A method, system, device and storage medium for predicting changes in icing areas

Abstract

The application discloses an icing area change prediction method, system, device and storage medium, applied to the icing area change prediction field, comprising the following steps: acquiring three types of time series data of sea ice density, polar vortex characteristics and covariants, fusing the covariants and the sea ice density to obtain a first associated sequence, analyzing the polar vortex characteristics under the constraint condition of the sequence to obtain polar vortex state characteristics; simultaneously, determining icing area grid data according to air temperature and precipitation data; finally, inputting the polar vortex state characteristics and the icing area grid data into a visual Transformer network to output an icing area distribution prediction result. The icing area change prediction method provided by the application combines large-scale climate factors and near-surface meteorological conditions, so that the icing area prediction is no longer dependent on the historical statistical extrapolation of a single meteorological element, and the accuracy, timeliness and spatial coverage capability of the prediction are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of icing zone change prediction technology, and in particular to a method, system, device and storage medium for predicting icing zone changes. Background Technology

[0002] In fields such as power system disaster prevention and mitigation, transportation line operation and maintenance, and communication facility protection, the prediction of changes in the extent of icing zones is an important basis for formulating disaster prevention plans and dispatching decisions. The formation and evolution of icing zones are directly affected by near-surface meteorological conditions such as temperature and precipitation, and are also indirectly regulated by anomalous activities of large-scale climate systems.

[0003] In existing technologies, icing zone prediction methods mainly rely on threshold judgments of single meteorological elements. The core idea is to establish icing determination models using near-surface meteorological parameters such as temperature, precipitation, and wind speed, and extrapolate future icing trends based on statistical patterns in historical data. However, these methods generally neglect the indirect impact of the intrinsic correlations between various climatic factors in the atmospheric circulation system on icing zone changes. For example, climate variability in high-latitude regions affects the circulation configuration in mid- and low-latitude regions through atmospheric teleconnection processes, thereby altering the hydrothermal conditions required for icing formation. This problem makes it difficult for existing prediction methods to explain and predict the trend of icing zones expanding into new areas, and also fails to accurately capture the annual fluctuations in icing zone extent, thus limiting the accuracy and timeliness of icing zone change predictions. Summary of the Invention

[0004] This invention provides a method, system, device, and storage medium for predicting changes in icing zones, in order to solve the technical problem that it is difficult to accurately predict the trend of changes in icing zones due to neglecting the intrinsic correlation between various climate factors in the atmospheric circulation system, thereby improving the accuracy, timeliness, and spatial coverage of icing zone change predictions.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for predicting changes in icing zones, comprising: Acquire time-series data on sea ice concentration, polar eddy characteristics, and covariates for the region to be predicted. By fusing the covariate time series data with the sea ice concentration time series data, a first correlation sequence characterizing the evolution of polar eddies under covariate constraints is obtained; by fusing the covariate time series data with the polar eddy characteristic time series data, a second correlation sequence is obtained. Using the first correlation sequence as a constraint, the second correlation sequence is analyzed to obtain the polar vortex state characteristics; Obtain meteorological time-series data of the area to be predicted, and determine the grid data of the icing area based on the meteorological time-series data; The polar vortex state characteristics and the icing zone grid data are analyzed to obtain the icing zone distribution prediction results.

[0006] As one preferred embodiment, the covariate time-series data includes Arctic Oscillation time-series data and Siberian High time-series data. The process of fusing the covariate time-series data with the sea ice concentration time-series data to obtain a first correlation sequence characterizing the evolution of the polar eddy under covariate constraints includes: The Arctic Oscillation time series data and the Siberian High pressure time series data are obtained as the covariate time series data; The Arctic Oscillation time series data, the Siberian High time series data, and the sea ice concentration time series data are time-aligned, and the aligned data are then spliced ​​together to obtain the first associated sequence.

[0007] As one preferred embodiment, the step of analyzing the second correlation sequence using the first correlation sequence as a constraint to obtain polar vortex state characteristics includes: The first associated sequence number is input into the first recurrent neural network to obtain the environmental constraint features; The second associated sequence is input into the first recurrent neural network to obtain the historical evolution characteristics of the polar vortex. The polar vortex state features are generated by fusing the environmental constraint features and the polar vortex historical evolution features through an attention mechanism.

[0008] As one preferred embodiment, the polar vortex state characteristics are obtained through bidirectional constraint training, including: The second associated sequence and the sea ice concentration time series data are input into the second recurrent neural network to obtain the predicted sea ice concentration sequence; With the goal of minimizing the difference between the polar vortex state characteristics and the actual polar vortex, and the difference between the predicted sea ice concentration sequence and the actual sea ice concentration sequence, the parameters of the first recurrent neural network and the second recurrent neural network are updated.

[0009] As one preferred embodiment, the step of acquiring meteorological time-series data of the area to be predicted and determining the icing zone grid data based on the meteorological time-series data includes: The time-series data of temperature and precipitation in the region to be predicted are obtained as the meteorological time-series data. Based on the time series data of temperature and precipitation, wet and cold composite events are screened. Sub-regions within the predicted area whose duration of the wet and cold composite event meets a preset duration condition are marked as icing zones; The region to be predicted is divided into grids, and the icing area is marked in the grids to obtain the grid data of the icing area.

[0010] As one preferred embodiment, the analysis of the polar vortex state characteristics and the icing zone grid data to obtain the icing zone distribution prediction results includes: The grid data of the icing area is encoded into feature embeddings; The polar vortex state features are fused with the feature embedding, and the fused features are input into a visual Transformer network to obtain the predicted distribution of the icing area.

[0011] Another embodiment of the present invention provides a prediction system for changes in icing zones, comprising: The data acquisition module is used to acquire time series data of sea ice concentration, polar eddy characteristics, and covariate characteristics of the area to be predicted. The data fusion module is used to fuse the covariate time series data with the sea ice concentration time series data to obtain a first correlation sequence characterizing the evolution of polar eddies under covariate constraints, and to fuse the covariate time series data with the polar eddy characteristic time series data to obtain a second correlation sequence. The polar vortex state analysis module is used to analyze the second correlation sequence with the first correlation sequence as a constraint to obtain polar vortex state characteristics. The meteorological data analysis module is used to acquire meteorological time-series data of the area to be predicted, and to determine the grid data of the icing area based on the meteorological time-series data; The icing zone prediction module is used to analyze the polar vortex state characteristics and the icing zone grid data to obtain the icing zone distribution prediction results.

[0012] As one preferred embodiment, the polar vortex state analysis module includes: An environmental constraint feature generation unit is used to input the first associated sequence number into a first recurrent neural network to obtain environmental constraint features; The polar vortex history evolution feature generation unit is used to input the second correlation sequence into the first recurrent neural network to obtain polar vortex history evolution features; The polar vortex state generation unit is used to generate the polar vortex state features by fusing the environmental constraint features and the polar vortex historical evolution features through an attention mechanism.

[0013] As one preferred embodiment, the polar vortex state characteristics are obtained through bidirectional constraint training, and the polar vortex state analysis module further includes: The sea ice prediction unit is used to input the second association sequence and the sea ice concentration time series data into the second recurrent neural network to obtain the predicted sea ice concentration sequence. A bidirectional constraint training unit is used to update the parameters of the first recurrent neural network and the second recurrent neural network with the goal of minimizing the difference between the polar vortex state characteristics and the actual polar vortex, and the difference between the predicted sea ice concentration sequence and the actual sea ice concentration sequence.

