

[0001] This application belongs to the field of power system disaster prevention technology, and in particular relates to prediction methods, devices, electronic equipment and storage media for icing areas. Background Technology

[0002] Ice accumulation is a major natural disaster that seriously threatens the safe operation of the power grid. The accumulation of ice on transmission lines can lead to a series of chain faults such as conductor breakage, tower collapse, and insulator flashover, causing large-scale power outages.

[0003] The current data foundation used for icing simulation limits the accuracy and precision of icing risk assessment. Icing data acquisition primarily relies on two sources: the first is discretely distributed ground meteorological station observation data, such as the China Meteorological Administration's station network; while this type of data has high single-point observation accuracy and reliability, its spatial representativeness is insufficient. The second type is large-scale atmospheric reanalysis data generated based on numerical assimilation techniques, such as the European Centre for Medium-Range Weather Forecasts' ERA5 dataset (i.e., the fifth-generation atmospheric reanalysis dataset); this type of data has the advantages of global coverage, spatiotemporal continuity, and complete variables, providing a valuable foundational field for meteorological research. However, when directly driving icing physical models such as the internationally accepted Makkonen model, this type of data exhibits inherent systematic biases. Furthermore, current models used to predict icing thickness have low accuracy for preset areas and cannot meet the needs of power system disaster prevention scenarios.

[0004] Existing technologies suffer from low accuracy in icing simulation data and icing thickness prediction models, which fails to meet the needs of power system disaster prevention scenarios. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and storage medium for predicting icing areas, which can solve the problem that the low prediction accuracy of icing simulation data and icing thickness prediction models cannot meet the needs of power system disaster prevention scenarios.

[0006] Firstly, this application provides a method for predicting icing areas, including: Acquire atmospheric reanalysis dataset, meteorological observation dataset from Chinese ground stations, and topographic elevation dataset; Preprocessing and icing element screening were performed on atmospheric reanalysis dataset, meteorological observation data from Chinese ground stations, and topographic elevation data to obtain a freezing rain climate fusion dataset based on a preset time step, extracted topographic feature tensor, and icing element data. The icing element data included 2-meter high-temperature air, wet-bulb temperature, 10-meter high-wind speed, supercooled precipitation, relative humidity, and equivalent potential temperature of the 850 hPa pressure layer. The freezing rain climate fusion dataset, topographic feature tensor, and icing element data are input into a pre-trained terrain adaptive convolutional network Monte Carlo random deactivation model, which outputs the average probability of freezing rain and the standard deviation of uncertainty. Based on the average probability of freezing rain, standard deviation of uncertainty, 10-meter high wind speed, wet-bulb temperature, supercooled precipitation, preset time step, icing density and equivalent radius of the conductor, the Makkonen calibration model is input. The calibration parameters of the Makkonen calibration model are corrected and updated according to the average probability of freezing rain and standard deviation of uncertainty, and the predicted icing thickness and 90% confidence interval are output. Based on the preset area, predicted icing thickness, and preset conditions, the predicted icing area is determined.

[0007] In one embodiment, the atmospheric reanalysis dataset includes: relative humidity, specific humidity, temperature, and vertical velocity at 10 standard pressure layers from 400 hPa to 975 hPa; 2-meter high temperature, 2-meter high dew point temperature, surface pressure, total precipitation, east-west component of 10-meter high wind speed, north-south component of 10-meter high wind speed, total precipitation and precipitation type at a preset time step for China. The Chinese ground station meteorological observation dataset includes: ground temperature, dew point temperature, wind speed, altitude, total precipitation, relative humidity, and wet-bulb temperature observations at a preset time step; The topographic elevation dataset includes topographic elevation data for the Chinese region. The topographic feature tensor includes local slope, aspect, topographic openness, and windward slope index.

[0008] In one embodiment, preprocessing includes outlier removal, missing data imputation, and uniform spatiotemporal resolution; The preset conditions include: air temperature at a height of 2m is -3℃ to 0℃, relative humidity is ≥85%, supercooled precipitation exists, and icing frequency is ≥10%.

[0009] In one embodiment, the terrain-adaptive convolutional network Monte Carlo random deactivation model includes a terrain-adaptive convolutional network model, a Monte Carlo random deactivation model, and a model output layer; The terrain-adaptive convolutional network model includes an input layer, nine terrain-adaptive deformable offset learning modules, nine convolutional modules corresponding to the terrain-adaptive deformable offset learning modules, one global average pooling layer, one fully connected layer, and a convolutional output layer. Each convolutional module includes one 7×7 convolutional layer, one normalization layer, one Gaussian error linear activation layer, and one residual connection layer. The Monte Carlo random deactivation model consists of a drop-rate activation layer, a random deactivation layer, and a probability output layer.

[0010] In one embodiment, the freezing rain climate fusion dataset includes a freezing rain climate fusion training set and a freezing rain climate fusion validation set; The steps for training a terrain-adaptive convolutional network Monte Carlo random inactivation model include: Input the freezing rain climate fusion training set, topographic feature tensor, and icing element data; The first terrain-adaptive deformable offset learning module performs deformable convolution based on the terrain feature tensor and the terrain elevation data in the freezing rain climate fusion dataset, so that the deformable convolution sampling points are offset along the terrain contour line direction. The output of the first terrain adaptive deformable offset learning module is input into the corresponding first convolution module, and the first convolution module outputs the first residual value. The second terrain-adaptive deformable offset learning module performs deformable convolution again based on the terrain feature tensor and the terrain elevation data in the freezing rain climate fusion dataset, so that the deformable convolution sampling points are offset along the terrain contour line direction. The output of the second terrain adaptive deformable offset learning module is input into the corresponding second convolution module, and the second convolution module outputs the second residual value, until the ninth convolution module outputs the ninth residual value. The ninth residual value passes through a global average pooling layer and a fully connected layer, and the convolutional output layer outputs the freezing rain classification probability. The dropout rate activation layer receives the freezing rain classification probability and determines the dropout rate, which is positively correlated with the terrain complexity corresponding to the terrain feature tensor. Each time the random deactivation layer propagates forward, the random neurons in the random deactivation layer obtain the probability of freezing rain once based on the dropout rate. After a preset number of Monte Carlo samplings, a preset number of freezing rain probabilities are obtained. Based on the freezing rain probability of a preset number of times and the preset number of times, the probability output layer outputs the average probability of freezing rain; Based on the freezing rain probability of a preset number of times, the preset number of times, and the average freezing rain probability, the probability output layer outputs the uncertainty standard deviation. If the average probability of freezing rain is greater than or equal to the average probability threshold of freezing rain, it is determined that a freezing rain event has occurred in the preset area. The freezing rain climate fusion validation set is used to validate freezing rain events in the preset area until the accuracy of freezing rain events in the preset area in China is greater than or equal to the freezing rain event accuracy threshold, and / or until the total loss function value of the preset area in China is less than or equal to the total loss threshold. The model output layer outputs freezing rain event identifier, freezing rain average probability, uncertainty standard deviation and total precipitation.

