Transmission line ice disaster early warning method based on cross-grid semantic fusion

Abstract

The invention discloses a power transmission line ice disaster early warning method based on cross-grid semantic fusion, and belongs to the technical field of disaster prevention, disaster reduction and automatic data processing of a power system. According to the invention, the problem of low accuracy of ice disaster early warning is solved. According to the method, a semantic character string of a cross-grid assimilation semantic expression module is used for expressing a grid and all meteorological station information related to the grid on a large scale, meanwhile, for one grid, not only is the information of the grid integrated, but also cross-regional information related to the grid is integrated, and the situation that a rigid boundary exists in grid decision is prevented; in the aspect of obtaining decision attributes, the ice disaster feature semantic fusion module utilizes a text-to-vector conversion process to highly fuse key information of a region and align structure dimensions through semantic vectors. The data of different numbers of meteorological stations are fused together, so that the decision process is more efficient, and the requirements of high accuracy and high quality power transmission line ice disaster early warning are met. The method can be applied to the field of power transmission line ice disaster early warning.

Description

Technical Field

[0001] This invention belongs to the field of power system disaster prevention, mitigation and automated data processing technology, specifically involving a cross-grid semantic fusion method for early warning of ice disasters on transmission lines. Background Technology

[0002] Ice storms affecting power transmission lines severely threaten the safe and stable operation of the power grid. Accurate ice storm forecasting plays a crucial role in energy security and socio-economic development. Therefore, to ensure the smooth operation of this vital power lifeline, it is essential to conduct high-precision forecasting of ice storms affecting power transmission lines using complex meteorological data. This work has significant economic and social value for power grid management.

[0003] In traditional methods for predicting ice storms on power transmission lines, the main technical means include: 1) Simplifying complex meteorological data from real-world production environments into ideal laboratory conditions and directly dividing an area into regular grids, then making decisions about the existence of power line ice disasters within each grid area based on data from each grid: The underlying expectation of this model is that meteorological stations are distributed very evenly and densely, thus forming effective decision attributes. However, in real-world production environments, the spatial distribution of meteorological stations is extremely uneven, and the resolution of regional satellite data is still insufficient. Therefore, when faced with unevenly distributed meteorological station data, traditional methods directly perform spatial numerical interpolation, which leads to significant biases in sparsely populated areas of meteorological stations. Furthermore, this error is continuously amplified during the calculation process, directly causing rigid historical meteorological models to easily fail in applications.

[0004] 2) Existing technologies attempt to extract individual attributes from time-series meteorological station information and then establish correlations between these attributes and specific physical grids. Since the actual physical distances of different meteorological stations from the target grid are not necessarily the same, this correlation is not a strict one-to-one relationship. This non-strict correspondence easily leads to dimensional misalignment of the data itself. Once the data is misaligned, conventional decision-making models cannot obtain effective input; therefore, how to integrate these characteristics for scientific decision-making becomes a huge challenge. Moreover, distance-based grids often have rigid boundaries, which can disrupt continuous meteorological changes, making it impossible for the entire decision-making process to consistently and coherently predict the risk of power transmission line ice storms.

[0005] Therefore, there is an urgent need to propose a new forecasting method to overcome the above-mentioned shortcomings. The new method must have the ability to process information across grids, effectively integrate data to prevent decision-making errors caused by the rigid boundaries of traditional grids, and assimilate meteorological data from multiple meteorological stations and embed them into a high-dimensional space, thereby replacing the traditional forecasting model based on a single grid and rigid attributes, and achieving more reliable forecasts. Summary of the Invention

[0006] The purpose of this invention is to solve the problem of low accuracy in ice disaster early warning caused by uneven distribution of meteorological stations leading to bias in grid data and rigid boundaries of distance-based grids. Therefore, a cross-grid semantic fusion method for transmission line ice disaster early warning is proposed.

[0007] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a cross-grid semantic fusion method for early warning of ice storms on transmission lines, the method specifically including the following steps: Step S1: Preprocess the historical data collected by meteorological stations in the area to be detected to generate a meteorological color image QYColor; Step S2: Use a clustering algorithm to segment the meteorological color image QYColor to obtain each sub-image block, and store all sub-image blocks in a spatial reference list; And the meteorological color image QYColor is divided into grids; Step S3: Establish the semantic conversion module YUZHModel for ice disaster conditions of transmission lines. The YUZHModel is used to convert meteorological information collected by meteorological stations into structured semantic description text. Step S4: Establish the cross-mesh assimilation semantic representation module KWGModel. The module KWGModel obtains cross-mesh assimilation semantic features based on the mesh partitioning results and the module YUZHModel. Step S5: Establish the ice disaster feature semantic fusion module BZModel. The ice disaster feature semantic fusion module BZModel is used to convert cross-grid assimilation semantic features into high-dimensional semantic vectors, and obtain the DTFeature field of the decision table based on the high-dimensional semantic vectors. The DTDecision field of the decision table is determined according to the historical ice disaster situation in the area to be detected. Step S6: Establish a semantic embedding decision model NNModel for ice disasters on transmission lines, and train the established model using the information in the decision table; Step S7: Obtain real-time data collected by meteorological stations in the area to be detected, obtain the input of the trained model NNModel based on the real-time collected data, and output the decision result through the trained model NNModel.