[0014] As one preferred embodiment, the meteorological data analysis module includes: The meteorological time-series data acquisition unit is used to acquire the temperature time-series data and precipitation time-series data of the area to be predicted as the meteorological time-series data. The event filtering unit is used to filter cold and wet composite events based on the temperature time series data and the precipitation time series data. An icing zone marking unit is used to mark sub-regions within the area to be predicted whose duration of the wet and cold composite event meets a preset duration condition as icing zones. A grid division unit is used to divide the area to be predicted into a grid and mark the icing area in the grid to obtain the grid data of the icing area.

[0015] Another embodiment of the present invention provides a device for predicting changes in icing zones, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for predicting changes in icing zones as described above.

[0016] In another embodiment of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the method for predicting changes in icing zones as described above.

[0017] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: (1) This invention acquires three types of time-series data: sea ice concentration, polar eddy characteristics, and covariates. It then uses the fusion of covariates and sea ice as constraints to analyze polar eddy evolution, generating polar eddy conditional characteristics. These characteristics are then analyzed together with grid data of the ice-covered area determined by meteorological data to ultimately obtain a predicted distribution of the ice-covered area. This method organically combines large-scale climate factors such as sea ice and polar eddies with near-surface meteorological conditions, enabling the prediction of ice-covered area changes to no longer rely solely on historical statistical extrapolation but rather on multi-factor coupled analysis. Therefore, this invention can effectively reveal the intrinsic driving forces behind the expansion of ice-covered areas into new regions and enhance the explanatory power of the prediction results for actual climate processes.

[0018] (2) This invention establishes a complete analytical chain from large-scale climate systems to regional icing conditions. This method can not only reflect the fluctuation characteristics of icing areas between years, but also capture their long-term migration trends, significantly improving the accuracy, timeliness, and spatial coverage of icing area change predictions. In practical applications, the prediction results can provide earlier and more reliable scientific basis for power line anti-icing upgrades, winter operation and maintenance scheduling, and emergency resource pre-positioning, thereby reducing the operational risks of icing disasters to power grids, transportation, and communication facilities. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for predicting changes in icing zones in one embodiment of the present invention. Figure 2 This is a spatial distribution map of Arctic sea ice concentration in the cold and warm seasons, according to one embodiment of the present invention. Figure 3 This is a diagram showing the variation of Arctic sea ice extent in one embodiment of the present invention; Figure 4 This is a time variation diagram of the area index and intensity index of different regions of the Arctic polar vortex in one embodiment of the present invention during the cold and warm seasons; Figure 5 This is a composite distribution map of Arctic sea ice concentration in high and low polar vortex area index years, according to one embodiment of the present invention. Figure 6 This is a composite distribution map of Arctic sea ice concentration in one embodiment of the present invention, showing the high and low values ​​of the polar vortex intensity index in different years. Figure 7 This is a composite distribution map of Arctic sea ice concentration in one embodiment of the present invention, showing the high and low values ​​of the intensity index of the Northern Hemisphere polar vortex center. Figure 8 This is a distribution diagram of the correlation coefficients between Arctic sea ice concentration and 11 Arctic polar vortex indices in one embodiment of the present invention; Figure 9 This is a diagram showing the annual evolution and trend of the northern boundary of the icing zone in one embodiment of the present invention. Figure 10 This is a schematic diagram of a prediction system for icing zone changes in one embodiment of the present invention; Figure 11 This is a structural block diagram of a device for predicting changes in icing zones according to one embodiment of the present invention.

[0020] Figure label: Among them, 11 is the data acquisition module, 12 is the data fusion module, 13 is the polar vortex state analysis module, 14 is the meteorological data analysis module, 15 is the icing area prediction module, 21 is the processor, and 22 is the memory. Detailed Implementation

[0021] 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. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0023] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0024] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0025] One embodiment of the present invention provides a method for predicting changes in icing zones. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The method for predicting changes in icing zones, as shown in one embodiment of the present invention, includes steps S1 to S5: S1: Obtain time series data of sea ice concentration, polar eddy characteristics, and covariate time series data for the region to be predicted; In this embodiment, three types of basic data are needed to construct the prediction model: time-series data on sea ice concentration, time-series data on polar eddy characteristics, and time-series data on covariates. If the polar eddy time-series data is directly input into the recurrent neural network, the model can only learn the time dependence of the polar eddy itself and cannot know what external factors are driving the changes in the polar eddy. By concatenating the covariates and sea ice concentration into a first correlation sequence, and then inputting it along with the polar eddy data into the recurrent neural network in subsequent steps, the model has the opportunity to learn how the polar eddy will evolve when sea ice and covariates are in a certain state.

[0026] Sea ice concentration refers to the proportion of sea ice coverage to the total area of ​​the sea area, ranging from 0% to 100%. Sea ice concentration time-series data refers to a sequence of sea ice concentration observations arranged chronologically. In this embodiment, the sea ice concentration time-series data originates from satellite remote sensing-derived sea ice concentration grid products. Both the polar eddy characteristic data and covariate data are derived from atmospheric reanalysis data. Atmospheric reanalysis data is a gridded dataset formed by reprocessing historical observation data using numerical assimilation techniques. Specifically, the European Centre for Medium-Range Weather Forecasts Generation 5 (ERA5) reanalysis dataset or the National Center for Environmental Prediction / National Center for Atmospheric Research (NCEP / NCARReanalysis) reanalysis dataset can be used. These datasets simultaneously contain multiple meteorological elements such as geopotential height, sea level pressure, temperature, and precipitation, and their time coverage extends back to 1940 or earlier.

[0027] To demonstrate the stable correlation between sea ice and polar eddies that can be used for subsequent modeling, and to provide a basis for selecting key sea areas, this embodiment requires the following statistical analysis of historical data on sea ice and polar eddies.

[0028] (1) Analysis of the long-term changing trends of sea ice and polar eddies.

[0029] according to Figure 2 Given the spatial distribution of Arctic sea ice concentration during the cold season (November of the previous year to April of the current year) and the warm season (May to October of the current year), we analyze the spatiotemporal variation characteristics of Arctic sea ice. Figure 2 a and Figure 2 b represents the cold season distribution. Figure 2 c and Figure 2 d represents the warm season distribution. Compare this to the data from 1961-1990 (…). Figure 2 a, Figure 2 c) and 1991-2020 ( Figure 2 b、 Figure 2d) It can be observed in two time periods: the Arctic sea ice area decreases overall during the cold season, with the sea ice concentration in the waters around Greenland, Novaya Zemlya and the Chukchi Sea decreasing by more than 20% on average; the sea ice area decreases more significantly during the warm season, with the sea ice concentration in the nearshore areas on the northern side of Eurasia and North America less than 30%.

[0030] Figure 3 'a' represents the monthly series of sea ice extent from 1961 to 2020, where the thin gray solid line represents the annual sea ice extent, the thick black dashed line is the multi-year average from 1961 to 1990, and the thick black solid line is the multi-year average from 1991 to 2020. It can be seen that... Figure 3 This further illustrates the annual cycle and long-term trend of sea ice extent, with September being the period of minimum sea ice extent and around March being the period of maximum sea ice extent. The multi-year average sea ice extent from 1991 to 2020 was about 1 million square kilometers less than the average from 1961 to 1990, with the most significant decrease occurring from late summer to early autumn, averaging about 2 million square kilometers. Figure 3 b represents the long-term series of average sea ice extent for the cold season (solid blue line) and the warm season (dashed red line). Linear trend calculations show that the sea ice extent decreases significantly at a rate of 240,000 square kilometers per decade in the cold season and at a rate of 530,000 square kilometers per decade in the warm season, with the rate of decrease in the warm season being approximately twice that of the cold season.