[0011] In one embodiment, the correction parameters of the Makkonen correction model include the correction collision efficiency coefficient, the correction freezing coefficient, and the correction liquid water content; The correction parameters of the Makkonen correction model are updated based on the mean probability of freezing rain and the standard deviation of uncertainty, including: The initial liquid water content is determined by the formula based on the supercooled precipitation, the 10-meter high wind speed and the preset time step. Based on the initial liquid water content and the average probability of freezing rain, the corrected liquid water content is determined by the formula for calculating the corrected liquid water content. The initial collision efficiency coefficient is determined based on a 10-meter high wind speed using the initial collision efficiency coefficient calculation formula, and the initial freezing coefficient is determined based on a wet-bulb temperature using the initial freezing coefficient calculation formula. Determine the observation noise covariance based on the uncertainty standard deviation and scale factor; The Kalman gain is determined based on the prior mean of the uncertain standard deviation and the covariance of the observation noise. Based on the initial collision efficiency coefficient, Kalman gain, and prior mean of the collision efficiency coefficient, the posterior mean of the collision efficiency coefficient is determined until the number of times the posterior mean of the obtained collision efficiency coefficient is equal to the preset number of updates, and the corrected collision efficiency coefficient is obtained. Based on the initial freeze coefficient, Kalman gain, and prior mean of the freeze coefficient, the posterior mean of the freeze coefficient is determined until the number of times the posterior mean of the freeze coefficient is obtained equals the preset number of updates, thus obtaining the corrected freeze coefficient.

[0012] In one embodiment, the step of outputting the predicted icing thickness includes: Based on the corrected collision efficiency coefficient, corrected freezing coefficient, corrected liquid water content, 10-meter high wind speed, preset time step, icing density, and equivalent radius of the conductor, the predicted icing thickness is determined by the icing thickness correction calculation formula.

[0013] Secondly, this application provides a device for predicting icing areas, including: The acquisition module is used to acquire atmospheric reanalysis datasets, meteorological observation datasets from Chinese ground stations, and topographic elevation datasets. The preprocessing and filtering module is used to preprocess the atmospheric reanalysis dataset, meteorological observation data from Chinese ground stations, and topographic elevation data and filter icing elements to obtain a freezing rain climate fusion dataset based on a preset time step, extracted topographic feature tensors, and icing element data. The icing element data includes 2-meter high-temperature air, wet-bulb temperature, 10-meter high-wind speed, supercooled precipitation, relative humidity, and equivalent potential temperature of the 850 hPa pressure layer. The first model module is used to input the freezing rain climate fusion dataset, terrain feature tensor and icing element data into the pre-trained terrain adaptive convolutional network Monte Carlo random deactivation model, and output the average probability of freezing rain and the standard deviation of uncertainty. The second model module is used to input the Makkonen calibration model based on the average probability of freezing rain, the standard deviation of uncertainty, the 10-meter high wind speed, the wet-bulb temperature, the supercooled precipitation, the preset time step, the icing density, and the equivalent radius of the conductor. The calibration parameters of the Makkonen calibration model are corrected and updated according to the average probability of freezing rain and the standard deviation of uncertainty, and the output is the predicted icing thickness and the 90% confidence interval. The icing area prediction module is used to determine the predicted icing area based on a preset area, predicted icing thickness, and preset conditions.

[0014] Thirdly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method for predicting icing areas as described in any one of the first aspects above.

[0015] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for predicting icing areas as described in any one of the first aspects above.

[0016] Fifthly, this application provides a computer program product that, when run on an electronic device, causes the electronic device to perform the method described in any one of the first aspects above.

[0017] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect above, and will not be repeated here.

[0018] The advantages of this application compared to the prior art are: This application acquires atmospheric reanalysis datasets, meteorological observation datasets from Chinese ground stations, and topographic elevation datasets. It preprocesses these datasets and filters icing elements to obtain a frozen rain climate fusion dataset based on a preset time step, extracted topographic feature tensors, and icing element data. The icing element data includes 2-meter high-temperature air, wet-bulb temperature, 10-meter high-speed wind, supercooled precipitation, relative humidity, and equivalent potential temperature of the 850 hPa pressure layer. The frozen rain climate fusion dataset, topographic feature tensor, and icing element data are input into a pre-trained topographic adaptive convolutional network Monte Carlo random inactivation model, which outputs the average probability and uncertainty standard deviation of frozen rain. Based on the average probability, uncertainty standard deviation, 10-meter high-speed wind, wet-bulb temperature, and supercooled precipitation... The parameters of the ice accumulation, preset time step, ice density, and equivalent radius of the conductor are input into the Makkonen calibration model. The calibration parameters of the Makkonen calibration model are updated based on the average probability of freezing rain and the standard deviation of uncertainty. The output is the predicted ice thickness and a 90% confidence interval. Based on the preset region, the predicted ice thickness, and preset conditions, the predicted ice accumulation area is determined. Compared with existing technologies, the use of atmospheric reanalysis dataset, real meteorological observation data from Chinese ground stations, and topographic elevation data of the preset region for preprocessing and fusion filtering makes the ice accumulation simulation data more suitable for predicting ice thickness. At the same time, the use of a terrain-adaptive convolutional network Monte Carlo random deactivation model and the Makkonen calibration model improves the prediction accuracy of ice thickness in the preset region and also improves the accuracy of the predicted ice accumulation area. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the method for predicting icing areas provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of the terrain-adaptive convolutional network Monte Carlo random deactivation model provided in the embodiments of this application; Figure 3 This is a schematic diagram illustrating the process of updating and predicting ice thickness based on the average probability of freezing rain and the standard deviation of uncertainty of the Makkonen correction model provided in this application embodiment. Figure 4 This is a schematic diagram of the ice thickness distribution during a typical freezing rain event in 2008, provided in an embodiment of this application. Figure 5 This is a schematic diagram of the structure of the icing area prediction device provided in the embodiments of this application. Detailed Implementation