[0008] The beneficial effects of this invention are: This invention constructs a cross-grid assimilation semantic representation module, using semantic strings instead of simple numerical values ​​to represent a grid and all meteorological station information related to that grid on a large scale. Furthermore, it integrates not only the grid's own information but also cross-regional information related to the grid, preventing rigid boundaries in grid-based decision-making. Simultaneously, it establishes an ice disaster feature semantic fusion module. In acquiring decision attributes, it utilizes a text-to-vector conversion process, ensuring consistent text vector lengths even with varying text lengths. This semantic vector highly integrates key regional information with aligned structural dimensions. This method can integrate data from different numbers of meteorological stations, making the decision-making process more efficient, meeting the requirements for high-accuracy, high-quality transmission line ice disaster early warning, and is unaffected by uneven distribution of meteorological stations. Attached Figure Description

[0009] Figure 1 This is a flowchart of a cross-grid semantic fusion method for early warning of ice disasters on power transmission lines according to the present invention. Detailed Implementation

[0010] Specific implementation method one: Combining Figure 1 This embodiment describes a cross-grid semantic fusion method for early warning of ice storms on transmission lines, which specifically includes the following steps: Step S1: Preprocess the historical data collected by meteorological stations in the area to be detected to generate a meteorological color image QYColor; Step S2: Use a clustering algorithm to segment the meteorological color image QYColor to obtain each sub-image block, and store all sub-image blocks in a spatial reference list; And the meteorological color image QYColor is divided into grids; Step S3: Establish the semantic conversion module YUZHModel for ice disaster conditions of transmission lines. The YUZHModel is used to convert meteorological information collected by meteorological stations into structured semantic description text. Step S4: Establish the cross-mesh assimilation semantic representation module KWGModel. The module KWGModel obtains cross-mesh assimilation semantic features based on the mesh partitioning results and the module YUZHModel. Step S5: Establish the ice disaster feature semantic fusion module BZModel. The ice disaster feature semantic fusion module BZModel is used to convert cross-grid assimilation semantic features into high-dimensional semantic vectors, and obtain the DTFeature field of the decision table based on the high-dimensional semantic vectors. The DTDecision field of the decision table is determined according to the historical ice disaster situation in the area to be detected. Step S6: Establish a semantic embedding decision model NNModel for ice disasters on transmission lines, and train the established model using the information in the decision table; Step S7: Obtain real-time data collected by meteorological stations in the area to be detected, obtain the input of the trained model NNModel based on the real-time collected data, and output the decision result through the trained model NNModel.

[0011] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that the specific process of step S1 is as follows: Step S101: Divide the area to be detected into grids and obtain the historical meteorological data corresponding to each grid. The meteorological data includes temperature SWD, humidity SSD, air pressure SQY and wind speed TFS. The unit of temperature is degrees Celsius, the unit of humidity is relative humidity percentage, the unit of air pressure is hectopascals, and the unit of wind speed is meters per second. It should be noted that after the grid is divided, it is necessary to ensure that there is a meteorological station in each grid. Since the distribution of meteorological stations in the area to be detected may be uneven, if a grid contains only one meteorological station, then the temperature SWD, humidity SSD, and air pressure SQY of the grid are the temperature SWD, humidity SSD, and air pressure SQY collected by that meteorological station. If a grid contains more than one meteorological station, then the temperature SWD of the grid is the average of the temperature SWD collected by all meteorological stations in the grid, the humidity SSD of the grid is the average of the humidity SSD collected by all meteorological stations in the grid, and the air pressure SQY of the grid is the average of the air pressure SQY collected by all meteorological stations in the grid. Step S102: Each grid cell is treated as a pixel in the meteorological color image QYColor. In the meteorological color image QYColor, the R channel value of the pixel is determined based on the average temperature within the grid cell, the G channel value of the pixel is determined based on the average humidity within the grid cell, and the B channel value of the pixel is determined based on the average air pressure within the grid cell.

[0012] The other steps and parameters are the same as in Specific Implementation Method 1.

