Monitoring and early warning method and system for icing condition of power transmission line

Abstract

The invention provides a power transmission line icing condition monitoring and early warning method and system, and relates to the technical field of power transmission line icing early warning. According to the scheme, line image data are acquired based on a power transmission panoramic system, and grid meteorological data are acquired from a meteorological center; based on a pre-trained icing identification model and in combination with a multi-head attention mechanism, carrying out icing identification on the line image data to obtain an image identification result, and based on a support vector machine algorithm, predicting an icing condition according to grid meteorological data to obtain a meteorological prediction result; and matching the grid meteorological data with the monitoring and photographing device point location to obtain positioning information, fusing a meteorological prediction result with an image recognition result for correction, and performing icing degree division according to a preset threshold value to obtain an icing degree report. According to the method, the traditional machine learning technology, the deep learning technology and the data processing and analysis technology are comprehensively utilized, the precision, the real-time performance and the reliability of icing monitoring are improved, meanwhile, electric power accidents caused by icing are reduced, and support is provided for development of the electric power career.

Description

Technical Field

[0001] The present invention relates to the technical field of transmission line icing warning technology, and in particular to a monitoring and warning method and system for icing conditions of transmission lines. Background Art

[0002] With the rapid development of my country's economy, the demand for electricity is growing. As an important carrier of power transmission, the safe and stable operation of transmission lines is of vital importance. However, due to the impact of global climate change, extreme weather events occur frequently, especially in cold areas in winter. Transmission lines are easily affected by icing, which leads to increased line weight, elongated conductors, deformation of hardware and even disconnection accidents, which seriously affect the safe and stable operation of the power grid.

[0003] In recent years, my country has achieved certain research results in the field of transmission line icing monitoring technology, which can reflect the icing situation of transmission lines to a certain extent. However, there are still some shortcomings, such as the real-time performance needs to be improved, the lack of unified analysis and processing methods, and the difficulty in accurately assessing the icing situation. In order to reduce the occurrence of icing accidents and ensure power supply, it is of great significance to study a method for monitoring and improving the early warning capabilities of transmission line icing. Therefore, it is urgent to develop a solution to solve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide a monitoring and early warning method and system for the icing condition of power transmission lines, which effectively improves the accuracy, real-time nature and reliability of icing monitoring by comprehensively using traditional machine learning technology, modern deep learning technology and data processing and analysis technology.

[0005] The present invention provides a method and system for monitoring and warning of ice coverage of power transmission lines, which adopts the following technical solutions: In a first aspect, a method for monitoring and early warning of ice coverage of a power transmission line specifically comprises: Acquire line image data based on the transmission panoramic system and obtain grid meteorological data from the meteorological center; Based on the pre-trained ice recognition model combined with the multi-head attention mechanism, ice recognition is performed on the line image data to obtain an image recognition result, and based on the support vector machine algorithm, ice conditions are predicted according to the grid meteorological data to obtain a meteorological forecast result; The grid meteorological data is matched with the monitoring device point to obtain the positioning information, the meteorological forecast result is fused with the image recognition result for correction, and the ice coverage degree is divided according to the preset threshold to obtain the ice coverage degree report; The ice location information and ice degree are sent as ice warning messages and an ice monitoring and warning report is output.

[0006] The beneficial effects of the monitoring and early warning method for icing of power transmission lines provided by the present invention are as follows: the present invention effectively improves the accuracy and real-time performance of icing monitoring of power transmission lines by comprehensively using traditional machine learning technology, modern deep learning technology, and data processing and analysis technology, which helps to timely grasp the icing situation of the lines; through data preprocessing and result fusion correction, the reliability of icing prediction is enhanced; through real-time push and scheduled push, the icing situation of power transmission lines can be fed back and the icing situation of power transmission lines can be counted in time, so that the staff can take corresponding preventive measures in time, and retrospective analysis and dynamic adjustment can be carried out at the same time. The invention provides a scientific and systematic analysis method for early warning of icing of transmission lines, which helps to formulate targeted prevention and treatment measures, ensure the safe and stable operation of the power grid, reduce power accidents caused by icing, and provide strong support for the development of my country's power industry.