[0021] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0022] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0023] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0024] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0025] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0026] The current data foundation for icing simulation limits the accuracy and refinement of icing risk assessment. Icing data acquisition primarily relies on two sources: the first is discretely distributed ground meteorological station observation data, such as the station network of the China Meteorological Administration (CMA). While this type of data offers high single-point observation accuracy and reliability, its spatial representativeness is insufficient. Stations are typically deployed based on administrative divisions rather than power grid corridors, resulting in sparse distribution or even monitoring gaps in high-risk areas like mountainous regions and plateaus. This discrete and non-uniform distribution makes it impossible to continuously and completely depict the spatially heterogeneous distribution pattern of icing under complex terrain, failing to meet the urgent need of power grid companies for comprehensive gridded data for refined risk assessments at the transmission line and regional power grid scales. The second type is large-scale atmospheric reanalysis data generated based on numerical assimilation techniques, such as the ERA5 dataset (the fifth-generation atmospheric reanalysis dataset) from the European Centre for Medium-Range Weather Forecasts. This second type of data has the advantages of global coverage, spatiotemporal continuity, and complete variables, providing a valuable foundational field for meteorological research. However, when directly driving icing physics models such as the internationally accepted Makkonen model, the second type of data inherently suffers from systematic biases. First, atmospheric reanalysis data is not specifically assimilated and optimized for the specific and complex microphysical process of icing. The primary goal of the second type of data is to reproduce large-scale atmospheric circulation, limiting its ability to characterize key micrometeorological elements affecting near-surface icing formation, such as supercooled liquid water content and fine temperature stratification. Second, in physical models like the Makkonen model, many key parameters, such as raindrop collision efficiency coefficient and freezing coefficient, are typically set based on global averages or physical experiments in specific regions. China has a vast territory with extremely complex topography and climate types, especially the high mountains and hills of Southwest China and the hilly areas of Central China, where significant local microclimate effects exist. Directly applying globally accepted parameters inevitably leads to significant systematic biases in the simulation results regarding icing intensity, growth rate, and spatial distribution, potentially underestimating the threat of high-risk areas or misjudging icing boundaries.

[0027] In summary, relying solely on site observations presents a coverage dilemma of being partial but lacking comprehensive coverage, while directly applying reanalysis data raises questions about its applicability due to its limited scope and imprecise analysis. Simply performing spatial interpolation or statistical stitching between the two types of data cannot fundamentally resolve the core contradiction of mismatched physical mechanisms and missing regional characteristics. It is difficult to generate a high-quality gridded dataset that is both physically consistent and accurately reflects the icing characteristics of China, thus hindering the prediction of icing thickness in a predetermined region.

[0028] Furthermore, current models used to predict icing thickness have low accuracy for preset areas and cannot meet the needs of power system disaster prevention scenarios.

[0029] To address the aforementioned technical issues, the icing region prediction method of this application acquires an atmospheric reanalysis dataset, a meteorological observation dataset from Chinese ground stations, and a topographic elevation dataset. Preprocessing and icing element filtering are performed on the atmospheric reanalysis dataset, the meteorological observation data from Chinese ground stations, and the topographic elevation data to obtain a frozen rain climate fusion dataset based on a preset time step, extracted topographic feature tensors, and icing element data. The icing element data includes 2-meter high-temperature air, wet-bulb temperature, 10-meter high-wind speed, supercooled precipitation, relative humidity, and equivalent potential temperature of the 850 hPa pressure layer. The frozen rain climate fusion dataset, topographic feature tensor, and icing element data are input into a pre-trained topographic adaptive convolutional network Monte Carlo random inactivation model, which outputs the average probability and uncertainty standard deviation of frozen rain. Based on the average probability of frozen rain, the uncertainty standard deviation, and the 10-meter high-wind speed... The system inputs wet-bulb temperature, supercooled precipitation, preset time step, icing density, and equivalent radius of the conductor into the Makkonen calibration model. The calibration parameters of the Makkonen calibration model are updated based on the average probability of freezing rain and the standard deviation of uncertainty. The output is the predicted icing thickness and a 90% confidence interval. Based on the preset region, the predicted icing thickness, and preset conditions, the predicted icing area is determined. Compared with existing technologies, the use of atmospheric reanalysis datasets, real meteorological observation data from Chinese ground stations, and topographic elevation data of the preset region for preprocessing and fusion filtering makes the icing simulation data more suitable for predicting icing thickness. At the same time, the use of a terrain-adaptive convolutional network Monte Carlo random deactivation model and the Makkonen calibration model improves the prediction accuracy of icing thickness in the preset region and also improves the accuracy of the predicted icing area.

[0030] The technical solution of this application will be described below through specific embodiments.

[0031] Firstly, such as Figure 1 As shown, this application provides a method for predicting icing areas, including: S1, acquires atmospheric reanalysis dataset, meteorological observation dataset from Chinese ground stations, and topographic elevation dataset.