[0013] Specifically, if a grid contains only one weather station a, the temperature of the grid is the average temperature of all time points in the temperature time series data collected by weather station a; if a grid contains weather stations b and c, the temperature of the grid is the average temperature of all time points in the temperature time series data collected by weather stations b and c; similarly, the temperature of a grid containing more weather stations can be calculated. Next, set upper limits for temperature, humidity, and air pressure values. These upper limits should be set slightly higher than the historical highest temperature, humidity, and air pressure values. Then, normalize the average temperature within the grid to between 0 and 255 based on the set temperature upper limit (let the set temperature upper limit value be 255), and use the normalized value as the R channel value of the corresponding pixel in the grid. Similarly, normalize the average humidity within the grid to between 0 and 255 based on the set humidity upper limit, and use the normalized value as the G channel value of the corresponding pixel in the grid. Finally, normalize the average air pressure within the grid to between 0 and 255 based on the set air pressure upper limit, and use the normalized value as the B channel value of the corresponding pixel in the grid.

[0014] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that the specific process of step S2 is as follows: Step S201: Use the KMeans algorithm to segment the meteorological color image QYColor, and put each segmented sub-image block into the spatial reference list SegmentList; Step S202: Divide the area to be detected into horizontal and vertical grids; Each grid has a width of KWidth (in this invention, KWidth is 1 / 10 of the total width of the area to be detected by default), and each grid has a height of KHeight (in this invention, KHeight is 1 / 10 of the total height of the area to be detected by default). Step S203: Place each grid obtained in step S202 into the grid list GridList. Each list item includes the grid coverage area WGQY, the average temperature WGWD, the average humidity WGSSD, and the average air pressure WGSQY within the grid. Similarly, the average temperature within the grid is also the average of all time-point temperature data collected by all meteorological stations within the grid, the average humidity within the grid is also the average of all time-point humidity data collected by all meteorological stations within the grid, and the average air pressure within the grid is also the average of all time-point air pressure data collected by all meteorological stations within the grid. Step S204: Each list item in the grid ice disaster status list WGBZ is BZFS. When BZFS=1, it means that an ice disaster has occurred. When BZFS=0, it means that no ice disaster has occurred. Steps S205 and S2 are now complete.

[0015] Other steps and parameters are the same as in specific implementation method one or two.

[0016] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that the specific process of step S3 is as follows: Step S301: Define the input and output of the YUZHModel semantic conversion module for ice disaster conditions of transmission lines; The input to the semantic conversion module YUZHModel for ice disaster conditions of transmission lines is a numerical value YUZHInput; The output of the semantic conversion module YUZHModel for ice disaster conditions of transmission lines is a semantic description of the numerical value YUZHInput, YUZHOutput. Step S302: Define two strings, String1 and String2; Step S303, the assignment rule for string String1 is as follows: A. When the input value YUZHInput is a temperature: If -10 < YUZHInput < -2, then the string String1 = "At the current temperature, there is a risk of ice storms occurring on power transmission lines"; Otherwise, if YUZHInput≤-10 or YUZHInput≥-2, then the string String1 = "There is no risk of transmission line ice disaster at the current temperature"; B. When the input value YUZHInput is humidity: If 80 < YUZHInput < 95, then the string String1 = "Under the current humidity, there is a risk of ice storms occurring on power transmission lines"; Otherwise, if YUZHInput≤80 or YUZHInput≥95, then the string String1 = "Under the current humidity, there is no risk of power transmission line ice disaster". C. When the input value YUZHInput is air pressure: If 990 < YUZHInput < 1010, then the string String1 = "Under the current air pressure, there is a risk of ice disaster on power transmission lines". Otherwise, if YUZHInput≤990 or YUZHInput≥1010, then the string String1 = "Under the current air pressure, there is no risk of ice disaster on power transmission lines"; Step S304, the assignment rule for string String2 is as follows: D. When the input value YUZHInput is a temperature: If the following condition is met: the value of abs(YUZHInput - TAVWD of AverageSeqence) / (TAVWD of AverageSeqence) is greater than 0.1, then the string String2 = "The temperature value fluctuates greatly compared to the average temperature of all meteorological stations in the area to be detected."; Otherwise, the string String2 = "The temperature value fluctuates slightly compared to the average temperature of all meteorological stations in the area to be detected."; Where, TAVWD of AverageSeqence represents the average temperature of all meteorological stations in the area to be detected, / represents the division operation of the two items before and after, and abs(·) represents taking the absolute value; E. When the input value YUZHInput is humidity: If the following condition is met: the value of abs(YUZHInput - TAVSD of AverageSeqence) / (TAVSD of AverageSeqence) is greater than 0.15, then the string String2 = "The humidity value fluctuates greatly compared to the average humidity of all meteorological stations in the area to be detected."; Otherwise, the string String2 = "The humidity value fluctuates slightly compared to the average humidity of all meteorological stations in the area to be detected."; Where, TAVSD of AverageSeqence represents the average humidity of all meteorological stations in the area to be detected; F. When the input value YUZHInput is air pressure: If the following condition is met: the value of abs(YUZHInput - TAVQY of AverageSeqence) / (TAVQY of AverageSeqence) is greater than 0.01, then the string String2 = "The air pressure value fluctuates greatly compared to the average air pressure of all meteorological stations in the area to be detected."; Otherwise, the string String2 = "The air pressure value fluctuates slightly compared to the average air pressure of all meteorological stations in the area to be detected."; Where, TAVQY of AverageSeqence represents the average air pressure of all meteorological stations in the area to be detected; G. When the input value YUZHInput is wind speed: If the value of abs(YUZHInput - TAVFS of AverageSeqence) / (TAVFS of AverageSeqence) is greater than 0.4, then the string String2 = "The wind speed value fluctuates greatly compared to the average wind speed of all meteorological stations in the area to be detected."; Otherwise, the string String2 = "The wind speed value fluctuates slightly compared to the average wind speed of all meteorological stations in the area to be detected."; In this context, AverageSeqence's TAVFS represents the average wind speed of all meteorological stations within the area to be detected. Step S305: Let the semantic description of the input value YUZHInput be YUZHOutput = String1 + String2; Step S306: Output YUZHOutput as the result of the YUZHModel module; Steps S307 and S3 are now complete.