[0007] Optionally, obtaining line image data based on the power transmission panoramic system and obtaining grid meteorological data from a meteorological center includes: The line image data acquired based on the power transmission panoramic system includes the power transmission line images, shooting time, and shooting points of each node shot on the power transmission panoramic line; Obtaining grid meteorological data from a meteorological center, wherein the grid meteorological data is divided into meteorological data and grid data, wherein the meteorological data includes temperature, humidity, wind speed, and wind direction, and the grid data includes forecast time, report time, longitude, and latitude; The acquired line image data are enhanced and cropped, and the raster meteorological data are cleaned, missing values ​​are filled and formatted.

[0008] Optionally, ice recognition is performed on the line image data based on a pre-trained ice recognition model in combination with a multi-head attention mechanism to obtain an image recognition result, including: The deep autoencoder replaces the input layer of the deep convolutional neural network model to reconstruct the network. After the deep convolutional neural network model extracts features, global pooling is performed to reduce the dimension of the feature map, and then the feature map is input into the multi-head attention layer for classification to construct the initial model. The initial model is pre-trained based on the known transmission line image database to obtain an ice recognition model; Ice recognition is performed on the line image data based on an ice recognition model to obtain an image recognition result.

[0009] Optionally, the initial model is pre-trained based on a known transmission line image database to obtain an ice recognition model, including: Obtain unlabeled training data based on a known transmission line image database and pre-train the deep autoencoder; After pre-training the deep autoencoder, the entire ice recognition model needs to be jointly trained.

[0010] Optionally, the entire ice recognition model is jointly trained, including: Randomly extract samples based on the known transmission line image database and divide them into training set and test set; The training set is input into the initial model for processing to obtain feature representation, and the loss function is constructed by combining the cross entropy loss and mean square error loss. The loss function is minimized based on the gradient descent method, and the initial model network weights are iteratively updated. The network parameters are updated based on the Adam optimization algorithm until the loss function tends to be stable or the preset maximum number of training rounds is reached to end the joint training. The ice recognition model is obtained by evaluation and fine-tuning based on the test set.

[0011] Optionally, the meteorological forecast result is obtained by predicting the ice cover condition according to the grid meteorological data based on a support vector machine algorithm, including: Based on the meteorological database, samples are obtained and randomly divided into training samples and test samples, a linear kernel function is selected to build an SVM model, and the SVM model is obtained by training and testing the training samples and test samples; The icing probability of the grid meteorological data is predicted based on the trained SVM model to obtain a meteorological forecast result.

[0012] Optionally, the grid meteorological data is matched with the monitoring device point to obtain positioning information, the meteorological forecast result is fused with the image recognition result for correction, and the ice coverage degree is divided according to a preset threshold to obtain an ice coverage degree report, including: Match the monitoring device point data and meteorological grid data in time and area, and match the prediction results of the monitoring device in the corresponding area with the meteorological results to obtain positioning information; Perform linear regression on the image recognition results and the meteorological forecast results, and correct the ice cover to obtain a corrected result; The correction results are divided into three levels of icing according to preset thresholds, namely, light icing, moderate icing and heavy icing, to obtain an icing degree report.

[0013] Optionally, the ice location information and ice degree are used as a message push to warn of ice and output an ice monitoring and warning report, including: Provide graded warnings for different degrees of icing, and issue icing warnings through message push mode when icing occurs; wherein the message push mode of icing warning includes real-time push and scheduled push; Develop corresponding prevention and treatment measures based on the degree of icing, and automatically initiate emergency procedures when the degree of icing reaches the emergency threshold.

[0014] Optionally, the message push mode of ice cover warning includes real-time push and scheduled push, including: The real-time push notification means that when the ice coverage reaches a preset threshold, the corresponding emergency procedure is activated, and the real-time push notification is made via SMS and WeChat platform digital assistants to notify the on-duty personnel to handle the situation accordingly. The push notification content includes the route, time, occurrence area and ice coverage; The scheduled push is to set a scheduled program to push the ice coverage statistics on the WeChat platform through the digital assistant, and the push content includes the route, time period, occurrence area, number of occurrences, and ice coverage degree.