[0031] Figure 4 The temporal variations of the polar vortex area index and intensity index in different regions during the cold and warm seasons are presented, enabling the analysis of the spatiotemporal variation characteristics of the Arctic polar vortex. Figure 4 a and Figure 4 b represents the polar vortex area index for the cold season and the warm season, respectively. Figure 4 c and Figure 4 d represents the polar vortex intensity index during the cold and warm seasons, respectively. Looking at different regions, the Northern Hemisphere exhibits the largest variations in both area and intensity index of the polar vortex, followed by the Atlantic-European region and the Asian region. Comparing cold and warm seasons within the same region, the Asian region shows the most significant difference in polar vortex area index, with its warm season variation being approximately five times that of the cold season; the North American region shows the largest difference in polar vortex intensity index between cold and warm seasons, reaching up to twice the variation.

[0032] The above analysis shows that both Arctic sea ice and polar eddies exhibit significant long-term variation trends, and there is a certain correspondence between the two in time and space, providing a basis for further exploration of their relationship.

[0033] (2) Composite analysis of sea ice concentration in years with high and low polar vortex index To further explore the correlation between polar vortex and sea ice, the Asian and Northern Hemisphere regions, where the changes are most significant, need to be selected as the subjects of analysis. The polar vortex area index and polar vortex intensity index of these two regions are standardized, and the years with absolute values ​​greater than one standard deviation are taken as the high and low values ​​of the polar vortex index. The Arctic sea ice concentration distribution of these years is then synthesized.

[0034] For details, please refer to Figure 5 , Figure 5 The composite distribution of sea ice concentration for years with high and low polar eddy area indices is presented. Figure 5 a and Figure 5 b represents the composite distribution of the polar vortex area index in Asia for years with high and low values, respectively. Figure 5 c and Figure 5 d represents the composite distribution of the polar vortex area index in the Northern Hemisphere for years with high and low values, respectively. (Comparison) Figure 5 a and Figure 5 b shows that when the polar vortex area in the Asian region is large, the sea ice concentration is higher and the area covered is larger in the waters east of Greenland, west of Novaya Zemlya, and north of the Bering Strait; when the polar vortex area in the Asian region is small, the sea ice concentration in the above-mentioned areas is significantly reduced. (Synthetic results of the Northern Hemisphere polar vortex area index) Figure 5 c. Figure 5 d) Exhibits similar characteristics.

[0035] Figure 6 The composite distribution of sea ice concentration for high and low polar vortex intensity index years is presented, and its spatial distribution characteristics are basically consistent with the composite results of polar vortex area index. Figure 7 The composite distribution of sea ice concentration for high and low values ​​of the polar vortex center intensity index in the Northern Hemisphere is presented. Figure 5 , Figure 6 Conversely, when the intensity at the polar vortex center is high ( Figure 7 a) The sea ice concentration in the above-mentioned areas is low; when the polar vortex center intensity is low ( Figure 7 (b) The sea ice concentration in the aforementioned region is relatively high. This indicates that the relationship between the polar vortex center intensity and sea ice concentration is inversely related to the relationship between the polar vortex area and intensity index.

[0036] The above content further confirms that there is a stable correlation between sea ice and polar eddies, but the strength of the correlation varies in different sea areas, and it is necessary to quantitatively identify the sea areas with the most significant correlation.

[0037] (3) Correlation analysis between sea ice concentration and polar vortex index.

[0038] To quantitatively identify the sea areas with the most significant correlation between sea ice and polar vortexes, Pearson correlation analysis was performed on sea ice concentration and 11 Arctic polar vortex indices. The 11 polar vortex indices include: Asian polar vortex area index, Pacific polar vortex area index, North American polar vortex area index, Atlantic-European polar vortex area index, Northern Hemisphere polar vortex area index, Asian polar vortex intensity index, Pacific polar vortex intensity index, North American polar vortex intensity index, North Atlantic-European polar vortex intensity index, Northern Hemisphere polar vortex intensity index, and Northern Hemisphere polar vortex center intensity index. The analysis period covers the cold season (November of the previous year to April of the current year) from 1961 to 2020.

[0039] Correlation coefficient distribution as follows Figure 8 As shown in the figure. The black scatter dots in the graph indicate that the correlation coefficient passed the 95% significance test. From Figure 8 It can be seen that more than two-thirds of the polar vortex indices show a significant positive correlation in the first region (27°W to 85°E, 66°34′N to 80°N) and a significant negative correlation in the second region (150°W to 90°W, 70°N to 80°N).

[0040] The sea ice area sequences for the first and second regions were further calculated. The sea ice area was calculated using the following formula: multiply the sea ice concentration of each grid within the region by the area of ​​the unit sea area represented by that grid, and then sum over all grids. The correlation coefficients between the calculated sea ice area sequences for the first and second regions and each polar eddy index were calculated, and the results are shown in Tables 1 and 2.

[0041] Table 1. Correlation coefficients between sea ice area in Region 1 and Arctic polar vortex index. Table 2. Correlation coefficients between sea ice area in the second region and the Arctic polar vortex index. As shown in Tables 1 and 2, the polar vortex area index and polar vortex intensity index in the Asian region have the highest correlation coefficients compared to other regions, and are also the dominant factors for the polar vortex area index and polar vortex intensity index in the Northern Hemisphere. Specifically, the correlation coefficient between sea ice area and the polar vortex area index in the first region is 0.46, and the correlation coefficient between sea ice area and the polar vortex intensity index in the Asian region is 0.66, both reaching a significance level of 95%. The correlation coefficient between sea ice area and the polar vortex area index in the second region is -0.05, and the correlation coefficient between sea ice area and the polar vortex intensity index in the Asian region is -0.20, both failing the significance test.

[0042] The above analysis shows a significant statistical correlation between sea ice and polar vortex, with the strongest correlation between sea ice in the first region and the polar vortex index in the Asian region. Therefore, this embodiment selects the first region as the extraction range for sea ice concentration and focuses on the polar vortex index in the Asian region for subsequent modeling.

[0043] Based on the above three analyses, the sea ice concentration of each sub-region within the first region (27°W to 85°E, 66°34′N to 80°N) was extracted from the sea ice concentration data, and the regional average was performed by sub-region to obtain the sea ice concentration time series data.

[0044] Then, the 500 hPa geopotential height field is extracted from the atmospheric reanalysis data, and the polar vortex characteristic data are calculated as follows.

[0045] The polar vortex area index is calculated as follows: Within the defined area under investigation, the fan-shaped area enclosed north of the characteristic contour line of the southern boundary of the polar vortex on the 500 hPa geopotential height field in the Northern Hemisphere is used as the polar vortex area index. The monthly values ​​of the characteristic contour line of the southern boundary of the polar vortex in the Northern Hemisphere at the 500 hPa level are: January 548 dagpm, February 552 dagpm, March 552 dagpm, April 552 dagpm, May 560 dagpm, June 568 dagpm, July 572 dagpm, August 572 dagpm, September 568 dagpm, October 564 dagpm, November 556 dagpm, and December 552 dagpm. The polar vortex area index is calculated using the formula for the area of ​​a curved surface on a sphere. Where R is the Earth's radius, taken as 6378 km; λ1 and λ2 are the longitudes of the western and eastern boundaries of the investigated area, respectively, in radians; and φ is the latitude of the southern boundary of the polar vortex.

[0046] Following the above method, the polar vortex area index is calculated for the following regions: Asia (60°E to 150°E), the Pacific (150°E to 120°W), North America (120°W to 30°W), the Atlantic Europe (30°W to 60°E), and the entire Northern Hemisphere (0° to 360°).