[0032] In one embodiment, the fifth-generation atmospheric reanalysis dataset (ERA5) is obtained from the climate data repository of the European Centre for Medium-Range Weather Forecasts (ECMWF). The ERA5 dataset includes: relative humidity, specific humidity, temperature, and vertical velocity at 10 standard pressure levels from 400 hPa to 975 hPa; 2-meter high temperature, 2-meter high dew point temperature, surface pressure, total precipitation (decumulated), east-west component of 10-meter high wind speed, and north-south component of 10-meter high wind speed at a preset time step for land over China; and total precipitation and precipitation type at a single level (i.e., near-surface) for a preset time step, using xarray and other methods. The library reads and converts downloaded NetCDF format files into structured data. Specifically, temperature, relative humidity, and specific humidity at 10 standard pressure layers between 400 hPa and 975 hPa are used to calculate the equivalent potential temperature at 850 hPa to characterize large-scale conditions of the frontal background. Relative humidity serves as a feature input for upper-level humidity condition analysis. Vertical velocity is used to analyze atmospheric rising and sinking motion to determine the dynamic conditions for freezing rain formation. Accumulated total precipitation is used for subsequent supercooled precipitation conversion. Total precipitation is combined with freezing rain identification results to convert it into supercooled precipitation. Precipitation type is used as an auxiliary feature input model to distinguish between rain, snow, and freezing rain. The China Meteorological Administration (CMA) surface station meteorological observation dataset includes: ground temperature, dew point temperature, wind speed, altitude, total precipitation, relative humidity, and wet-bulb temperature observations at a preset time step; the topographic elevation dataset includes: topographic elevation data for the Chinese region. The preset time step is any one of 1h, 2h, 3h, 4h, 5h, or 6h; ground temperature is used to construct freezing rain event labels (i.e., freezing rain occurrence identifiers observed by stations) and to filter preset conditions; wind speed is used to verify the accuracy of ERA5 wind speed data and as a model input feature; altitude is used to fuse with DEM topographic data to verify topographic accuracy; total precipitation is used to verify the accuracy of ERA5 precipitation and to assist in freezing rain identification; relative humidity is used to verify the accuracy of ERA5 relative humidity and to participate in the filtering of preset conditions.

[0033] S2 preprocesses the atmospheric reanalysis dataset, meteorological observation data from Chinese ground stations, and topographic elevation data, and filters icing elements to obtain a frozen rain climate fusion dataset based on a preset time step, as well as extracted topographic feature tensors and icing element data.

[0034] In one embodiment, preprocessing includes outlier removal, missing data imputation, and unifying spatiotemporal resolution. Specifically, unifying spatiotemporal resolution involves removing outliers from the original data using extreme value tests and consistency tests (such as the 3σ principle), unifying the spatial resolution between CMA station data and ERA5 grid data using interpolation, and aligning the preset time steps of all data to construct a spatiotemporally matched fusion dataset. For example, the spatial resolution is 0.1°×0.1° or 0.5°×0.5°, and the preset time step is 6 hours.

[0035] In one embodiment, icing data includes 2-meter altitude air temperature, wet-bulb temperature, 10-meter altitude wind speed, supercooled precipitation, relative humidity, and equivalent potential temperature of the 850 hPa pressure layer. Specifically, 2-meter altitude air temperature... Temperature range: -6℃ to 0℃; wet-bulb temperature range: -4℃ to 2℃. Based on the temperature at a height of 2 meters The calculation is based on relative humidity (RH) and surface air pressure; the wind speed (WS) at a height of 10 meters is 1 m / s to 8 m / s, calculated using the formula... Obtain; supercooled precipitation >0.1mm / 6h, converted from freezing rain identification results to total precipitation (TP).

[0036] Among them, the relative humidity (RH) is greater than 85%, based on the temperature at a height of 2 meters. Given the air temperature T (i.e., the dew point temperature at 2 meters), the relative humidity RH is obtained using the Magnus formula, which is as follows:

[0037]

[0038] in, The saturated vapor pressure, This is the actual water vapor pressure. , It is a constant. It is a natural exponential function.

[0039] Among them, the equivalent potential temperature of the 850 hPa pressure layer characterizes the large-scale conditions of the frontal background, and the calculation formula is:

[0040] in, for P =Equivalent potential temperature of the 850 hPa pressure layer; Temperature in Kelvin (K) at the pressure level; This is the gas constant for dry air, for example, 287 J / (kg·K); For constant pressure specific heat, for example, 1004 J / (kg·K); For example, the latent heat of vaporization J / (kg·K); The water vapor mixing ratio (the mass of water vapor per unit mass of dry air, expressed in kg / kg) is represented by the specific humidity variable data in the ERA5 downscaling data.

[0041] In one embodiment, the terrain feature tensor includes local slope, aspect, terrain openness, and windward slope index, used to determine terrain complexity. The local slope S ranges from 0° to 90°, and the aspect A ranges from 0° to 360°, but is generally converted to the aspect change rate. Topographical openness The smaller the value, the more complex the terrain, and the higher the windward slope index. The larger the absolute value, the more intense the airflow lift and the higher the complexity.

[0042] S3 inputs the freezing rain climate fusion dataset, topographic feature tensor, and icing element data into a pre-trained topographic adaptive convolutional network Monte Carlo random deactivation model, and outputs the average probability of freezing rain and the standard deviation of uncertainty.

[0043] In one embodiment, such as Figure 2 As shown, the terrain-adaptive convolutional network Monte Carlo random deactivation model (TA-ConvNeXt / MC-Dropout) includes a terrain-adaptive convolutional network model (TA-ConvNeXt), a Monte Carlo random deactivation model (MC-Dropout), and a model output layer. The terrain-adaptive convolutional network model includes an input layer, nine terrain-adaptive deformable offset learning modules, nine convolutional modules corresponding to the terrain-adaptive deformable offset learning modules, one global average pooling layer, one fully connected layer, and a convolutional output layer. Each convolutional module includes one 7×7 convolutional layer, one normalization layer, one Gaussian error linear activation layer, and one residual connection layer. The Monte Carlo random deactivation model includes a dropout rate activation layer, a random deactivation layer, and a probability output layer. The model output layer... The average probability and standard deviation of freezing rain are calculated. Because a terrain-adaptive deformable offset learning module is inserted before each convolutional module, the convolutional sampling points are offset along the terrain contour lines, thus improving the adaptability of the terrain-adaptive convolutional network model to various terrains, increasing the accuracy of freezing rain probability, and consequently improving the prediction accuracy of ice thickness. Furthermore, the higher the terrain complexity (e.g., steep mountains, canyons, abrupt changes in windward slopes), the greater the uncertainty of the Monte Carlo random deactivation model of the terrain-adaptive convolutional network. Since the dropout rate of the Monte Carlo random deactivation model is positively correlated with the terrain complexity corresponding to the terrain feature tensor, the accuracy of the average probability and standard deviation of freezing rain output by the terrain-adaptive convolutional network Monte Carlo random deactivation model is improved, further enhancing the prediction accuracy of ice thickness.

[0044] In one embodiment, the freezing rain climate fusion dataset includes a freezing rain climate fusion training set and a freezing rain climate fusion validation set.