[0015] In the second aspect, a monitoring and early warning system for icing of power transmission lines specifically comprises: The data acquisition module acquires line image data based on the transmission panoramic system and obtains grid meteorological data from the meteorological center; A model prediction module, which performs ice recognition on the line image data based on a pre-trained ice recognition model combined with a multi-head attention mechanism to obtain an image recognition result, and predicts ice conditions based on grid meteorological data based on a support vector machine algorithm to obtain a meteorological forecast result; The positioning and grading module is used to match the grid meteorological data with the monitoring device points to obtain positioning information, fuse the meteorological forecast results with the image recognition results for correction, and divide the ice coverage degree according to the preset threshold to obtain an ice coverage degree report; The warning reporting module is used to send ice location information and ice degree information in the form of message push to warn of ice and output ice monitoring warning reports.

[0016] The beneficial effects of the second aspect can refer to the description of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 An overall flow chart of a method for monitoring and early warning of icing conditions of a power transmission line provided by the present invention; Figure 2 A schematic diagram of the deep autoencoder training process for a method for monitoring and warning of icing conditions on power transmission lines provided by the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0019] The embodiment of the present invention provides a method for monitoring and warning of ice coverage of a power transmission line. Figure 1 , including the following steps: S1. Acquire line image data based on the power transmission panoramic system and obtain grid meteorological data from the meteorological center; S2. Based on the pre-trained ice recognition model combined with the multi-head attention mechanism, ice recognition is performed on the line image data to obtain an image recognition result, and ice conditions are predicted according to the grid meteorological data based on the support vector machine algorithm to obtain a meteorological forecast result; S3, matching the grid meteorological data with the monitoring device point to obtain positioning information, fusing the meteorological forecast result with the image recognition result to make corrections, and dividing the ice coverage degree according to the preset threshold to obtain an ice coverage degree report; S4. The ice location information and ice degree are pushed as a message to warn of ice and output an ice monitoring and warning report.

[0020] In some embodiments, when executing step S1, acquiring line image data based on the power transmission panoramic system and acquiring grid meteorological data from a meteorological center includes: S1-1. The line image data acquired based on the power transmission panoramic system includes the power transmission line images, shooting time, and shooting points of each node shot on the power transmission panoramic line; S1-2, obtaining grid meteorological data from the meteorological center, wherein the grid meteorological data is divided into meteorological data and grid data, wherein the meteorological data includes temperature, humidity, wind speed, and wind direction, and the grid data includes forecast time, report time, longitude, and latitude; S1-3. Enhance and crop the acquired line image data, clean, fill missing values ​​and format the raster meteorological data.

[0021] Specifically, when executing step S1 - 1 , data such as the power transmission line image Imagex, the shooting time Time, the point position point_loc, etc. of each node photographed on the power transmission panoramic line are obtained from the power transmission panoramic system.

[0022] Specifically, when executing step S1-2, raster meteorological data is obtained from the meteorological center, that is, meteorological data and raster data are obtained from the meteorological center. The meteorological data includes meteorological parameters such as temperature temp, humidity humidity, wind speed wind_s, and wind direction wind_d; the raster data includes parameters such as prediction time preTime, report time fcstTime, longitude x, and latitude y.

[0023] Specifically, when executing step S1-3, the acquired line image data is enhanced and cropped, and the raster meteorological data is cleaned, missing values ​​are filled and formatted to improve the quality of the data, enhance the availability of the data, and ensure the accuracy of the analysis.

[0024] In some embodiments, when executing step S2, it specifically includes: S2-1, performing ice recognition on the line image data based on a pre-trained ice recognition model combined with a multi-head attention mechanism to obtain an image recognition result; S2-2. Based on the support vector machine algorithm, the icing situation is predicted according to the grid meteorological data to obtain the meteorological forecast results.