[0047] The polar vortex intensity index is calculated as follows: the total air mass between the 500 hPa isobaric surface and the contour surface containing the characteristic contour line of the southern boundary of the polar vortex is used as the polar vortex intensity index. For the investigated area, the grid is divided with a meridional grid spacing of 10 degrees and a zonal grid spacing of 5 degrees, and the index is calculated using the following formula: Where ρ is atmospheric density; R is Earth's radius, taken as 6378 km; Δλ and Δφ are the meridional and zonal grid spacings, respectively, taken as Δλ=10° and Δφ=5°, converted to radians; H0 is the geopotential height value of the characteristic contour line of the southern boundary of the polar vortex; is the geopotential height value at the grid point in the i-th row and j-th column north of the characteristic contour line.

[0048] Calculate the polar vortex intensity index for the following regions in the manner described above: Asia (60°E to 150°E), Pacific (150°E to 120°W), North America (120°W to 30°W), North Atlantic-Europe (30°W to 60°E), and the entire Northern Hemisphere (0° to 360°).

[0049] The latitude and longitude of the polar vortex center are determined as follows: On the 500 hPa geopotential field in the Northern Hemisphere, the location of the lowest geopotential height center in a high-latitude region is selected as the location of the Northern Hemisphere polar vortex center. The geopotential height value at the smallest grid point near the polar vortex center is taken as the polar vortex center intensity. When two low-center values ​​are the same, the low-vortex center with a relatively larger central closed loop or more minimum value points is selected; if the two low-vortex ranges are also roughly equivalent, the geopotential height value of the low-vortex center located further south is selected.

[0050] Polar vortex morphology parameters are used to describe the shape characteristics of the polar vortex. One possible implementation is to fit an ellipse to the polar vortex boundary, using the ratio of the major axis to the minor axis of the ellipse as the morphology parameter; another possible approach is to calculate the ratio of the polar vortex area to its perimeter. The specific parameter used can be determined based on the actual application scenario.

[0051] By arranging the above polar vortex area index, polar vortex intensity index, polar vortex center latitude and longitude, and polar vortex morphology parameters in chronological order, we obtain the polar vortex characteristic time series data.

[0052] Finally, covariate time-series data are extracted from the atmospheric reanalysis data. In one embodiment of the invention, the covariate time-series data includes Arctic Oscillation time-series data and Siberian High time-series data.

[0053] The Arctic Oscillation Index is calculated as follows: In the region from 20°N to 90°N and from 0° to 360°, empirical orthogonal function analysis (EOF) is performed on the 1000 hPa geopotential height anomaly field. The time coefficient of the first mode is taken and standardized to obtain the Arctic Oscillation Index sequence.

[0054] The Siberian High Pressure Index is calculated as follows: The Siberian region (40°N to 60°N, 80°E to 120°E) is selected as the monitoring area. The winter sea level pressure (SLP) in this region is averaged regionally and then standardized to obtain the Siberian High Pressure Index sequence.

[0055] The physical basis for introducing the Arctic Oscillation (ANO) and the Siberian High (SHB) as covariates is as follows: When the ANO is in a negative phase, the polar vortex weakens, and cold air is more likely to erupt southward, affecting the southern boundary of the ice-covered area; when the ANO is in a positive phase, the polar vortex contracts, and the ice-covered area shrinks northward accordingly. When the SHB strengthens, the East Asian winter monsoon strengthens, which is conducive to transporting cold air to lower latitudes, where it converges with warm and moist air currents to form wet and cold conditions favorable for icing.

[0056] S2: The covariate time series data and the sea ice concentration time series data are fused to obtain the first correlation sequence characterizing the evolution of polar eddies under covariate constraints. The covariate time series data and the polar eddy characteristic time series data are fused to obtain the second correlation sequence. Preferably, in one embodiment of the present invention, the covariate time series data includes Arctic Oscillation time series data and Siberian High time series data. The covariate time series data are fused with sea ice concentration time series data to obtain a first correlation sequence characterizing the evolution of the polar eddy under covariate constraints, including: Time series data of the Arctic Oscillation and the Siberian High were obtained as covariate time series data; The time series data of the Arctic Oscillation, the Siberian High, and the sea ice concentration were time-aligned, and the aligned data were then spliced ​​together to obtain the first associated sequence.

[0057] Specifically, the aforementioned time-series data on sea ice concentration, polar eddy characteristics, and covariate characteristics are aligned along the time dimension so that all three types of data have corresponding values ​​at the same time points. If the time sampling points of different data sources are inconsistent, linear interpolation or nearest neighbor sampling methods can be used to unify them.

[0058] The aligned sea ice concentration time series data, polar eddy characteristic time series data, and covariate time series data were standardized separately. The standardization method was as follows: for each data series, the mean was subtracted and then divided by the standard deviation, so that the mean of the standardized data series was 0 and the standard deviation was 1.

[0059] The time-series data of the Arctic Oscillation (AOC), the Siberian High (SH), and sea ice concentration are time-series aligned. Since the sea ice concentration data and covariate data have already been aligned in the previous steps, this step can be omitted, but it is still explained to maintain the completeness of the fusion process. The aligned data are then concatenated. The concatenation method is as follows: for each time point, the AOC, SH, and sea ice concentration values ​​are combined into a multi-dimensional vector in a fixed order. All the combined vectors from all time points are arranged in ascending order of time to obtain the first correlation sequence. Each time point in this sequence corresponds to a vector, which simultaneously contains the covariate state (characterized by the AOC and SH) and the sea ice state (characterized by sea ice concentration) at that moment.

[0060] Physically speaking, the first correlation sequence represents the "set of external driving factors influencing polar vortex changes." Among these, sea ice concentration reflects the thermal state of the Arctic Ocean, while the Arctic Oscillation and the Siberian High reflect anomalous atmospheric circulation patterns. These three types of factors collectively control the location, intensity, and morphological changes of the polar vortex. By integrating these three factors into a unified sequence input network, this invention enables subsequent models to automatically learn the comprehensive regulatory effects of these factors on the polar vortex, rather than manually pre-setting a certain functional relationship.

[0061] In this embodiment, the second correlation sequence refers to the set of polar eddy states under covariate constraints. Unlike the first correlation sequence, which focuses on the "driving effect of the external environment (sea ice + covariates) on the polar eddy," the second correlation sequence focuses on the "characteristic expression of the polar eddy itself under the influence of covariates." Both describe the relationship between sea ice, polar eddy, and covariates from different perspectives.

[0062] Specifically, the generation process of the second correlation sequence is as follows: First, the time series data of the Arctic Oscillation, the Siberian High, and the polar vortex characteristic time series are time-aligned. The time alignment method is the same as that of the first correlation sequence, that is, by using linear interpolation or nearest neighbor sampling, data from different sources have corresponding values ​​at the same time point.

[0063] The aligned data is then concatenated to obtain the second correlation sequence: for each time point, the Arctic Oscillation value, Siberian High value, polar vortex area index, polar vortex intensity index, polar vortex center latitude and longitude, and polar vortex morphology parameters are combined into a multi-dimensional vector in a fixed order. The combined vectors of all time points are arranged in ascending order of time to obtain the second correlation sequence.

[0064] The core function of the second correlation sequence is to serve as a bidirectional constraint for the training phase of subsequent steps. Specifically, during the subsequent training process, the second correlation sequence, along with the sea ice concentration time series data, is input into the second recurrent neural network to predict the sea ice concentration sequence. By minimizing the difference between this prediction and the actual sea ice concentration, the learning process of the first recurrent neural network can be constrained inversely, forcing the polar vortex state features generated by the first recurrent neural network not only to accurately predict polar vortices but also to effectively reconstruct sea ice (with the participation of covariates).

