Three-span line state intelligent evaluation system based on numerical weather forecast and machine learning

Abstract

The invention provides a three-span line state intelligent evaluation system based on numerical weather forecast and machine learning, and relates to the technical field of data processing, and the system comprises the steps: carrying out the arrangement of the vibration mode, frequency and amplitude of a lead, and forming a vibration analysis data set, which is used for simulating the dynamic behavior of the lead in a finite element; establishing a finite element model of the three-span line, operating finite element analysis, and simulating response of the three-span line under different meteorological conditions and vibration loads so as to identify key areas and factors causing line faults; and according to the key areas and factors causing the line fault, evaluating the safety state of the line, and determining a potential fault type. According to the invention, the efficiency and accuracy of line state evaluation can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a three-span line status intelligent evaluation system based on numerical weather forecasting and machine learning. Background Art

[0002] With the rapid development of the power industry, the safe operation of three-span lines (i.e., power lines that cross railways, highways, and important transmission channels) has become increasingly prominent. The operating status of these lines is directly related to the stability and power supply reliability of the power system. Therefore, accurate and efficient status assessment is crucial.

[0003] However, some traditional three-span line status assessment methods rely on manual inspections and regular testing, and therefore have the following main defects:

[0004] For example, some traditional methods are unable to monitor the operating status of the line in real time, making it difficult to promptly detect and deal with potential fault hazards, thereby increasing the risk of line failures. When processing and analyzing the line status, some lack intelligent data processing and analysis methods, making it difficult to deeply explore the inherent laws and characteristics of the line operating status. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a three-span line status intelligent evaluation system based on numerical weather forecasting and machine learning, which can improve the efficiency and accuracy of line status evaluation.

[0006] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0007] In the first aspect, a three-span line status intelligent assessment system based on numerical weather forecast and machine learning includes:

[0008] The recognition module is used to automatically identify and classify abnormal conditions of the line from the pre-processed images by building, training and tuning a convolutional neural network model. The abnormal conditions include equipment damage data, foreign object intrusion data and temperature abnormality data;

[0009] The meteorological analysis module is used to obtain meteorological parameters in real time, including temperature, humidity, wind speed, precipitation, air pressure and radiation; based on the meteorological parameters, the influence of the meteorological parameters on the line operation status is analyzed to obtain the meteorological analysis results, including the influence of temperature, humidity and wind speed on the line operation status;

[0010] An evaluation module is used to monitor and identify the vibration mode and characteristics of the three-span line conductors in real time to evaluate the impact of vibration on the line operation status;

[0011] The finite element analysis module is used to organize the equipment damage data, foreign body intrusion data and temperature anomaly data to form an abnormal condition data set, which is used as the boundary condition of the finite element analysis; the meteorological analysis results obtained based on the meteorological parameter analysis are summarized to form a meteorological analysis data set, which is used as the environmental load for the finite element analysis; the vibration mode, frequency and amplitude of the conductor are organized to form a vibration analysis data set, which is used to simulate the dynamic behavior of the conductor in the finite element; a finite element model of the three-span line is established, and the finite element analysis is run to simulate the response of the three-span line under different meteorological conditions and vibration loads to identify the key areas and factors that cause line failures; based on the key areas and factors that cause line failures, the safety status of the line is evaluated and the potential fault type is determined.

[0012] Furthermore, by building, training and tuning a convolutional neural network model, the abnormal conditions of the line can be automatically identified and classified from the pre-processed images. The abnormal conditions include equipment damage data, foreign body intrusion data and temperature abnormality data, including:

[0013] The real-time image data of the three-span lines are collected by video monitoring equipment, and image preprocessing is performed to obtain preprocessed images;

[0014] Screen out samples containing equipment damage, foreign matter intrusion, and temperature abnormalities from the preprocessed images, and annotate the samples to create a labeled dataset;

[0015] Divide the labeled dataset into training, validation, and test sets;

[0016] Build a convolutional neural network model and initialize the convolutional neural network model parameters, including weights and biases;

[0017] Use the training set to train the convolutional neural network model, calculate the predicted value through forward propagation, and optimize the convolutional neural network model parameters through the back propagation algorithm to obtain the trained convolutional neural network model;

[0018] Use the test set to evaluate the trained convolutional neural network model and calculate the accuracy; adjust the trained convolutional neural network model according to the accuracy to obtain the final convolutional neural network model;

[0019] The preprocessed new image is input into the final convolutional neural network model to obtain the abnormal conditions of the line, which include equipment damage, foreign body intrusion and temperature abnormality.