[0025] Specifically, when executing step S2-1, ice recognition is performed on the line image data based on the pre-trained ice recognition model combined with the multi-head attention mechanism to obtain an image recognition result, including: S2-1-1. Replace the input layer of the deep convolutional neural network model with the deep autoencoder to reconstruct the network. After extracting the features of the deep convolutional neural network model, perform global pooling to reduce the dimension of the feature map, and then input it into the multi-head attention layer for classification to construct the initial model. S2-1-2. Pre-train the initial model based on a known power transmission line image database to obtain an ice recognition model, and perform ice recognition on the line image data based on the ice recognition model to obtain an image recognition result.

[0026] In fact, when executing step S2-1-1, the deep autoencoder (DAE) is combined with the VGG16 network to reconstruct the network. The deep autoencoder DAE is used to replace the input layer of the VGG16 network to reduce noise sensitivity and enhance the network feature extraction capability. At the same time, after the last convolutional layer of the VGG16 network, global pooling is used to reduce the dimension of the feature map and input it to the multi-head attention layer for classification.

[0027] Furthermore, the Deep Autoencoder (DAE) introduces a noise injection strategy based on the traditional autoencoder. By injecting noise into the input, the "corrupted" samples are used to reconstruct the "clean" input without noise, thereby more effectively learning the essential characteristics of the data. The basic structure of DAE is mainly composed of an encoder (encode) and a decoder (decode).

[0028] Furthermore, the encoder part of DAE uses multiple convolutional layers to gradually extract image features, connects batch normalization and ReLU activation functions after each convolutional layer, and introduces pooling layers between convolutional layers to reduce the spatial dimension of the features. Finally, it connects multiple fully connected layers to output low-dimensional feature representations. The fully connected layer of the decoder part corresponds to the last fully connected layer of the encoder, which is used to map the low-dimensional features back to a higher-dimensional space, and uses deconvolution operations to gradually increase the size of the feature map and restore the original size of the image.

[0029] Specifically, when executing step S2-1-2, the initial model is pre-trained based on a known transmission line image database to obtain an ice recognition model, including: S2-1-2-1. Obtain unlabeled training data based on a known power transmission line image database and pre-train the deep autoencoder; S2-1-2-2. After pre-training the deep autoencoder, the entire ice cover recognition model needs to be jointly trained.

[0030] In fact, see Figure 2 When executing step S2-1-2-1, DAE needs to be pre-trained on an unlabeled image dataset to learn effective feature extraction and noise reduction capabilities. For the pre-training of DAE, Gaussian noise needs to be added to the original input image data, and then the noisy input samples are encoded to extract image features, and finally a decoding operation is performed to restore them to the original image structure to obtain the final result image; throughout the process, the parameters in the encoding and decoding process are adjusted and optimized with the goal of minimizing the error between the original input image and the final result image.

[0031] Furthermore, after inputting the training data, the image is preprocessed to obtain a training input feature vector of the DAE network, and an encoding operation is performed on the training input feature vector; The processing flow of the encoding network for the input image can be expressed as: ; In the formula, the output is the high-dimensional feature of the encoded image data, is the weight matrix of the encoding network, is the bias term corresponding to the encoding network, To encode image data, is the feature vector.

[0032] Furthermore, after obtaining the encoded image data, decoding is performed to obtain a noise-reduced and optimized reconstructed image. The decoding process can be expressed as: ; In the formula, the output That is the reconstructed image optimized for noise reduction. is the bias term of the decoding network, To decode the encoded image data, is the weight matrix, It is the high-dimensional feature of image data.

[0033] Furthermore, the mean square error loss (MSE) is used to measure the difference between the reconstructed image and the original image: ; In the formula, is the original image data, The reconstructed image data after noise reduction optimization is is the total number of image data samples, Is the sequence number parameter.

[0034] In fact, when executing step S2-1-2-2, after pre-training the deep autoencoder, the ice recognition model as a whole needs to be jointly trained, and the joint training of the ice recognition model as a whole includes: Randomly extract samples based on the known transmission line image database and divide them into training set and test set; The training set is input into the initial model for processing to obtain feature representation, and the loss function is constructed by combining the cross entropy loss and mean square error loss. The loss function is minimized based on the gradient descent method, and the initial model network weights are iteratively updated. The network parameters are updated based on the Adam optimization algorithm until the loss function tends to be stable or the preset maximum number of training rounds is reached to end the joint training. The ice recognition model is obtained by evaluation and fine-tuning based on the test set.