[0065] S3: Using the first correlation sequence as a constraint, analyze the second correlation sequence to obtain the polar vortex state characteristics; Step S3 uses the first correlation sequence generated in step S2 as a constraint to analyze the second correlation sequence and generate polar vortex state features. The polar vortex state features are an encoded vector sequence that integrates the historical evolution information of the polar vortex itself and the constraint information of the external environment (covariates and sea ice) on the polar vortex, and will serve as the input conditions for the subsequent ice-covered area prediction model.

[0066] In this embodiment, two recurrent neural networks are included. The first recurrent neural network is used to predict polar eddy state characteristics with a first correlation sequence as a constraint; the second recurrent neural network is used to predict sea ice concentration sequences with a second correlation sequence as a constraint. The two networks are jointly optimized through bidirectional constraints during the training phase, but only the first recurrent neural network is used during the prediction phase.

[0067] The polar vortex state characteristics refer to a compact vector representation of the future evolution trend of the polar vortex under the joint constraints of sea ice and covariates (Arctic Oscillation, Siberian High). It is not a physical quantity that can be intuitively interpreted, but a high-dimensional feature encoding. The subsequent visual Transformer network will directly use this feature as an input condition for predicting the distribution of ice-covered areas.

[0068] Preferably, in one embodiment of the present invention, the second correlation sequence is analyzed using the first correlation sequence as a constraint to obtain polar vortex condition characteristics, including: The first associated sequence number is input into the first recurrent neural network to obtain the environmental constraint features; The second correlation sequence is input into the first recurrent neural network to obtain the historical evolution characteristics of the polar vortex. By integrating environmental constraint features with the historical evolution features of polar vortices through an attention mechanism, polar vortex state features are generated.

[0069] Specifically, the first correlation sequence is first input into the first recurrent neural network. This first correlation sequence contains information from the concatenation of Arctic Oscillation time-series data, Siberian High time-series data, and sea ice concentration time-series data, representing the "environmental constraints." The first recurrent neural network, through its recurrent structure, processes each vector in the first correlation sequence sequentially along the time steps, outputting a hidden state at each time step. These hidden states are then arranged in chronological order to obtain the environmental constraint feature sequence, which characterizes the evolution of the environmental constraints over time.

[0070] Simultaneously, the second correlation sequence needs to be input into the same first recurrent neural network. The polar vortex feature time series data includes the polar vortex area index, polar vortex intensity index, polar vortex center latitude and longitude, and polar vortex morphological parameters, representing the "historical state of the polar vortex itself." The first recurrent neural network also processes each vector in the second correlation sequence sequentially along the time step to obtain the polar vortex historical evolution feature sequence, which characterizes the law of polar vortex evolution over time.

[0071] In this embodiment, the first recurrent neural network can employ a bidirectional LSTM (Long Short-Term Memory) network. A bidirectional LSTM includes two processing directions: the forward direction processes the sequence from front to back, capturing the influence of past information on the present; the backward direction processes the sequence from back to front, capturing the influence of future information on the present. The outputs of the two directions are concatenated to obtain the hidden state at each time step. This bidirectional structure allows the extracted environmental constraint features and polar vortex historical evolution features to simultaneously utilize past and future information in the sequence, thereby more comprehensively characterizing temporal dependencies.

[0072] After obtaining the environmental constraint feature sequence and the polar eddy history evolution feature sequence, the two are fused using an attention mechanism, the calculation method of which is as follows: The historical evolution feature sequence of polar vortices is used as the query, and the environmental constraint feature sequence is used as the key and value. For each time step, the similarity between the query and each key is calculated, and the similarity is normalized into attention weights through the Softmax function. Then, the values ​​are weighted and summed using these weights to obtain the fusion result for that time step.

[0073] Let Q be the characteristic sequence of the polar vortex's historical evolution, and K and V be the characteristic sequences of environmental constraints. First, calculate the attention score matrix: Then, Softmax normalization is performed on each row of S to obtain the attention weight matrix A, and finally, the fusion result is calculated: Each vector in F represents the "polar vortex state characteristic" for that time step.

[0074] The role of this attention mechanism is that, when processing the historical evolution characteristics of polar eddies, the model can dynamically focus on the most important parts of the environmental constraints that are relevant to the current state of the polar eddy. For example, when there are abnormal changes in sea ice concentration, the model will give higher attention weight to the environmental constraints at that moment. This dynamic attention capability makes the generation of polar eddy state features not a simple feature splicing, but a selective information fusion. Arranging the above fusion results in chronological order yields the polar eddy state feature sequence.

[0075] The generation process of the aforementioned polar vortex state characteristics depends on the parameters of the first recurrent neural network. In order for the network to learn the mapping relationship between reasonable environmental constraints and polar vortex evolution, this embodiment uses a bidirectional constraint training strategy to train the network.

[0076] Preferably, in one embodiment of the present invention, the polar vortex state characteristics are obtained through bidirectional constraint training, including: The second association sequence and sea ice concentration time series data are input into the second recurrent neural network to obtain the predicted sea ice concentration sequence; The parameters of the first and second recurrent neural networks are updated with the goal of minimizing the difference between the polar eddy state characteristics and the actual polar eddy, and the difference between the predicted sea ice concentration sequence and the actual sea ice concentration sequence.

[0077] The second association sequence is generated in step S2 and is used directly here without being generated again. The difference between the second association sequence and the first association sequence is that the first association sequence is "covariate + sea ice" and is used to constrain the prediction of polar eddies; the second association sequence is "covariate + polar eddies" and is used to constrain the prediction of sea ice. Together, they form the basis of bidirectional constraint training.

[0078] The second correlation sequence and the sea ice concentration time series data are then input into the second recurrent neural network. The structure of the second recurrent neural network is the same as that of the first recurrent neural network, which is a bidirectional LSTM. Its input includes two parts: the second correlation sequence represents the "conditional information composed of polar eddies and covariates", and the sea ice concentration time series data represents the "target true value".

[0079] The processing of the second recurrent neural network is similar to that of the first recurrent neural network: firstly, conditional features are extracted from the second associated sequence, and sea ice historical evolution features are extracted from the sea ice concentration time series data. Then, the two are fused through an attention mechanism, and finally, they are mapped to a predicted sea ice concentration sequence through a fully connected layer.

[0080] During training, two loss functions are calculated simultaneously. The first loss function is the difference between the polar eddy state characteristics and the actual polar eddy, which can be calculated using mean squared error (MSE). The second loss function is the difference between the predicted sea ice concentration sequence output by the second recurrent neural network and the actual sea ice concentration sequence, also calculated using MSE. The total loss function is the weighted sum of the two losses, where the weighting coefficients can be set according to the specific application scenario; in this embodiment, they can be selected as 0.5.

[0081] With the goal of minimizing the total loss function, the parameters of the first and second recurrent neural networks are updated simultaneously using gradient descent algorithms (such as the Adam optimizer).

[0082] The advantage of this bidirectional constraint training is that the first recurrent neural network not only learns to predict the "polar eddy state" from the "environmental constraints," but also ensures that the predicted polar eddy state can (with the participation of covariates) accurately reconstruct sea ice. If the polar eddy state characteristics deviate from the actual physical laws, it will be unable to accurately reconstruct sea ice, resulting in a significant second loss. Therefore, the second loss acts as a constraint on the first recurrent neural network, forcing the network to learn feature representations that conform to the bidirectional physical relationship between sea ice and polar eddies.

[0083] After training, in actual prediction, only the first recurrent neural network is needed to generate polar vortex state features, and the second recurrent neural network is no longer involved.

[0084] S4: Obtain meteorological time-series data for the area to be predicted, and determine the grid data for the icing zone based on the meteorological time-series data; Step S4 first acquires the meteorological time-series data of the area to be predicted, and then determines the icing zone grid data based on the meteorological time-series data. The icing zone grid data is a numerical representation of the spatial distribution of the icing zone and will serve as one of the inputs to the subsequent visual Transformer model.