[0045] In one embodiment, the steps of training a terrain-adaptive convolutional network Monte Carlo stochastic inactivation model include: S30, input freezing rain climate fusion training set, topographic feature tensor and icing element data.

[0046] S31, the first terrain adaptive deformable offset learning module performs deformable convolution based on the terrain feature tensor and the terrain elevation data in the freezing rain climate fusion dataset, so that the deformable convolution sampling points are offset along the terrain contour line direction.

[0047] S32, the output of the first terrain adaptive deformable offset learning module is input into the corresponding first convolution module, and the first convolution module outputs the first residual value.

[0048] Specifically, the formula for calculating the residual value is:

[0049] in, These are the input values ​​for the convolution module. The output value of the convolution module. It is the residual mapping function, which includes the mapping of the convolution kernel, normalization function and Gaussian error linear activation function of the convolutional layer.

[0050] S33, the second terrain adaptive deformable offset learning module performs deformable convolution again based on the terrain feature tensor and the terrain elevation data in the freezing rain climate fusion dataset, so that the deformable convolution sampling points are offset along the terrain contour line direction.

[0051] S34, the output of the second terrain adaptive deformable offset learning module is input into the corresponding second convolution module, the second convolution module outputs the second residual value, until the ninth convolution module outputs the ninth residual value.

[0052] By setting up a 9-layer terrain-adaptive deformable offset learning module and corresponding 9-layer convolutional modules, it can fully capture the spatial and geographical continuous features of the meteorological field, thereby obtaining a more accurate freezing rain classification probability.

[0053] S35, the ninth residual value passes through a global average pooling layer and a fully connected layer, and the convolutional output layer outputs the freezing rain classification probability.

[0054] S36, the dropout rate activation layer receives the freezing rain classification probability and determines the dropout rate, which is positively correlated with the terrain complexity corresponding to the terrain feature tensor.

[0055] S37, each time the random deactivation layer propagates forward, the random neurons of the random deactivation layer obtain the probability of freezing rain once based on the dropout rate. After a preset number of Monte Carlo samplings, a preset number of freezing rain probabilities are obtained. S38, based on the freezing rain probability of a preset number of times and the preset number of times, the probability output layer outputs the average probability of freezing rain.

[0056] In one implementation, the average probability of freezing rain is calculated as follows:

[0057] in, The average probability of freezing rain reflects the likelihood that supercooled water droplets will arrive and freeze under current meteorological conditions. The number of random forward propagation samples. For the first i The probability of freezing rain in a random forward propagation sample. Input for each spatial grid point.

[0058] It should be noted that the input for each spatial grid point includes: icing element data, 2-meter dew point temperature, surface air pressure, total precipitation, precipitation type, temperature, relative humidity, specific humidity, vertical velocity, and topographic feature tensor for 10 standard pressure layers from 400hPa to 975hPa.

[0059] S39, based on the freezing rain probability of a preset number of times, the preset number of times, and the average freezing rain probability, the probability output layer outputs the uncertainty standard deviation.

[0060] In one implementation, the standard deviation of uncertainty is calculated as follows:

[0061] in, The standard deviation of uncertainty is denoted as when the input x is removed. , This represents the average probability of freezing rain. The number of random forward propagation samples. For the first i The probability of freezing rain in a random forward propagation sample. Input for each spatial grid point.

[0062] S310, if the average probability of freezing rain is greater than or equal to the average probability threshold of freezing rain, it is determined that a freezing rain event has occurred in the preset area.

[0063] In one embodiment, the average probability threshold for freezing rain is greater than or equal to 0.5. For example, when the average probability of freezing rain is 0.5, it is determined that a freezing rain event has occurred in the preset area.

[0064] S311 uses the freezing rain climate fusion validation set to validate freezing rain events in the preset area until the accuracy of freezing rain events in the preset area in China is greater than or equal to the freezing rain event accuracy threshold, and / or until the total loss function value of the preset area in China is less than or equal to the total loss threshold. The model output layer outputs the freezing rain event identifier, the average probability of freezing rain, the standard deviation of uncertainty, and the total precipitation.

[0065] In one embodiment, the accuracy threshold for freezing rain events is greater than or equal to 90%.

[0066] In one embodiment, the loss function is calculated as follows:

[0067] in, This is the total loss function value. Classify the loss values ​​for freezing rain events. The loss weight is determined by the temperature physical constraint. This represents the temperature-related physical constraint loss value. For terrain-constrained loss weights, This represents the terrain constraint loss value. .

[0068] It should be noted that the temperature physical constraint loss value represents the upper limit penalty imposed on the predicted probability of freezing rain events based on the temperature at 2 meters. When the temperature at 2 meters is greater than 0°, the probability of freezing rain is forcibly constrained to approach 0. The terrain constraint loss value represents the spatial gradient of the freezing rain probability, which is consistent with the terrain elevation gradient, so that the model prediction results conform to the spatial continuity characteristics of freezing rain distribution under complex terrain.

[0069] In one implementation, the terrain complexity calculation formula is:

[0070] in, Due to terrain complexity, For slope, For the maximum slope, The slope aspect change rate, For the openness of the terrain, The windward slope index and the local terrain complexity calculation formula can quickly calculate terrain complexity, improving calculation efficiency.

[0071] In another embodiment, the terrain complexity calculation formula can also be:

[0072] in, Due to terrain complexity, For slope, The standard deviation is circular in aspect. For the openness of the terrain, As the windward slope index, the local topographic complexity calculation formula assigns weights to the topographic feature tensors of each region, improving the accuracy of topographic complexity.

[0073] The formula for calculating the discard rate is:

[0074] in, For the discard rate, Due to terrain complexity, The base drop rate for flat terrain, for example, is 0.1. This represents the maximum drop rate for the most complex terrain, for example, 0.5.

[0075] S4, based on the average probability of freezing rain, standard deviation of uncertainty, 10-meter high wind speed, wet-bulb temperature, supercooled precipitation, preset time step, icing density and equivalent radius of the conductor, is input into the Makkonen calibration model. The calibration parameters of the Makkonen calibration model are corrected and updated according to the average probability of freezing rain and standard deviation of uncertainty, and the predicted icing thickness and 90% confidence interval are output.

[0076] In one embodiment, the correction parameters of the Makkonen correction model include the correction collision efficiency coefficient, the correction freezing coefficient, and the correction liquid water content.