[0035] Furthermore, based on the provided labeled ice-covered and non-ice-covered image datasets, random sampling is performed and divided into a training set M and a test set N in a ratio of 7:3; Furthermore, we define a combined loss function that combines the cross entropy loss of the classification task ( ) and the mean squared error loss for the denoising task ( ); The cross entropy loss ( ) is defined as: ; In the formula, is the sample size, is the number of categories, is the true label, is the predicted result; Furthermore, the combined loss function is defined as: ; In the formula, is a hyperparameter used to balance the importance of the two loss terms. is the cross entropy loss, is the mean square error loss; Furthermore, the training set is input into DAE to obtain feature representation, and then these features are input into the VGG16 network for processing. Finally, they are classified through the multi-head attention mechanism, the gradient is calculated according to the loss function, the weights of the DAE encoder and the VGG16 network are updated, and the parameters of the network are updated using the Adam optimization algorithm. The iterative training is repeated until the loss of the model on the training set converges or the preset number of training rounds is reached; It should be noted that during the training process, the features of DAE are combined with the intermediate layer features of VGG16 through jump connections to further fuse the features of DAE and VGG16, and then the network parameters are fine-tuned to adapt to the classification task while preserving the denoising ability of DAE. Through joint training, DAE and VGG16 networks can learn together, so that VGG16 has better performance in processing noisy images while maintaining its accuracy in image classification tasks.

[0036] Specifically, when executing step S2-2, the meteorological forecast result is obtained by predicting the ice cover situation according to the grid meteorological data based on the support vector machine algorithm, including: S2-2-1. Obtain samples based on the meteorological database and randomly divide them into training samples and test samples, select a linear kernel function to build an SVM model, and train and test the SVM model based on the training samples and test samples; S2-2-2. Based on the trained SVM model, the icing probability of the grid meteorological data is predicted to obtain the meteorological forecast results.

[0037] In fact, when executing step S2-2-1, since SVM is sensitive to the scale of the data, it is necessary to perform Z-score normalization on the features; The collected samples are divided into training set and test set in a ratio of 7:3, a linear kernel function is selected according to the data characteristics, and then the training samples are used to train the SVM model; Furthermore, the goal of SVM is to minimize the following loss function: ; Further satisfying the constraints: ; In the formula, is the true label, is the feature vector, is a function that maps the input space to a high-dimensional feature space. is the penalty parameter, is the slack variable, is the weight matrix of the encoding network, is the total number of samples, is the function intercept constant; Furthermore, the decision function of SVM is: ; In the formula, is the function intercept constant, is the weight matrix of the encoding network, is the mapping function.

[0038] Furthermore, in order to obtain the probability of ice cover, a formula similar to Platt scaling is used to convert the original output of the model into a probability estimate; Among them, the Platt scaling formula is: ; In the formula, A and B are parameters obtained by maximum likelihood estimation. To estimate the probability; In fact, when executing step S2-2-2, the grid meteorological data obtained in S1 is used to predict the probability of icing to obtain a meteorological forecast result.

[0039] In some embodiments, when executing step S3, it includes: Match the monitoring device point data and meteorological grid data in time and area, and match the prediction results of the monitoring device in the corresponding area with the meteorological results to obtain positioning information; Perform linear regression on the image recognition results and the meteorological forecast results, and correct the ice cover to obtain a corrected result; The correction results are divided into three levels of icing according to preset thresholds, namely, light icing, moderate icing and heavy icing, to obtain an icing degree report.

[0040] Specifically, the icing conditions are divided into three levels according to the degree of icing, namely light icing, moderate icing and heavy icing, and an icing degree report is obtained. When the ice thickness is less than the preset threshold (such as 10mm), it is regarded as light icing. When the ice thickness is greater than 10mm and less than 20mm (ice density 0.9g / cm³), it can be called a medium ice zone, and the ice within this range can be regarded as moderate icing. When the ice thickness reaches or exceeds 20mm (ice density 0.9g / cm³), it is called a heavy ice zone, and the ice within this range is heavy icing.