[0085] Preferably, in one embodiment of the present invention, acquiring meteorological time-series data of the area to be predicted and determining icing zone grid data based on the meteorological time-series data includes: Acquire time-series temperature and precipitation data of the area to be predicted as meteorological time-series data; Based on time-series temperature and precipitation data, we screened for cold and wet composite events. Sub-regions within the area to be predicted whose duration of wet and cold combined events meets the preset duration condition are marked as icing zones; The area to be predicted is divided into grids, and the icing areas are marked in the grids to obtain the icing area grid data.

[0086] Icing requires specific meteorological conditions. In practice, icing on conductors or object surfaces typically requires a temperature range of 0°C to -10°C and a sufficient supply of water vapor. However, due to the significant non-uniformity of relative humidity observation data, it is difficult to directly use relative humidity for accurate analysis of icing areas. Therefore, this embodiment defines a "wet and cold composite event" that is conducive to icing formation based on the combination of air temperature and precipitation conditions.

[0087] The definition of a wet-cold composite event requires that the following four conditions be met simultaneously: (1) Temperature is lower than the historical temperature percentile threshold for the same period. The specific calculation method is as follows: For each day within a set year interval (e.g., 1961 to 2022), take a set number of days before and after that day (e.g., 15 days before and after) as a sample window. Sort all temperature samples near the same day in all years within the window, and determine the 10th percentile value of that day as the temperature percentile threshold. When the temperature of the day is lower than this threshold, this condition is met. The physical meaning of this condition is: the current temperature is significantly lower than the climate average, belonging to a "colder than average" case.

[0088] (2) Precipitation exceeds the historical precipitation threshold for the same period. The specific calculation method is similar to that for temperature: for each day, precipitation samples from the preceding and following days are sorted, and the 85th percentile value of that day is determined as the precipitation threshold. This condition is met when the precipitation on a given day exceeds this threshold. The physical meaning of this condition is: the current precipitation is significantly higher than the climate average, which is a case of "wetter".

[0089] (3) The air temperature is below the upper limit for icing formation. According to meteorological observations on icing formation, when the air temperature is above 1°C, supercooled water droplets are difficult to maintain a liquid state, and icing is not easy to form. Therefore, an air temperature below 1°C is set as the absolute threshold condition. This condition is met when the air temperature on a given day is below 1°C.

[0090] (4) Precipitation exceeds the minimum precipitation required for icing. A precipitation level of 1 mm or more is set as an absolute threshold condition. This condition is met when the daily precipitation is greater than 1 mm.

[0091] A single day that simultaneously meets all four of the above conditions is defined as a wet-cold composite event that is conducive to icing formation. For the region to be predicted, the occurrence of wet and cold complex events during the winter (November to March of the following year) is statistically analyzed annually. Spatially, the region to be predicted is divided into several sub-regions. For each sub-region, the duration of wet and cold complex events is statistically analyzed. When the number of consecutive days of wet and cold complex events in a sub-region exceeds a preset duration threshold (e.g., 5 days), the sub-region is designated as an icing zone. The boundary of the icing zone can be defined as follows: each winter, the northernmost location within a set longitude range where the duration of wet and cold complex events exceeds 5 days is defined as the northern boundary of the icing zone, and the southernmost location where the same condition occurs is defined as the southern boundary of the icing zone.

[0092] For details, please refer to Figure 9 , Figure 11 The changes in the northern boundary of the regional icing zone, statistically obtained based on the above definition, are shown. Figure 9 (a) shows the annual evolution of the northern boundary of the overall (70°E to 135°E) icing zone; the blue solid line represents its long-term linear trend. From Figure 9(a) It can be seen that the northern boundary of the ice cover zone in the region has shown a significant northward expansion trend over the past 60 years, a trend that passed the 99% significance test. Figure 9 (b) shows the annual evolution of the northern boundary of the icing zone in the eastern part of the region (105°E to 135°E), which also exhibits a significant northward expansion. Notably, in addition to this long-term northward expansion trend, the location of the icing zone also shows significant interannual fluctuations, indicating that the area of ​​winter icing in a given year depends on the combined effects of long-term climate change and interannual variability. This suggests that icing zone prediction methods based solely on historical averages are insufficient to capture this dynamic change; therefore, it is necessary to develop prediction models that consider large-scale climate factors.

[0093] In order to use the spatial distribution of the icing area in subsequent visual Transformer models, the icing area needs to be represented in a grid.

[0094] Specifically, the area to be predicted is divided into regular grids at fixed latitude and longitude intervals (e.g., 0.5°×0.5° or 1°×1°). Each grid cell corresponds to a fixed area on the actual map. For each point in time (e.g., each winter), based on the above-mentioned icing zone determination results, each grid cell is marked: if the location of the grid cell belongs to the icing zone determined for that year, the value of the grid cell is marked as 1; otherwise, it is marked as 0.

[0095] The values ​​of all grid cells are arranged spatially to form a two-dimensional matrix, which is the icing zone mask at that time point. The icing zone masks at multiple time points are arranged in chronological order to obtain the icing zone mask sequence, which is the icing zone grid data referred to in this embodiment.

[0096] S5: Analyze the polar vortex state characteristics and icing zone grid data to obtain the predicted icing zone distribution results.

[0097] Step S5: Analyze the polar vortex state characteristics generated in the above steps with the grid data of the icing area, and output the final predicted results of the icing area distribution.

[0098] Preferably, in one embodiment of the present invention, the polar vortex state characteristics and icing zone grid data are analyzed to obtain icing zone distribution prediction results, including: Encode the grid data of the icy area into feature embeddings; The polar vortex state features are fused with feature embeddings, and the fused features are input into a visual Transformer network to obtain the predicted distribution of icing areas.

[0099] Specifically, the icing area grid data is first encoded into feature embeddings. The icing area grid data is a two-dimensional matrix generated in the previous steps, where each grid cell has a value of 0 or 1, indicating whether the location belongs to the icing area. This binary matrix is ​​essentially a single-channel spatial distribution map.

[0100] The visual Transformer network cannot directly process the raw two-dimensional matrix and needs to convert it into a vector sequence. The encoding process is as follows: The icing area grid data is divided into several non-overlapping image patches according to a preset patch size (e.g., 16×16 or 8×8). Each image patch contains information about the icing distribution in that local area. For each image patch, it is flattened and mapped into a fixed-dimensional vector through a linear mapping layer; this vector is called a "feature embedding". The feature embeddings of all image patches are arranged in spatial order to obtain a vector sequence. At the same time, a learnable positional encoding is added to each location to preserve the spatial relationship between different image patches. After encoding, the original icing area grid data is converted into a set of feature embedding sequences, which can be further processed by the visual Transformer network.

[0101] After obtaining the feature embedding sequence of the icing zone grid data, it is fused with the polar vortex state features generated in the previous steps. The purpose of fusion is to enable the visual Transformer network to simultaneously refer to the current spatial distribution pattern of the icing zone and the large-scale climate constraints represented by the polar vortex state features when predicting the future distribution of the icing zone.

[0102] In one embodiment of the present invention, fusion is achieved through a cross-attention mechanism. Specifically, a visual Transformer network constructs a cross-attention layer, using the feature embedding sequence as the query and the polar vortex state features as the key and value. For the embedding vector corresponding to each image patch in the feature embedding sequence, the similarity between it and each element in the polar vortex state features is calculated. Based on the similarity, the information in the polar vortex state features is aggregated in a weighted manner to generate the conditional embedding of the image patch.