[0017] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0018] Specific Implementation Method 5: This implementation method differs from Specific Implementation Methods 1 to 4 in that the number of meteorological stations contained in the area to be detected is denoted as NStation. Each meteorological station collects meteorological data at NSeq time points, and the elements at each time point include temperature TWD, humidity TSD, air pressure TQY, and wind speed TFS. Taking temperature data as an example, the average temperature of all meteorological stations in the area to be tested is: in, For the first The weather station was at the first Temperature data collected at various time points This indicates the total number of meteorological stations within the area to be monitored. This represents the total number of time points at which data was collected for each time point.

[0019] The other steps and parameters are the same as those in one of the specific implementation methods one to four.

[0020] The calculation methods for average humidity, average air pressure, and average wind speed are the same as those for average temperature.

[0021] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One to Five in that the specific process of step S4 is as follows: Step S401: Define the input and output of the cross-mesh assimilation semantic representation module KWGModel; The input to the cross-grid assimilation semantic expression module KWGModel is: grid temperature mean WGWD, humidity mean WGSSD, air pressure mean WGSQY, and meteorological station data within the cross-grid region KWGQY; The output of the cross-mesh assimilation semantic representation module KWGModel is an assimilation embedding attribute list KWGFTList; Step S402: Create the assimilation embedding attribute list KWGFTList and initialize KWGFTList to an empty list; Step S403: Obtain the total number of elements in the grid list GridList, GridTotalNum, and initialize the grid loop variable GridCounter = 1; Step S404: Calculate the average temperature (WGWD), average humidity (WGSSD), and average air pressure (WGSQY) of all meteorological stations within the GridCounter grid. Step S405: Use the module YUZHModel to process the average temperature WGWD, average humidity WGSSD, and average air pressure WGSQY of all meteorological stations in the GridCounter grid respectively, and append the processing results to the assimilated embedded string TString1. Step S406: For the coverage area WGQY of the GridCounter grid, find all sub-images in the spatial reference list SegmentList that intersect with the coverage area WGQY of the GridCounter grid, and merge all the found sub-images into a cross-grid area KWGQY. Step S407: Find all meteorological stations within the coverage space of the cross-grid region KWGQY and form a list FGTZList, and obtain the total number of elements in the list FGTZList StationTotalNum; Step S408: Initialize the weather station loop variable StationCounter = 1; Step S409: Retrieve the time series data of the StationCounter meteorological station in the list FGTZList, and initialize the time series loop variable SeqCounter = 1; Step S410: Obtain the SeqCounter-th element of the time series data, process each attribute data in the SeqCounter-th element using the YUZHModel module, and append the processing result to the assimilated embedded string TString2; Step S411: Increment SeqCounter by 1; If SeqCounter ≤ NSequceNumber, proceed to step S410; otherwise, proceed to step S412. Wherein, NSequceNumber represents the total number of time points corresponding to the time series data of the StationCounter meteorological station; Step S412: Increment StationCounter by 1; If StationCounter ≤ StationTotalNum, proceed to step S409; otherwise, proceed to step S413. Step S413: Add a list item element to the list KWGFTList. Each list item element includes two attributes. The two attributes include the local grid attribute and the cross-grid assimilation attribute, where the local grid attribute is BWGSX=TString1 and the cross-grid assimilation attribute is KWGSHSX=TString2; Step S414: Increment GridCounter by 1; If GridCounter ≤ GridTotalNum, proceed to step S404; otherwise, proceed to step S415. Step S415: Use the list KWGFTList as the output of the cross-mesh assimilation semantic representation module KWGModel; Step S416, Step S4 ends.

[0022] The other steps and parameters are the same as those in one of the specific implementation methods one to five.