[0020] Furthermore, the test set is used to evaluate the trained convolutional neural network model and calculate the accuracy, including:

[0021] Use the trained convolutional neural network model to predict each image in the test set and record the predicted label of each image; compare the predicted label of each image in the test set with the actual label;

[0022] Construct a confusion matrix to record the number of true and predicted classifications of each category. The confusion matrix includes true positive examples TP, false positive examples FP, true negative examples TN, and false negative examples FN;

[0023] According to the confusion matrix, calculate the accuracy, that is, the proportion of all correctly predicted samples to the total samples;

[0024] Initialize a list to save the accuracy of each iteration; for each of the K subsets, record it as a validation set, and perform the following steps:

[0025] Merge the remaining K-1 subsets to form a temporary "test set"; use the trained convolutional neural network model to predict each image in the temporary "test set" and record the predicted label; compare the predicted label of each image in the temporary "test set" with the true label, and update the confusion matrix; calculate the accuracy of the current iteration based on the confusion matrix, add it to the accuracy record, reset the confusion matrix to prepare for the next iteration; calculate the average accuracy of K iterations as the final accuracy of the model.

[0026] Furthermore, the vibration modes and characteristics of the three-span line conductors are monitored and identified in real time to evaluate the impact of vibration on the line operation status, including:

[0027] Define the initial temperature T of the simulated annealing algorithm o , termination temperature T f , cooling coefficient α and the number of iterations L at each temperature; set the initial solution S o , i.e., the initial position set of the vibration sensor;

[0028] Define an objective function to evaluate the quality of the position set S;

[0029] For each iteration at the current temperature T, a new solution S is generated new ;

[0030] Calculate the objective function value f(S new ), if f(S new ) is greater than the objective function value f(S current ), then accept the new solution and update S current =S new , where S current represents the current solution; if f(S new )≤f(S current), then accept the new solution, and after completing the inner cycle, lower the temperature. If the current temperature T drops to the termination temperature T f Below, the simulated annealing process is stopped and the final solution, i.e., the final position set of the vibration sensor, is output;

[0031] Install multiple vibration sensors at key locations of the three-span line to monitor the vibration data of the conductor in real time and pre-process the vibration data to obtain pre-processed vibration data;

[0032] The vibration characteristics of the pre-processed vibration data are extracted by Fourier transform, and the vibration characteristics include frequency, amplitude and phase;

[0033] Based on the vibration characteristics, a vibration analysis model is established to identify the vibration mode of the conductor and analyze the impact of vibration on the line operation status to obtain the vibration analysis results, which include the vibration mode, frequency and amplitude of the conductor.

[0034] Furthermore, based on the vibration characteristics, a vibration analysis model is established to identify the vibration mode of the conductor and analyze the impact of the vibration on the line operation status to obtain the vibration analysis results. The vibration analysis results include the vibration mode, frequency and amplitude of the conductor, including:

[0035] According to the characteristics of the vibration data, the vibration data is labeled, and the corresponding vibration mode labels are determined and labeled to form a labeled data set;

[0036] The labeled data set is divided into a training set, a validation set, and a test set. The training set is used to train the recurrent neural network model. During the training process, the weights and biases of the recurrent neural network model are continuously adjusted through the back propagation algorithm and the optimizer to obtain the trained recurrent neural network model.

[0037] Use the trained recurrent neural network model to classify new vibration data to obtain the corresponding predicted vibration pattern labels;

[0038] By comparing the predicted vibration pattern labels with predefined vibration pattern labels, different vibration patterns, including breeze vibration and dancing, are identified;

[0039] Analyze and identify the correlation between different vibration modes and line operation status;

[0040] Based on the correlation, the risk level of the line under different vibration modes is evaluated.

[0041] Furthermore, according to the characteristics of the vibration data, the vibration data is labeled, and the corresponding vibration mode label is determined and labeled to form a labeled data set, including:

[0042] Identify categories of vibration patterns, including breeze vibrations and dancing;

[0043] Define a unique label for each vibration pattern, and convert the characteristics of the vibration data, namely frequency or amplitude, into genes in the genetic algorithm. Genes are used to constitute the chromosomes of individuals.