[0103] After fusion, the conditionalized feature embedding sequence is input into the main body of the visual Transformer network. The visual Transformer network consists of multiple stacked Transformer encoder layers, each containing a multi-head self-attention mechanism and a feedforward neural network. The self-attention mechanism enables the network to capture dependencies between different spatial locations—for example, the icing state at one location may be correlated with its neighboring or distant locations. The feedforward neural network performs a non-linear transformation on the embedding vector at each location, extracting higher-level spatial features.

[0104] This cross-attention fusion approach enables the visual Transformer network to dynamically extract climate constraint information related to the polar vortex state features based on the distribution characteristics of different locations within the icing zone. For example, for image patches in higher latitude regions, the model may pay more attention to changes in the polar vortex intensity index; while for image patches in lower latitude regions, the model may pay more attention to changes in the polar vortex area index.

[0105] The task of the visual Transformer network is to model the mapping relationship between historical icing distribution and future icing distribution. During the training phase, the network takes the icing mask sequence of historical time periods (after feature embedding and fusion, and then adding polar vortex state features) as input and the icing mask of future time periods as supervision signal to learn the transition pattern between the two.

[0106] In this embodiment, an autoregressive approach can be used for prediction: the grid data of the icing area at the past T time points is used as input, and the network outputs the predicted distribution of the icing area at the next time point. This prediction result is then used as input for subsequent time points to progressively derive predictions for longer time ranges.

[0107] The output layer of the visual Transformer network maps the feature embeddings of each image patch to probability values ​​between 0 and 1 through a linear mapping and a sigmoid activation function, representing the probability that the corresponding region of the image patch belongs to the icing area. The probability values ​​of all image patches are rearranged according to their spatial location to form a two-dimensional matrix of the same size as the input, which is the predicted icing area mask. This mask is the output of step S5—the predicted icing area distribution result.

[0108] The aforementioned visual Transformer network employs a self-supervised training strategy. Specifically, during the training phase, the input time-series data is randomly masked; that is, the grid data of the icing area at certain time points or spatial locations is randomly replaced with mask markers. The network then needs to predict the content of the masked data based on the unmasked data and the polar vortex state characteristics. In this way, the model can learn the inherent patterns of the icing area distribution in the spatiotemporal dimensions, enhancing the robustness of the predictions.

[0109] After training, during the prediction phase, grid data of icing areas for all historical periods and corresponding polar vortex state characteristics are input into the network. The network outputs predictions of icing area distribution for future periods. These results are further used for regional icing early warning. For example, the predicted icing area mask can be integrated with a geographic information system to identify power lines, traffic routes, or communication facilities that may be affected by icing, allowing for the early issuance of warnings and providing a scientific basis for icing prevention and disaster reduction decisions.

[0110] Because the polar vortex state characteristics include leading signals from sea ice and covariates (e.g., if sea ice concentration is low in the previous warm season, the icing area will be further south in the following winter), and combined with long-term, slowly changing climate factors such as the Arctic Oscillation and the Siberian High, the prediction model can capture the interannual and interdecadal trends of polar vortex evolution. Compared to traditional icing prediction methods that rely on short-term weather forecasts, this invention can predict icing area change trends on seasonal to interannual scales, significantly improving the prediction lead time and providing earlier decision-making basis for power grid disaster prevention and mitigation. Another embodiment of this invention provides a prediction system for icing area changes; for details, please refer to [link to relevant documentation]. Figure 10 , Figure 10 The diagram shown is a schematic representation of a prediction system for icing zone changes in an embodiment of the present invention, comprising: Data acquisition module 11 is used to acquire time series data of sea ice concentration, polar eddy characteristics and covariate time series data of the area to be predicted; Data fusion module 12 is used to fuse covariate time series data with sea ice concentration time series data to obtain a first correlation sequence characterizing the evolution of polar eddies under covariate constraints, and to fuse covariate time series data with polar eddy characteristic time series data to obtain a second correlation sequence. The polar vortex state analysis module 13 is used to analyze the second correlation sequence with the first correlation sequence as a constraint to obtain the polar vortex state characteristics. Meteorological data analysis module 14 is used to acquire meteorological time-series data of the area to be predicted and determine the grid data of the icing area based on the meteorological time-series data; The icing zone prediction module 15 is used to analyze the polar vortex state characteristics and icing zone grid data to obtain the icing zone distribution prediction results.

[0111] Preferably, in one embodiment of the present invention, the polar vortex state analysis module includes: An environmental constraint feature generation unit is used to input the first associated sequence number into the first recurrent neural network to obtain environmental constraint features; The polar vortex history evolution feature generation unit is used to input the second correlation sequence into the first recurrent neural network to obtain the polar vortex history evolution features. The polar vortex state generation unit is used to generate polar vortex state features by fusing environmental constraint features and polar vortex historical evolution features through an attention mechanism.

[0112] Preferably, in one embodiment of the present invention, the polar vortex state characteristics are obtained through bidirectional constraint training, and the polar vortex state analysis module further includes: a sea ice prediction unit, used to input the second correlation sequence and sea ice concentration time series data into the second recurrent neural network to obtain a predicted sea ice concentration sequence; A bidirectional constraint training unit is used to update the parameters of the first and second recurrent neural networks with the goal of minimizing the difference between the polar eddy state characteristics and the actual polar eddy, and the difference between the predicted sea ice concentration sequence and the actual sea ice concentration sequence.

[0113] Preferably, in one embodiment of the present invention, the meteorological data analysis module includes: The meteorological time-series data acquisition unit is used to acquire temperature time-series data and precipitation time-series data of the area to be predicted as meteorological time-series data. The event filtering unit is used to filter cold and wet composite events based on temperature time series data and precipitation time series data. The icing zone marking unit is used to mark the sub-regions within the area to be predicted where the duration of wet and cold composite events meets a preset duration condition as icing zones. The grid division unit is used to divide the area to be predicted into grids and mark the icing area in the grids to obtain the icing area grid data.

[0114] See Figure 11 This is a structural block diagram of an ice-covered area change prediction device provided in an embodiment of the present invention. The ice-covered area change prediction device 20 provided in this embodiment includes a processor 21, a memory 22, and a computer program stored in the memory 22 and configured to be executed by the processor 21. When the processor 21 executes the computer program, it implements the steps as described in the above embodiment of the ice-covered area change prediction method, for example... Figure 1 The steps S1 to S5 described above; or, when the processor 21 executes the computer program, it implements the functions of each module in the above-described device embodiments, such as the data acquisition module 11.

[0115] For example, the computer program can be divided into one or more modules, which are stored in the memory 22 and executed by the processor 21 to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the icing zone change prediction device 20. For example, the computer program can be divided into a data acquisition module 11, a data fusion module 12, a polar vortex state analysis module 13, a meteorological data analysis module 14, and an icing zone prediction module 15, with the specific functions of each module as follows: Data acquisition module 11 is used to acquire time series data of sea ice concentration, polar eddy characteristics and covariate time series data of the area to be predicted; Data fusion module 12 is used to fuse the covariate time series data with the sea ice concentration time series data to obtain a first correlation sequence characterizing the evolution of polar eddies under covariate constraints, and to fuse the covariate time series data with the polar eddy characteristic time series data to obtain a second correlation sequence. The polar vortex state analysis module 13 is used to analyze the second correlation sequence with the first correlation sequence as a constraint to obtain the polar vortex state characteristics. Meteorological data analysis module 14 is used to acquire meteorological time-series data of the area to be predicted and determine the grid data of the icing area based on the meteorological time-series data; The icing zone prediction module 15 is used to analyze the polar vortex state characteristics and icing zone grid data to obtain the icing zone distribution prediction results.