[0023] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One to Six in that the specific process of step S5 is as follows: Step S501: Define the input and output of the ice disaster feature semantic fusion module BZModel respectively; The input to the ice disaster feature semantic fusion module BZModel is a list KWGFTList; The output of the ice disaster feature semantic fusion module BZModel is the decision information table DecisionTable; Step S502: Create a decision information table DecisionTable and initialize the decision information table DecisionTable as an empty decision table containing two fields: input attribute DTFeature and decision DTDecision; Initialize the list KWGItem to an empty list; Step S503: Set the counter BZModelCounter1=1 for BZModel; Step S504: Take out the BZModelCounter1 element of the list KWGFTList and put it into the list KWGItem; Step S505: Use a text embedding model to convert the attribute BWGSX in the list KWGItem into a vector Vector1. The text embedding model used in this invention is Sentence-BERT (Sentence-Bidirectional Encoder Representations from Transformers). Step S506: Use the text embedding model to convert the attribute KWGSHSX in the list KWGItem into a vector Vector2; Both Vector1 and Vector2 have a dimension of NDimension, and the default value of NDimension is 768. Step S507: Merge vectors Vector1 and Vector2 into a single attribute tensor NTensor with dimensions [2, NDimension]. The first horizontal dimension of the tensor NTensor corresponds to vector Vector1, and the second horizontal dimension corresponds to vector Vector2. Feature fusion is performed on the attribute tensor NTensor, vector Vector1, and vector Vector2 using a self-attention mechanism to obtain the fused features. ; Step S508: Create a row of data for the Decision Information Table DecisionTable. Each newly created row of data includes the DTFeature field and the DTDecision field. Wherein, DTFeature= DTDecision = the 1st element of the list BZFS; Step S509: Initialize the list KWGItem to be empty, and increment BZModelCounter1 by 1; If BZModelCounter1 ≤ GridTotalNum, proceed to step S504; otherwise, proceed to step S510. Step S510: Use the decision information table DecisionTable as the output result of the ice disaster feature semantic fusion module BZModel; Step S511 and Step S5 are complete.

[0024] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0025] Specific Implementation Method Eight: This implementation method differs from Specific Implementation Methods One through Seven in that it uses a self-attention mechanism to fuse the attribute tensor NTensor, vector Vector1, and vector Vector2 to obtain the fused features. Specifically: in, The fused features are represented by Tanh, which is the hyperbolic tangent function, and L2 represents the L2 norm of the calculated vector. Representation matrix Each element in the expression is added to 1.

[0026] The other steps and parameters are the same as those in any of the specific implementation methods one to seven.

[0027] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One to Eight in that the specific process of step S6 is as follows: Step S601: The transmission line ice disaster semantic embedding decision model NNModel is a four-layer neural network: The first layer is the input layer; The second layer is a fully connected layer; The third layer is a fully connected layer; The fourth layer is the output layer; The input to the input layer is the DTFeature field of the DecisionTable. The DTFeature field then passes through the second layer (which consists of ten neurons and uses the Sigmoid activation function) and the third layer (which also consists of ten neurons and uses the Sigmoid activation function). The output of the third layer is then output through the output layer. Step S602: Train the model NNModel using the data from the decision information table DecisionTable. The training label is the DTDecision field of the decision information table DecisionTable, so that it has decision-making ability.

[0028] The other steps and parameters are the same as those in one of the specific implementation methods one to eight.

[0029] Specific Implementation Method Ten: This implementation method differs from Specific Implementation Methods One to Nine in that the specific process of step S7 is as follows: Step S701: Obtain the DTFeature field of the prediction decision information table DecisionTable based on real-time collected meteorological data; Step S702: Obtain the number of rows in the prediction decision information table DecisionTableNumber, and set a counter DecisionTableCounter=1; Step S703: Store the DecisionTableCounter row of the prediction decision information table DecisionTable into the decision tableRow object to be decided; Step S704: Use the DTFeature field of DecisionTableRow as input to the trained model NNModel, and obtain the decision result NNModelResult through the model NNModel; Step S705: Write the decision result NNModelResult into the DTDecision field of the corresponding row of DecisionTableRow; Step S706: If the NNModelResult result is 1, output the grid coverage area WGQY of the DecisionTableCounter element of the grid list GridList, and output that the grid coverage area WGQY of the DecisionTableCounter element has the risk of ice disaster on the transmission line. If the NNModelResult result is 0, then no output is needed; Step S707: Increment DecisionTableCounter by 1; If DecisionTableCounter ≤ DecisionTableNumber, proceed to step S703; otherwise, proceed to step S708. Step S708: The entire testing process is complete.

[0030] The other steps and parameters are the same as those in any of the specific implementation methods one to nine.