[0041] In some embodiments, when executing step S4, ice location information and ice degree are used as message push to issue ice warning and output an ice monitoring warning report, including: Provide graded warnings for different degrees of icing, and issue icing warnings through message push mode when icing occurs; wherein the message push mode of icing warning includes real-time push and scheduled push; Formulate corresponding prevention and treatment measures according to the degree of icing, and automatically initiate emergency procedures when the degree of icing reaches the emergency threshold; Specifically, the message push modes of ice cover warning include real-time push and scheduled push, including: The real-time push means that when the degree of ice coverage reaches a preset threshold, the corresponding emergency procedure is activated, and real-time push is carried out through SMS and WeChat platform digital assistants to notify the on-duty personnel to take corresponding measures. The push content includes the route, time, occurrence area and degree of ice coverage.

[0042] Furthermore, the timed push is to set a timed program to push ice coverage statistics on the WeChat platform through a digital assistant.

[0043] Furthermore, the pushed content includes ice-covered routes, time periods, occurrence areas, number of occurrences, and degree of ice coverage.

[0044] The present invention also provides a monitoring and early warning system for icing of power transmission lines, including: The data acquisition module acquires line image data based on the transmission panoramic system and obtains grid meteorological data from the meteorological center; A model prediction module, which performs ice recognition on the line image data based on a pre-trained ice recognition model combined with a multi-head attention mechanism to obtain an image recognition result, and predicts ice conditions based on grid meteorological data based on a support vector machine algorithm to obtain a meteorological forecast result; The positioning and grading module is used to match the grid meteorological data with the monitoring device points to obtain positioning information, fuse the meteorological forecast results with the image recognition results for correction, and divide the ice coverage degree according to the preset threshold to obtain an ice coverage degree report; The warning reporting module is used to send ice location information and ice degree information in the form of message push to warn of ice and output ice monitoring warning reports.

[0045] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.

Claims

1. A method for monitoring and early warning of ice coverage of power transmission lines, characterized in that: include: Acquire line image data based on the transmission panoramic system and obtain grid meteorological data from the meteorological center; Based on the pre-trained ice recognition model combined with the multi-head attention mechanism, ice recognition is performed on the line image data to obtain an image recognition result, and based on the support vector machine algorithm, ice conditions are predicted according to the grid meteorological data to obtain a meteorological forecast result; The grid meteorological data is matched with the monitoring device point to obtain the positioning information, the meteorological forecast result is fused with the image recognition result for correction, and the ice coverage degree is divided according to the preset threshold to obtain the ice coverage degree report; The ice location information and ice degree are sent as ice warning messages and an ice monitoring and warning report is output.

2. A method for monitoring and early warning of ice coverage of power transmission lines according to claim 1, characterized in that: Obtain line image data based on the transmission panoramic system and obtain raster meteorological data from the meteorological center, including: The line image data acquired based on the power transmission panoramic system includes the power transmission line images, shooting time, and shooting points of each node shot on the power transmission panoramic line; Obtaining grid meteorological data from a meteorological center, wherein the grid meteorological data is divided into meteorological data and grid data, wherein the meteorological data includes temperature, humidity, wind speed, and wind direction, and the grid data includes forecast time, report time, longitude, and latitude; The acquired line image data are enhanced and cropped, and the raster meteorological data are cleaned, missing values ​​are filled and formatted.

3. A method for monitoring and early warning of ice coverage of power transmission lines according to claim 1, characterized in that: Based on the pre-trained ice recognition model combined with the multi-head attention mechanism, ice recognition is performed on the line image data to obtain image recognition results, including: The deep autoencoder replaces the input layer of the deep convolutional neural network model to reconstruct the network. After the deep convolutional neural network model extracts features, global pooling is performed to reduce the dimension of the feature map, and then the feature map is input into the multi-head attention layer for classification to construct the initial model. The initial model is pre-trained based on the known transmission line image database to obtain an ice recognition model; Ice recognition is performed on the line image data based on an ice recognition model to obtain an image recognition result.