[0077] In one embodiment, such as Figure 3 As shown, the correction parameters of the Makkonen correction model are updated based on the mean probability of freezing rain and the standard deviation of uncertainty, including: S41 determines the initial liquid water content based on the supercooled precipitation, 10-meter high wind speed, and preset time step using the initial liquid water content calculation formula.

[0078] In one embodiment, the initial liquid water content is calculated as follows:

[0079] in, This represents the initial liquid water content. For supercooled precipitation, A wind speed of 10 meters per second. This is the preset time step.

[0080] S42, based on the initial liquid water content and the average probability of freezing rain, the corrected liquid water content is determined by the formula for calculating the corrected liquid water content.

[0081] In one embodiment, the formula for calculating the corrected liquid water content is:

[0082] in, To correct the liquid water content, This represents the initial liquid water content. This represents the average probability of freezing rain. The input is for each spatial grid point. The corrected liquid water content in this embodiment indicates that even if liquid water is present, the actual liquid water content contributing to icing should be reduced if the temperature does not meet the freezing rain conditions (e.g., above 0°C).

[0083] In another embodiment, the formula for calculating the corrected liquid water content can also be:

[0084] in, To correct the liquid water content, This represents the initial liquid water content. This represents the average probability of freezing rain. For each spatial grid point, the input, This is a safety factor (e.g., 1.645 corresponds to the lower limit of a 90% confidence interval). The standard deviation represents the uncertainty. In this embodiment, when the standard deviation of uncertainty for corrected liquid water content is relatively high, the lower bound of the confidence interval of the probability is used for estimation, thereby achieving risk avoidance for high uncertainty.

[0085] S43, the initial collision efficiency coefficient is determined by the initial collision efficiency coefficient calculation formula based on a 10-meter high wind speed, and the initial freezing coefficient is determined by the initial freezing coefficient calculation formula based on the wet-bulb temperature.

[0086] In one embodiment, the initial collision efficiency coefficient is calculated as follows:

[0087] in, This represents the initial collision efficiency coefficient. A wind speed of 10 meters per second. It is a natural exponential function.

[0088] In one embodiment, the initial freezing coefficient is calculated as follows:

[0089] in, This is the initial freezing coefficient. This refers to the wet-bulb temperature, which ranges from -3℃ to 0℃.

[0090] S44, determine the observation noise covariance based on the uncertainty standard deviation and scale factor.

[0091] In one embodiment, the formula for calculating the observation noise covariance is:

[0092] in, To observe the noise covariance, As a scale factor, The standard deviation of uncertainty is used; that is, the uncertainty of the average probability of freezing rain is mapped to the observation noise covariance through the calculation formula of the observation noise covariance, which facilitates the updating of the correction parameters.

[0093] S45, determine the Kalman gain based on the prior mean of the uncertain standard deviation and the covariance of the observation noise.

[0094] In one embodiment, the Kalman gain is calculated as follows:

[0095] in, For Kalman gain, Let be the prior mean of the uncertainty bias. To observe the noise covariance.

[0096] S46. Based on the initial collision efficiency coefficient, Kalman gain, and prior mean of the collision efficiency coefficient, determine the posterior mean of the collision efficiency coefficient until the number of times the posterior mean of the obtained collision efficiency coefficient is equal to the preset number of updates, and obtain the corrected collision efficiency coefficient.

[0097] In one embodiment, the formula for updating the corrected collision efficiency coefficient is:

[0098] in, To correct the collision efficiency coefficient (that is, the posterior mean of the collision efficiency coefficient). Let be the prior mean of the collision efficiency coefficient. For Kalman gain, This represents the initial collision efficiency coefficient.

[0099] S47. Based on the initial freeze coefficient, Kalman gain, and prior mean of the freeze coefficient, determine the posterior mean of the freeze coefficient until the number of times the posterior mean of the freeze coefficient is obtained equals the preset number of updates, and then obtain the corrected freeze coefficient.

[0100] In one embodiment, the formula for updating the correction freeze coefficient is:

[0101] in, To correct for the freezing coefficient (i.e., the posterior mean of the freezing coefficient). Let be the prior mean of the freezing coefficient. For Kalman gain, This is the initial freezing coefficient.

[0102] In one embodiment, the step of outputting the predicted icing thickness includes: determining the predicted icing thickness using an icing thickness correction formula based on the corrected collision efficiency coefficient, corrected freezing coefficient, corrected liquid water content, 10-meter high wind speed, preset time step, icing density, and equivalent radius of the conductor.

[0103] In one embodiment, the formula for calculating the icing thickness correction is:

[0104] in, For ice thickness, To correct the collision efficiency coefficient, To correct the freezing coefficient, To correct the liquid water content, A wind speed of 10 meters per second. To preset the time step, The density of the ice layer. Let be the equivalent radius of the conductor.

[0105] S5 determines the predicted icing area based on the preset area, predicted icing thickness, and preset conditions.

[0106] In one embodiment, the preset conditions include: a 2m altitude temperature of -3℃ to 0℃, relative humidity ≥85%, supercooled precipitation, and an icing frequency ≥10%. The icing frequency represents the ratio of the number of days with freezing rain events and icing thickness ≥1mm to the total number of days that meet the conditions of a 2m altitude temperature of -3℃ to 0℃, relative humidity ≥85%, and supercooled precipitation. For example, if the total number of days meeting the conditions of a 2m altitude temperature of -3℃ to 0℃, relative humidity ≥85%, and supercooled precipitation is 100 days, and 10 of these days have an average probability of freezing rain greater than or equal to 50% and an icing thickness ≥1mm, then these 10 days are considered to have freezing rain events and icing thickness ≥1mm, and therefore the icing frequency is 10%.