[0031] The above examples of the present invention are merely illustrative of the computational model and process of the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is impossible to exhaustively list all possible implementations here. Any obvious variations or modifications derived from the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for early warning of ice storms on transmission lines through cross-grid semantic fusion, characterized in that, The method specifically includes the following steps: Step S1: Preprocess the historical data collected by meteorological stations in the area to be detected to generate a meteorological color image QYColor; Step S2: Use a clustering algorithm to segment the meteorological color image QYColor to obtain each sub-image block, and store all sub-image blocks in a spatial reference list; And the meteorological color image QYColor is divided into grids; Step S3: Establish the semantic conversion module YUZHModel for ice disaster conditions of transmission lines. The YUZHModel is used to convert meteorological information collected by meteorological stations into structured semantic description text. Step S4: Establish the cross-mesh assimilation semantic representation module KWGModel. The module KWGModel obtains cross-mesh assimilation semantic features based on the mesh partitioning results and the module YUZHModel. Step S5: Establish the ice disaster feature semantic fusion module BZModel. The ice disaster feature semantic fusion module BZModel is used to convert cross-grid assimilation semantic features into high-dimensional semantic vectors, and obtain the DTFeature field of the decision table based on the high-dimensional semantic vectors. The DTDecision field of the decision table is determined according to the historical ice disaster situation in the area to be detected. Step S6: Establish a semantic embedding decision model NNModel for ice disasters on transmission lines, and train the established model using the information in the decision table; Step S7: Obtain real-time data collected by meteorological stations in the area to be detected, obtain the input of the trained model NNModel based on the real-time collected data, and output the decision result through the trained model NNModel.

2. The method for early warning of ice storms on transmission lines based on cross-grid semantic fusion according to claim 1, characterized in that, The specific process of step S1 is as follows: Step S101: Divide the area to be detected into grids and obtain the historical meteorological data corresponding to each grid. The meteorological data includes temperature SWD, humidity SSD, air pressure SQY and wind speed TFS. Step S102: Each grid cell is treated as a pixel in the meteorological color image QYColor. In the meteorological color image QYColor, the R channel value of the pixel is determined based on the average temperature within the grid cell, the G channel value of the pixel is determined based on the average humidity within the grid cell, and the B channel value of the pixel is determined based on the average air pressure within the grid cell.

3. The method for early warning of ice storms on transmission lines based on cross-grid semantic fusion according to claim 2, characterized in that, The specific process of step S2 is as follows: Step S201: Use the KMeans algorithm to segment the meteorological color image QYColor, and put each segmented sub-image block into the spatial reference list SegmentList; Step S202: Divide the area to be detected into horizontal and vertical grids; Each grid cell has a width of KWidth and a height of KHeight. Step S203: Place each grid obtained in step S202 into the grid list GridList. Each list item includes the grid coverage area WGQY, the average temperature WGWD, the average humidity WGSSD, and the average air pressure WGSQY within the grid. Step S204: Each list item in the grid ice disaster situation list WGBZ is BZFS. When BZFS = 1, it indicates that an ice disaster has occurred; when BZFS = 0, it indicates that no ice disaster has occurred. Step S205: Step S2 ends.