4. A method for monitoring and early warning of ice coverage of power transmission lines according to claim 3, characterized in that: The initial model is pre-trained based on the known transmission line image database to obtain the ice recognition model, including: Obtain unlabeled training data based on a known transmission line image database and pre-train the deep autoencoder; After pre-training the deep autoencoder, the entire ice recognition model needs to be jointly trained.

5. A method for monitoring and early warning of ice coverage of power transmission lines according to claim 4, characterized in that: Joint training of the entire ice recognition model includes: Randomly extract samples based on the known transmission line image database and divide them into training set and test set; The training set is input into the initial model for processing to obtain feature representation, and the loss function is constructed by combining the cross entropy loss and mean square error loss. The loss function is minimized based on the gradient descent method, and the initial model network weights are iteratively updated. The network parameters are updated based on the Adam optimization algorithm until the loss function tends to be stable or the preset maximum number of training rounds is reached to end the joint training. The ice recognition model is obtained by evaluation and fine-tuning based on the test set.

6. A method for monitoring and early warning of ice coverage of power transmission lines according to claim 1, characterized in that: Based on the support vector machine algorithm, the icing situation is predicted according to the grid meteorological data to obtain the meteorological forecast results, including: Based on the meteorological database, samples are obtained and randomly divided into training samples and test samples, a linear kernel function is selected to build an SVM model, and the SVM model is obtained by training and testing the training samples and test samples; The icing probability of the grid meteorological data is predicted based on the trained SVM model to obtain a meteorological forecast result.

7. A method for monitoring and early warning of ice coverage of power transmission lines according to claim 1, characterized in that: The grid meteorological data is matched with the monitoring device point to obtain the positioning information, the meteorological forecast results are fused with the image recognition results for correction, and the ice coverage degree is divided according to the preset threshold to obtain the ice coverage degree report, including: Match the monitoring device point data and meteorological grid data in time and area, and match the prediction results of the monitoring device in the corresponding area with the meteorological results to obtain positioning information; Perform linear regression on the image recognition results and the meteorological forecast results, and correct the ice cover to obtain a corrected result; The correction results are divided into three levels of icing according to preset thresholds, namely, light icing, moderate icing and heavy icing, to obtain an icing degree report.

8. The method for monitoring and warning of ice coverage of power transmission lines according to claim 1, characterized in that: The ice location information and ice degree are pushed as messages to warn of ice and output ice monitoring and warning reports, including: Provide graded warnings for different degrees of icing, and issue icing warnings through message push mode when icing occurs; wherein the message push mode of icing warning includes real-time push and scheduled push; Develop corresponding prevention and treatment measures based on the degree of icing, and automatically initiate emergency procedures when the degree of icing reaches the emergency threshold.

9. A method for monitoring and early warning of ice coverage of power transmission lines according to claim 8, characterized in that: The message push modes of ice cover warning include real-time push and scheduled push, including: The real-time push notification means that when the ice coverage reaches a preset threshold, the corresponding emergency procedure is activated, and the real-time push notification is made via SMS and WeChat platform digital assistants to notify the on-duty personnel to handle the situation accordingly. The push notification content includes the route, time, occurrence area and ice coverage; The scheduled push is to set a scheduled program to push the ice coverage statistics on the WeChat platform through the digital assistant, and the push content includes the route, time period, occurrence area, number of occurrences, and ice coverage degree.

10. A monitoring and early warning system for icing of power transmission lines, characterized in that: include: The data acquisition module acquires line image data based on the transmission panoramic system and obtains grid meteorological data from the meteorological center; A model prediction module, which performs ice recognition on the line image data based on a pre-trained ice recognition model combined with a multi-head attention mechanism to obtain an image recognition result, and predicts ice conditions based on grid meteorological data based on a support vector machine algorithm to obtain a meteorological forecast result; The positioning and grading module is used to match the grid meteorological data with the monitoring device points to obtain positioning information, fuse the meteorological forecast results with the image recognition results for correction, and divide the ice coverage degree according to the preset threshold to obtain an ice coverage degree report; The warning reporting module is used to send ice location information and ice degree information in the form of message push to warn of ice and output ice monitoring warning reports.

Images