[0044] An initial population is randomly generated, in which each individual represents a vibration pattern label sequence;

[0045] Extract key features from each vibration data sample, including time domain and frequency domain features;

[0046] Define a fitness function to evaluate how well each individual matches the vibration data features;

[0047] According to the fitness function, the corresponding individuals are selected to enter the next generation, two individuals are randomly selected as parents, and a crossover operation is performed to produce new offspring. The genes of the offspring, i.e., the labels of the vibration patterns, are randomly mutated. The selection, crossover and mutation operations are repeated until the termination condition is met. When the termination condition is met, the iteration is stopped and the final individual is output as the labeling result of the vibration data.

[0048] The final individuals are decoded into corresponding vibration pattern label sequences, and the vibration pattern label sequences are matched with the original vibration data samples to form a labeled dataset.

[0049] Furthermore, a finite element model of the three-span line was established and finite element analysis was performed to simulate the response of the three-span line under different meteorological conditions and vibration loads to identify the key areas and factors that lead to line failures, including:

[0050] Determine the detailed structure of the three-span line, including the geometry, materials and connection methods of conductors, insulators and towers;

[0051] According to the specific structure of the three-span line, a three-span line geometric model is created, and corresponding material properties are specified for each component in the three-span line geometric model, including elastic modulus, density and Poisson's ratio; the connection relationship between each component is defined, including hinged or fixed connection;

[0052] Meshing the geometric model of the three-span line and setting the boundary conditions of the geometric model of the three-span line, including fixed constraints and load application positions;

[0053] Based on the meteorological analysis dataset, different meteorological conditions, including wind speed, wind direction and temperature, are defined and applied as environmental loads to the three-span line geometry model;

[0054] Using vibration analysis data sets, determine the vibration mode, frequency and amplitude of the conductor and apply vibration loads to the conductor section;

[0055] Set the solution parameters, including the time step and the number of iterations, start the solver, and start running the simulation analysis to obtain the simulation results, which include stress distribution cloud diagrams, deformation diagrams, and vibration response curves;

[0056] Based on the simulation results, identify key areas with stress concentration, large deformation or abnormal vibration response;

[0057] Analyze the response characteristics of key areas under different meteorological conditions and vibration loads to determine the factors that lead to line failures.

[0058] Furthermore, the three-span line geometric model is meshed, including:

[0059] Set the number of particles, velocity range, and position range for the particle swarm optimization algorithm; set the learning rate and number of iterations for the gradient descent algorithm;

[0060] Generate a set of initial particles, each particle represents a grid division scheme, and the position of the particle indicates the parameters of the grid division, including the grid size and distribution;

[0061] The grid division scheme represented by each particle is evaluated through the evaluation function, the speed and position of each particle are updated, and the fitness is evaluated and the particles are updated repeatedly until the convergence condition is met to obtain the global final particle;

[0062] Extract the meshing parameters from the global final particle as the initial scheme of gradient descent;

[0063] The initial grid solution is disturbed, and the gradient of the optimization function with respect to the grid partitioning parameters is calculated by the finite difference method. The grid partitioning parameters are adjusted along the direction of gradient descent, and the gradient calculation and grid parameter adjustment are repeated until the preset number of iterations is reached to obtain the optimized grid partitioning parameters.

[0064] The three-span line geometric model is meshed using the optimized meshing parameters.

[0065] In a second aspect, a three-span line status intelligent assessment method based on numerical weather forecast and machine learning, the method comprising the following steps:

[0066] By building, training, and tuning a convolutional neural network model, it can automatically identify and classify abnormal conditions of the line from pre-processed images, including equipment damage data, foreign object intrusion data, and temperature abnormality data;

[0067] Acquire meteorological parameters in real time, including temperature, humidity, wind speed, precipitation, air pressure and radiation; analyze the impact of meteorological parameters on the line operation status based on the meteorological parameters to obtain meteorological analysis results, including the impact of temperature, humidity and wind speed on the line operation status;

[0068] Real-time monitoring and identification of the vibration modes and characteristics of three-span line conductors to assess the impact of vibration on line operation status;

[0069] The equipment damage data, foreign body intrusion data and temperature anomaly data are collated to form an abnormal condition data set, which is used as the boundary condition for finite element analysis. The meteorological analysis results obtained based on meteorological parameter analysis are summarized to form a meteorological analysis data set, which is used as the environmental load for finite element analysis. The vibration mode, frequency and amplitude of the conductor are collated to form a vibration analysis data set, which is used to simulate the dynamic behavior of the conductor in the finite element. A finite element model of the three-span line is established, and finite element analysis is run to simulate the response of the three-span line under different meteorological conditions and vibration loads to identify the key areas and factors that cause line failures. According to the key areas and factors that cause line failures, the safety status of the line is evaluated and the potential fault type is determined.