[0116] The icing zone change prediction device 20 may include, but is not limited to, a processor 21 and a memory 22. Those skilled in the art will understand that the schematic diagram is merely an example of an icing zone change prediction device and does not constitute a limitation on the icing zone change prediction device 20. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the icing zone change prediction device 20 may also include input / output devices, network access devices, buses, etc.

[0117] The processor 21 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 21 is the control center of the icing zone change prediction device 20, connecting all parts of the icing zone change prediction device 20 via various interfaces and lines.

[0118] The memory 22 can be used to store the computer program and / or modules. The processor 21 implements various functions of the icing zone change prediction device 20 by running or executing the computer program and / or modules stored in the memory 22 and calling the data stored in the memory 22. The memory 22 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 22 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0119] The module integrated into the icing zone change prediction device 20, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also 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. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0120] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0121] Accordingly, embodiments of the present invention provide a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform steps in the icing zone change prediction method of the above embodiments, for example... Figure 1 Steps S1 to S5 as described above.

[0122] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: (1) This invention introduces covariate time-series data such as the Arctic Oscillation and the Siberian High, and trains a recurrent neural network in a bidirectional constraint manner, so that the model can take into account the physical consistency of both directions when learning the relationship between sea ice and polar eddies. This training strategy does not rely on manually labeled ice-covered samples, but uses the inherent coupling mechanism within the climate system as a supervision signal, so that it can still maintain good predictive ability in areas where historical ice-covered data is sparse or missing.

[0123] (2) This invention establishes the spatial distribution prediction results of icing areas on the basis of gridded data, and can output an icing area distribution map with clear geographical location information. Compared with the traditional method that only gives the icing level or range boundary, the gridded prediction results can be directly connected with the geographic information system, which is convenient for industries such as power and transportation to conduct risk assessment and resource scheduling at the level of specific lines or stations.

[0124] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for predicting changes in icing zones, characterized in that, include: Acquire time-series data on sea ice concentration, polar eddy characteristics, and covariates for the region to be predicted. By fusing the covariate time series data with the sea ice concentration time series data, a first correlation sequence characterizing the evolution of polar eddies under covariate constraints is obtained; by fusing the covariate time series data with the polar eddy characteristic time series data, a second correlation sequence is obtained. Using the first correlation sequence as a constraint, the second correlation sequence is analyzed to obtain the polar vortex state characteristics; Obtain meteorological time-series data of the area to be predicted, and determine the grid data of the icing area based on the meteorological time-series data; The polar vortex state characteristics and the icing zone grid data are analyzed to obtain the icing zone distribution prediction results.

2. The method for predicting changes in icing zones as described in claim 1, characterized in that, The covariate time-series data includes Arctic Oscillation time-series data and Siberian High time-series data. The process of fusing the covariate time-series data with the sea ice concentration time-series data to obtain a first correlation sequence characterizing the evolution of the polar eddy under covariate constraints includes: The Arctic Oscillation time series data and the Siberian High pressure time series data are obtained as the covariate time series data; The Arctic Oscillation time series data, the Siberian High time series data, and the sea ice concentration time series data are time-aligned, and the aligned data are then spliced ​​together to obtain the first associated sequence.

3. The method for predicting changes in icing zones as described in claim 1, characterized in that, The step of analyzing the second correlation sequence using the first correlation sequence as a constraint to obtain polar vortex state characteristics includes: The first associated sequence number is input into the first recurrent neural network to obtain the environmental constraint features; The second associated sequence is input into the first recurrent neural network to obtain the historical evolution characteristics of the polar vortex. The polar vortex state features are generated by fusing the environmental constraint features and the polar vortex historical evolution features through an attention mechanism.

4. The method for predicting changes in icing zones as described in claim 3, characterized in that, The polar vortex state characteristics are obtained through bidirectional constraint training, including: The second associated sequence and the sea ice concentration time series data are input into the second recurrent neural network to obtain the predicted sea ice concentration sequence; With the goal of minimizing the difference between the polar vortex state characteristics and the actual polar vortex, and the difference between the predicted sea ice concentration sequence and the actual sea ice concentration sequence, the parameters of the first recurrent neural network and the second recurrent neural network are updated.

5. The method for predicting changes in icing zones as described in claim 1, characterized in that, The step of acquiring meteorological time-series data of the area to be predicted and determining the icing zone grid data based on the meteorological time-series data includes: The time-series data of temperature and precipitation in the region to be predicted are obtained as the meteorological time-series data. Based on the time series data of temperature and precipitation, wet and cold composite events are screened. Sub-regions within the predicted area whose duration of the wet and cold composite event meets a preset duration condition are marked as icing zones; The region to be predicted is divided into grids, and the icing area is marked in the grids to obtain the grid data of the icing area.

6. The method for predicting changes in icing zones as described in claim 1, characterized in that, The analysis of the polar vortex state characteristics and the icing zone grid data yields the predicted icing zone distribution results, including: The grid data of the icing area is encoded into feature embeddings; The polar vortex state features are fused with the feature embedding, and the fused features are input into a visual Transformer network to obtain the predicted distribution of the icing area.

7. A prediction system for changes in icing zones, characterized in that, include: The data acquisition module is used to acquire time series data of sea ice concentration, polar eddy characteristics, and covariate characteristics of the area to be predicted. The data fusion module is used to fuse the covariate time series data with the sea ice concentration time series data to obtain a first correlation sequence characterizing the evolution of polar eddies under covariate constraints, and to fuse the covariate time series data with the polar eddy characteristic time series data to obtain a second correlation sequence. The polar vortex state analysis module is used to analyze the second correlation sequence with the first correlation sequence as a constraint to obtain polar vortex state characteristics. The meteorological data analysis module is used to acquire meteorological time-series data of the area to be predicted, and to determine the grid data of the icing area based on the meteorological time-series data; The icing zone prediction module is used to analyze the polar vortex state characteristics and the icing zone grid data to obtain the icing zone distribution prediction results.

8. The prediction system for icing zone changes as described in claim 7, characterized in that, The polar vortex state analysis module includes: An environmental constraint feature generation unit is used to input the first associated sequence number into a first recurrent neural network to obtain environmental constraint features; The polar vortex history evolution feature generation unit is used to input the second correlation sequence into the first recurrent neural network to obtain polar vortex history evolution features; The polar vortex state generation unit is used to generate the polar vortex state features by fusing the environmental constraint features and the polar vortex historical evolution features through an attention mechanism.

9. The prediction system for icing zone changes as described in claim 8, characterized in that, The polar vortex state characteristics are obtained through bidirectional constraint training. The polar vortex state analysis module further includes: The sea ice prediction unit is used to input the second association sequence and the sea ice concentration time series data into the second recurrent neural network to obtain the predicted sea ice concentration sequence. A bidirectional constraint training unit is used to update the parameters of the first recurrent neural network and the second recurrent neural network with the goal of minimizing the difference between the polar vortex state characteristics and the actual polar vortex, and the difference between the predicted sea ice concentration sequence and the actual sea ice concentration sequence.

10. The prediction system for icing zone changes as described in claim 7, characterized in that, The meteorological data analysis module includes: The meteorological time-series data acquisition unit is used to acquire the temperature time-series data and precipitation time-series data of the area to be predicted as the meteorological time-series data. The event filtering unit is used to filter cold and wet composite events based on the temperature time series data and the precipitation time series data. An icing zone marking unit is used to mark sub-regions within the area to be predicted whose duration of the wet and cold composite event meets a preset duration condition as icing zones. A grid division unit is used to divide the area to be predicted into a grid and mark the icing area in the grid to obtain the grid data of the icing area.

11. A device for predicting changes in icing zones, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the method for predicting changes in icing zones as described in any one of claims 1 to 6.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the method for predicting changes in icing zones as described in any one of claims 1 to 6.

Images