[0107] In one specific embodiment, atmospheric reanalysis dataset, meteorological observation dataset from Chinese ground stations, and topographic elevation dataset from 2008 are extracted. The spatial resolution of each grid point is 0.5° × 0.5°, and the preset time step is 6 hours. After obtaining the predicted icing thickness for each grid point, grid points meeting preset conditions are selected based on 2m altitude air temperature, relative humidity, presence of supercooled precipitation, and whether the icing frequency is greater than or equal to 10%. These grid points meeting the preset conditions are then used to generate an icing thickness distribution map of typical freezing rain events in 2008, based on the icing frequency. Figure 4 As shown. Figure 4An inversion and analysis of the major freezing rain disaster that occurred in central and eastern China during the winter of 2008 clearly revealed the spatial distribution pattern of the freezing rain impact, the approximate range of icing thickness, and the location of high-risk areas. The main affected areas were concentrated in mountainous and hilly areas with complex terrain and high altitudes, such as Guizhou, Hunan, Jiangxi, southern Hubei, and southern Anhui, which highly coincides with historical disaster records. Through the visualization of typical cases in this embodiment, not only is the reliability and physical rationality of the icing area prediction method of this application verified in reconstructing historical freezing rain icing events verified, but it also provides an intuitive reference for understanding the spatial distribution pattern of freezing rain-type icing and identifying key areas for power grid disaster prevention.

[0108] The method in this application is not a simple stacking of existing data sources, but a deep and organic fusion. It can effectively integrate high-precision point information from station observations with full coverage information from reanalysis data. On this basis, it closely combines the regional physical mechanisms of icing formation with localized observational knowledge. Through the synergy of intelligent algorithms and physical models, it achieves parameter correction and process optimization of the modified model. In the end, it constructs an icing grid dataset with clear physical meaning, high spatiotemporal resolution, strong applicability and high reliability in the context of China's complex underlying surface, thereby improving the accuracy of icing area prediction.

[0109] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0110] Secondly, such as Figure 5 As shown, this application provides a device 100 for predicting icing areas, comprising: The acquisition module 110 is used to acquire atmospheric reanalysis datasets, meteorological observation datasets from Chinese ground stations, and topographic elevation datasets.

[0111] The preprocessing and filtering module 120 is used to preprocess the atmospheric reanalysis dataset, meteorological observation data from Chinese ground stations, and topographic elevation data and filter icing elements to obtain a freezing rain climate fusion dataset based on a preset time step, extracted topographic feature tensors, and icing element data. The icing element data includes 2-meter high-temperature air, wet-bulb temperature, 10-meter high-wind speed, supercooled precipitation, relative humidity, and equivalent potential temperature of the 850 hPa pressure layer.

[0112] The first model module 130 is used to input the freezing rain climate fusion dataset, terrain feature tensor and icing element data into the pre-trained terrain adaptive convolutional network Monte Carlo random inactivation model, and output the average probability of freezing rain and the standard deviation of uncertainty.

[0113] The second model module 140 is used to input the Makkonen calibration model based on the average probability of freezing rain, the standard deviation of uncertainty, the 10-meter high wind speed, the wet-bulb temperature, the supercooled precipitation, the preset time step, the icing density, and the equivalent radius of the conductor. The calibration parameters of the Makkonen calibration model are corrected and updated according to the average probability of freezing rain and the standard deviation of uncertainty, and the predicted icing thickness and the 90% confidence interval are output.

[0114] The icing area prediction module 150 is used to determine the predicted icing area based on a preset area, the predicted icing thickness, and preset conditions.

[0115] It should be noted that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0116] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0117] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0118] The method for predicting icing areas provided in this application can be applied to terminal devices such as mobile phones, tablets, wearable devices, in-vehicle devices, augmented reality (AR) / virtual reality (VR) devices, laptops, ultra-mobile personal computers (UMPCs), netbooks, and personal digital assistants (PDAs). This application does not impose any restrictions on the specific type of terminal device.

[0119] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms.

[0120] The computer-readable medium may include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, recording media, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, portable hard drives, magnetic disks, or optical discs.

[0121] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0122] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0123] In the embodiments provided in this application, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0124] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0125] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method of predicting an icing area, characterized by, include: Acquire atmospheric reanalysis dataset, meteorological observation dataset from Chinese ground stations, and topographic elevation dataset; Preprocessing and icing element screening were performed on atmospheric reanalysis dataset, meteorological observation data from Chinese ground stations, and topographic elevation data to obtain a freezing rain climate fusion dataset based on a preset time step, extracted topographic feature tensor, and icing element data. The icing element data included 2-meter high-temperature air, wet-bulb temperature, 10-meter high-wind speed, supercooled precipitation, relative humidity, and equivalent potential temperature of the 850 hPa pressure layer. The freezing rain climate fusion dataset, topographic feature tensor, and icing element data are input into a pre-trained terrain adaptive convolutional network Monte Carlo random deactivation model, which outputs the average probability of freezing rain and the standard deviation of uncertainty. Based on the average probability of freezing rain, standard deviation of uncertainty, 10-meter high wind speed, wet-bulb temperature, supercooled precipitation, preset time step, icing density, and equivalent radius of the conductor, the Makkonen calibration model is input. The calibration parameters of the Makkonen calibration model are corrected and updated according to the average probability of freezing rain and standard deviation of uncertainty, and the predicted icing thickness and 90% confidence interval are output. Based on the preset area, predicted icing thickness, and preset conditions, the predicted icing area is determined.

2. The ice accretion area prediction method of claim 1, wherein The atmospheric reanalysis dataset includes: relative humidity, specific humidity, temperature, and vertical velocity at 10 standard pressure levels from 400 hPa to 975 hPa; 2-meter upper temperature, 2-meter upper dew point temperature, surface pressure, total precipitation, east-west component of 10-meter upper wind speed, north-south component of 10-meter upper wind speed, total precipitation and precipitation type at a preset time step for China. The Chinese ground station meteorological observation dataset includes: ground temperature, dew point temperature, wind speed, altitude, total precipitation, relative humidity, and wet-bulb temperature observations at a preset time step; The topographic elevation dataset includes topographic elevation data for the Chinese region. The topographic feature tensor includes local slope, aspect, topographic openness, and windward slope index.

3. The ice accretion area prediction method of claim 1, wherein Preprocessing includes outlier removal, missing data imputation, and uniform spatiotemporal resolution; The preset conditions include: air temperature at a height of 2m is -3℃ to 0℃, relative humidity is ≥85%, supercooled precipitation exists, and icing frequency is ≥10%.