4. The method for early warning of ice storms on transmission lines based on cross-grid semantic fusion according to claim 3, characterized in that, The specific process of the said Step S3 is as follows: Step S301: Define the input and output of the transmission line ice disaster situation semantic conversion module YUZHModel. The input of the transmission line ice disaster situation semantic conversion module YUZHModel is a numerical value YUZHInput. The output of the transmission line ice disaster situation semantic conversion module YUZHModel is the semantic description YUZHOutput of the numerical value YUZHInput. Step S302: Define two strings String1 and String2. Step S303: The assignment rule of the string String1 is as follows: A. When the input numerical value YUZHInput is temperature: If -10 < YUZHInput < -2, then the string String1 = "At the current temperature, there is a risk of transmission line ice disaster"; Otherwise, if YUZHInput ≤ -10 or YUZHInput ≥ -2, then the string String1 = "At the current temperature, there is no risk of transmission line ice disaster"; B. When the input numerical value YUZHInput is humidity: If 80 < YUZHInput < 95, then the string String1 = "At the current humidity, there is a risk of transmission line ice disaster"; Otherwise, if YUZHInput ≤ 80 or YUZHInput ≥ 95, then the string String1 = "At the current humidity, there is no risk of transmission line ice disaster"; C. When the input numerical value YUZHInput is air pressure: If 990 < YUZHInput < 1010, then the string String1 = "At the current air pressure, there is a risk of transmission line ice disaster"; Otherwise, if YUZHInput ≤ 990 or YUZHInput ≥ 1010, then the string String1 = "At the current air pressure, there is no risk of transmission line ice disaster"; Step S304: The assignment rule of the string String2 is as follows: D. When the input numerical value YUZHInput is temperature: If it satisfies that the value of abs(YUZHInput - TAVWD of AverageSeqence) / (TAVWD of AverageSeqence) is greater than 0.1, then the string String2 = "The temperature value fluctuates greatly compared with the average temperature of all meteorological stations in the area to be detected"; Otherwise, the string String2 = "The temperature value fluctuates little compared with the average temperature of all meteorological stations in the area to be detected"; Where, TAVWD of AverageSeqence represents the average temperature of all meteorological stations in the area to be detected, / represents the division operation of the two items before and after, and abs(·) represents taking the absolute value; E. When the input numerical value YUZHInput is humidity: If the following condition is met: the value of abs(YUZHInput - TAVSD of AverageSeqence) / (TAVSD of AverageSeqence) is greater than 0.15, then the string String2 = "The humidity value fluctuates greatly compared to the average humidity of all meteorological stations in the area to be detected"; Otherwise, the string String2 = "The humidity value fluctuates little compared to the average humidity of all meteorological stations in the area to be detected"; Where, TAVSD of AverageSeqence represents the average humidity of all meteorological stations in the area to be detected; F. When the input value YUZHInput is air pressure: If the following condition is met: the value of abs(YUZHInput - TAVQY of AverageSeqence) / (TAVQY of AverageSeqence) is greater than 0.01, then the string String2 = "The air pressure value fluctuates greatly compared to the average air pressure of all meteorological stations in the area to be detected"; Otherwise, the string String2 = "The air pressure value fluctuates little compared to the average air pressure of all meteorological stations in the area to be detected"; Where, TAVQY of AverageSeqence represents the average air pressure of all meteorological stations in the area to be detected; G. When the input value YUZHInput is wind speed: If the value of abs(YUZHInput - TAVFS of AverageSeqence) / (TAVFS of AverageSeqence) is greater than 0.4, then the string String2 = "The wind speed value fluctuates greatly compared to the average wind speed of all meteorological stations in the area to be detected"; Otherwise, the string String2 = "The wind speed value fluctuates little compared to the average wind speed of all meteorological stations in the area to be detected"; Where, TAVFS of AverageSeqence represents the average wind speed of all meteorological stations in the area to be detected; Step S305. Let the semantic description YUZHOutput of the input value YUZHInput = String1 + String2; Step S306. Output YUZHOutput as the result of the YUZHModel module; Step S307. End of step S3.

5. The method for early warning of ice storms on transmission lines based on cross-grid semantic fusion according to claim 4, characterized in that, The average temperature of all meteorological stations in the area to be detected is: in, For the first The weather station was at the first Temperature data collected at various time points This indicates the total number of meteorological stations within the area to be monitored. This represents the total number of time points at which data was collected for each time point.

6. The method for early warning of ice storms on transmission lines based on cross-grid semantic fusion according to claim 5, characterized in that, The specific process of step S4 is as follows: Step S401. Define the input and output of the cross-grid assimilation semantic expression module KWGModel; The input of the cross-grid assimilation semantic expression module KWGModel is: grid temperature average WGWD, humidity average WGSSD, air pressure average WGSQY, data of each meteorological station in the cross-grid area KWGQY; The output of the cross-grid assimilation semantic expression module KWGModel is the assimilation embedding attribute list KWGFTList; Step S402. Establish the assimilation embedding attribute list KWGFTList and initialize KWGFTList as an empty list; Step S403: Obtain the total number of elements in the grid list GridList, GridTotalNum, and initialize the grid loop variable GridCounter = 1; Step S404: Calculate the average temperature (WGWD), average humidity (WGSSD), and average air pressure (WGSQY) of all meteorological stations within the GridCounter grid. Step S405: Use the module YUZHModel to process the average temperature WGWD, average humidity WGSSD, and average air pressure WGSQY of all meteorological stations in the GridCounter grid respectively, and append the processing results to the assimilated embedded string TString1. Step S406: For the coverage area WGQY of the GridCounter grid, find all sub-images in the spatial reference list SegmentList that intersect with the coverage area WGQY of the GridCounter grid, and merge all the found sub-images into a cross-grid area KWGQY. Step S407: Find all meteorological stations within the coverage space of the cross-grid region KWGQY and form a list FGTZList, and obtain the total number of elements in the list FGTZList StationTotalNum; Step S408: Initialize the weather station loop variable StationCounter = 1; Step S409: Retrieve the time series data of the StationCounter meteorological station in the list FGTZList, and initialize the time series loop variable SeqCounter = 1; Step S410: Obtain the SeqCounter-th element of the time series data, process each attribute data in the SeqCounter-th element using the YUZHModel module, and append the processing result to the assimilated embedded string TString2; Step S411: Increment SeqCounter by 1; If SeqCounter ≤ NSequceNumber, proceed to step S410; otherwise, proceed to step S412. Wherein, NSequceNumber represents the total number of time points corresponding to the time series data of the StationCounter meteorological station; Step S412: Increment StationCounter by 1; If StationCounter ≤ StationTotalNum, proceed to step S409; otherwise, proceed to step S413. Step S413: Add a list item element to the list KWGFTList. Each list item element includes two attributes. The two attributes include the local grid attribute and the cross-grid assimilation attribute, where the local grid attribute is BWGSX=TString1 and the cross-grid assimilation attribute is KWGSHSX=TString2; Step S414: Increment GridCounter by 1; If GridCounter ≤ GridTotalNum, proceed to step S404; otherwise, proceed to step S415. Step S415: Use the list KWGFTList as the output of the cross-mesh assimilation semantic representation module KWGModel; Step S416, Step S4 ends.