4. The ice accretion area prediction method of claim 1, wherein The terrain-adaptive convolutional network Monte Carlo random deactivation model includes a terrain-adaptive convolutional network model, a Monte Carlo random deactivation model, and a model output layer; The terrain-adaptive convolutional network model includes an input layer, nine terrain-adaptive deformable offset learning modules, nine convolutional modules corresponding to the terrain-adaptive deformable offset learning modules, one global average pooling layer, one fully connected layer, and a convolutional output layer. Each convolutional module includes one 7×7 convolutional layer, one normalization layer, one Gaussian error linear activation layer, and one residual connection layer. The Monte Carlo random deactivation model consists of a drop-rate activation layer, a random deactivation layer, and a probability output layer.

5. The ice accretion area prediction method of claim 4, wherein The freezing rain climate fusion dataset includes a freezing rain climate fusion training set and a freezing rain climate fusion validation set; The steps for training a terrain-adaptive convolutional network Monte Carlo random inactivation model include: Input the freezing rain climate fusion training set, topographic feature tensor, and icing element data; The first terrain-adaptive deformable offset learning module performs deformable convolution based on the terrain feature tensor and the terrain elevation data in the freezing rain climate fusion dataset, so that the deformable convolution sampling points are offset along the terrain contour line direction. The output of the first terrain adaptive deformable offset learning module is input into the corresponding first convolution module, and the first convolution module outputs the first residual value. The second terrain-adaptive deformable offset learning module performs deformable convolution again based on the terrain feature tensor and the terrain elevation data in the freezing rain climate fusion dataset, so that the deformable convolution sampling points are offset along the terrain contour line direction. The output of the second terrain adaptive deformable offset learning module is input into the corresponding second convolution module, and the second convolution module outputs the second residual value, until the ninth convolution module outputs the ninth residual value. The ninth residual value passes through a global average pooling layer and a fully connected layer, and the convolutional output layer outputs the freezing rain classification probability. The dropout rate activation layer receives the freezing rain classification probability and determines the dropout rate, which is positively correlated with the terrain complexity corresponding to the terrain feature tensor. Each time the random deactivation layer propagates forward, the random neurons in the random deactivation layer obtain the probability of freezing rain once based on the dropout rate. After a preset number of Monte Carlo samplings, a preset number of freezing rain probabilities are obtained. Based on the freezing rain probability of a preset number of times and the preset number of times, the probability output layer outputs the average probability of freezing rain; Based on the freezing rain probability of a preset number of times, the preset number of times, and the average freezing rain probability, the probability output layer outputs the uncertainty standard deviation. If the average probability of freezing rain is greater than or equal to the average probability threshold of freezing rain, it is determined that a freezing rain event has occurred in the preset area. The freezing rain climate fusion validation set is used to validate freezing rain events in the preset area until the accuracy of freezing rain events in the preset area in China is greater than or equal to the freezing rain event accuracy threshold, and / or until the total loss function value of the preset area in China is less than or equal to the total loss threshold. The model output layer outputs freezing rain event identifier, freezing rain average probability, uncertainty standard deviation and total precipitation.

6. The ice accretion area prediction method of claim 1, wherein The correction parameters of the Makkonen correction model include the correction collision efficiency coefficient, the correction freezing coefficient, and the correction liquid water content; The correction parameters of the Makkonen correction model are updated based on the mean probability of freezing rain and the standard deviation of uncertainty, including: The initial liquid water content is determined by the formula based on the supercooled precipitation, the 10-meter high wind speed and the preset time step. Based on the initial liquid water content and the average probability of freezing rain, the corrected liquid water content is determined by the formula for calculating the corrected liquid water content. The initial collision efficiency coefficient is determined based on a 10-meter high wind speed using the initial collision efficiency coefficient calculation formula, and the initial freezing coefficient is determined based on a wet-bulb temperature using the initial freezing coefficient calculation formula. Determine the observation noise covariance based on the uncertainty standard deviation and scale factor; The Kalman gain is determined based on the prior mean of the uncertain standard deviation and the covariance of the observation noise. Based on the initial collision efficiency coefficient, Kalman gain, and prior mean of the collision efficiency coefficient, the posterior mean of the collision efficiency coefficient is determined until the number of times the posterior mean of the obtained collision efficiency coefficient is equal to the preset number of updates, and the corrected collision efficiency coefficient is obtained. Based on the initial freeze coefficient, Kalman gain, and prior mean of the freeze coefficient, the posterior mean of the freeze coefficient is determined until the number of times the posterior mean of the freeze coefficient is obtained equals the preset number of updates, thus obtaining the corrected freeze coefficient.

7. The ice accretion area prediction method of claim 6, wherein The steps for outputting the predicted icing thickness include: Based on the corrected collision efficiency coefficient, corrected freezing coefficient, corrected liquid water content, 10-meter high wind speed, preset time step, icing density, and equivalent radius of the conductor, the predicted icing thickness is determined by the icing thickness correction calculation formula.

8. The ice accretion area prediction device characterized by, include: The acquisition module is used to acquire atmospheric reanalysis datasets, meteorological observation datasets from Chinese ground stations, and topographic elevation datasets. The preprocessing and filtering module is used to preprocess the atmospheric reanalysis dataset, meteorological observation data from Chinese ground stations, and topographic elevation data and filter icing elements to obtain a freezing rain climate fusion dataset based on a preset time step, extracted topographic feature tensors, and icing element data. The icing element data includes 2-meter high-temperature air, wet-bulb temperature, 10-meter high-wind speed, supercooled precipitation, relative humidity, and equivalent potential temperature of the 850 hPa pressure layer. The first model module is used to input the freezing rain climate fusion dataset, terrain feature tensor and icing element data into the pre-trained terrain adaptive convolutional network Monte Carlo random deactivation model, and output the average probability of freezing rain and the standard deviation of uncertainty. The second model module is used to input the Makkonen calibration model based on the average probability of freezing rain, the standard deviation of uncertainty, the 10-meter high wind speed, the wet-bulb temperature, the supercooled precipitation, the preset time step, the icing density, and the equivalent radius of the conductor. The calibration parameters of the Makkonen calibration model are corrected and updated according to the average probability of freezing rain and the standard deviation of uncertainty, and the output is the predicted icing thickness and the 90% confidence interval. The icing area prediction module is used to determine the predicted icing area based on a preset area, predicted icing thickness, and preset conditions.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for predicting icing areas as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for predicting icing areas as described in any one of claims 1 to 7.

Images