7. The method for early warning of ice storms on transmission lines based on cross-grid semantic fusion according to claim 6, characterized in that, The specific process of step S5 is as follows: Step S501: Define the input and output of the ice disaster feature semantic fusion module BZModel respectively; The input to the ice disaster feature semantic fusion module BZModel is a list KWGFTList; The output of the ice disaster feature semantic fusion module BZModel is the decision information table DecisionTable; Step S502: Create a decision information table DecisionTable and initialize the decision information table DecisionTable as an empty decision table containing two fields: input attribute DTFeature and decision DTDecision; Initialize the list KWGItem to an empty list; Step S503: Set the counter BZModelCounter1=1 for BZModel; Step S504: Take out the BZModelCounter1 element of the list KWGFTList and put it into the list KWGItem; Step S505: Use the text embedding model to convert the attribute BWGSX in the list KWGItem into a vector Vector1; Step S506: Use the text embedding model to convert the attribute KWGSHSX in the list KWGItem into a vector Vector2; Both Vector1 and Vector2 have a dimension of NDimension; Step S507: Merge vectors Vector1 and Vector2 into a single attribute tensor NTensor with dimensions [2, NDimension]. The first horizontal dimension of the tensor NTensor corresponds to vector Vector1, and the second horizontal dimension corresponds to vector Vector2. Feature fusion is performed on the attribute tensor NTensor, vector Vector1, and vector Vector2 using a self-attention mechanism to obtain the fused features. ; Step S508: Create a row of data for the Decision Information Table DecisionTable. Each newly created row of data includes the DTFeature field and the DTDecision field. Wherein, DTFeature= DTDecision = the 1st element of the list BZFS; Step S509: Initialize the list KWGItem to be empty, and increment BZModelCounter1 by 1; If BZModelCounter1 ≤ GridTotalNum, proceed to step S504; otherwise, proceed to step S510. Step S510: Use the decision information table DecisionTable as the output result of the ice disaster feature semantic fusion module BZModel; Step S511 and Step S5 are complete.

8. The method for early warning of ice storms on transmission lines based on cross-grid semantic fusion according to claim 7, characterized in that, The method involves fusing features from the attribute tensor NTensor, vector Vector1, and vector Vector2 using a self-attention mechanism to obtain the fused features. Specifically: in, The fused features are represented by Tanh, which is the hyperbolic tangent function, and L2 represents the L2 norm of the calculated vector. Representation matrix Each element in the expression is added to 1.

9. The method for early warning of ice storms on transmission lines based on cross-grid semantic fusion according to claim 8, characterized in that, The specific process of step S6 is as follows: Step S601: The transmission line ice disaster semantic embedding decision model NNModel is a four-layer neural network: The first layer is the input layer; The second layer is a fully connected layer; The third layer is a fully connected layer; The fourth layer is the output layer; The input to the input layer is the DTFeature field of the Decision information table DecisionTable. The DTFeature field then passes through the second and third layers in sequence. The output of the third layer is then output through the output layer. Step S602: Train the NNModel model using the data from the Decision Information Table. The training label is the DTDecision field of the Decision Information Table.

10. A method for early warning of ice storms on transmission lines based on cross-grid semantic fusion according to claim 9, characterized in that, The specific process of step S7 is as follows: Step S701: Obtain the DTFeature field of the prediction decision information table DecisionTable based on real-time collected meteorological data; Step S702: Obtain the number of rows in the prediction decision information table DecisionTableNumber, and set a counter DecisionTableCounter=1; Step S703: Store the DecisionTableCounter row of the prediction decision information table DecisionTable into the decision tableRow object to be decided; Step S704: Use the DTFeature field of DecisionTableRow as input to the trained model NNModel, and obtain the decision result NNModelResult through the model NNModel; Step S705: Write the decision result NNModelResult into the DTDecision field of the corresponding row of DecisionTableRow; Step S706: If the NNModelResult result is 1, output the grid coverage area WGQY of the DecisionTableCounter element of the grid list GridList, and output that the grid coverage area WGQY of the DecisionTableCounter element has the risk of ice disaster on the transmission line. If the NNModelResult result is 0, then no output is needed; Step S707: Increment DecisionTableCounter by 1; If DecisionTableCounter ≤ DecisionTableNumber, proceed to step S703; otherwise, proceed to step S708. Step S708: The entire testing process is complete.

Images