[0070] According to a third aspect, a computing device includes:

[0071] one or more processors;

[0072] The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method described.

[0073] In a fourth aspect, a computer-readable storage medium stores a program, and when the program is executed by a processor, the method described is implemented.

[0074] The above solution of the present invention includes at least the following beneficial effects:

[0075] By building, training, and tuning a convolutional neural network model, the system can automatically identify and classify abnormal line conditions, such as equipment damage, foreign object intrusion, and temperature anomalies. This approach not only improves the accuracy of recognition, but also greatly improves recognition efficiency, helping to promptly detect and handle line problems.

[0076] It is able to monitor meteorological parameters in real time and analyze their impact on the line operation status, which enables the system to respond quickly under severe weather conditions and provide strong support for preventing line failures.

[0077] By real-time monitoring and identifying the vibration modes and characteristics of the three-span line conductors, it is possible to evaluate the impact of vibration on the line's operating status, help discover potential mechanical failures or structural problems, and make timely repairs and replacements.

[0078] The combination of equipment damage data, foreign object intrusion data, temperature anomaly data, meteorological analysis results and conductor vibration analysis provides comprehensive input conditions for the finite element analysis of the line, which makes the evaluation results more accurate and helps to identify the key areas and factors that cause line failures.

[0079] Through finite element analysis, the system can simulate the response of three-span lines under different meteorological conditions and vibration loads, so as to promptly discover and deal with potential safety hazards. This can not only extend the service life of the line, but also significantly reduce the probability of failure and improve the stability and safety of the entire power grid system. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is a schematic diagram of a three-span line status intelligent assessment system based on numerical weather forecasting and machine learning provided by an embodiment of the present invention.

[0081] Figure 2 It is a flow chart of a method for intelligently evaluating the status of three-span lines based on numerical weather forecasting and machine learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0082] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0083] like Figure 2 As shown, an embodiment of the present invention proposes a three-span line status intelligent evaluation method based on numerical weather forecasting and machine learning, comprising:

[0084] Step 1: Build, train, and tune a convolutional neural network model to automatically identify and classify abnormal conditions of the line from preprocessed images. The abnormal conditions include equipment damage data, foreign object intrusion data, and temperature abnormality data.

[0085] Step 2, obtaining meteorological parameters in real time, including temperature, humidity, wind speed, precipitation, air pressure and radiation; based on the meteorological parameters, analyzing the impact of the meteorological parameters on the line operation status to obtain meteorological analysis results, including the impact of temperature, humidity and wind speed on the line operation status;

[0086] Step 3: monitor and identify the vibration mode and characteristics of the three-span line conductors in real time to evaluate the impact of vibration on the line operation status;

[0087] Step 4: Organize the equipment damage data, foreign body intrusion data and temperature anomaly data to form an abnormal condition data set, and use it as the boundary condition of the finite element analysis; summarize the meteorological analysis results obtained based on the meteorological parameter analysis to form a meteorological analysis data set, and use it as the environmental load for the finite element analysis; organize the vibration mode, frequency and amplitude of the conductor to form a vibration analysis data set, which is used to simulate the dynamic behavior of the conductor in the finite element; establish a finite element model of the three-span line, run finite element analysis, and simulate the response of the three-span line under different meteorological conditions and vibration loads to identify the key areas and factors that cause line failures; evaluate the safety status of the line and determine the potential fault type based on the key areas and factors that cause line failures.

[0088] In the embodiment of the present invention, through the convolutional neural network model, the system can automatically identify and classify the abnormal conditions of the line from the preprocessed image, greatly improving the efficiency and accuracy of abnormality detection; compared with traditional manual inspections, the introduction of machine learning models reduces misjudgments and missed detections caused by human factors, enhances the reliability of abnormality detection, and the accumulated abnormal condition data can be used for further analysis and prediction, providing data support for line maintenance and management. Real-time acquisition of meteorological parameters and analysis of their impact on the operating status of the line, can obtain meteorological parameters in real time, and quickly analyze their impact on the operating status of the line, which helps to respond to sudden weather events in a timely manner. Through the analysis of meteorological parameters, possible problems of the line under certain weather conditions can be predicted in advance, so as to carry out preventive maintenance. Under extreme weather conditions, the system can issue an alarm in advance to help operators take measures to reduce line failures and improve the safety of the power grid.

[0089] By monitoring the vibration patterns of the conductors, the system can identify abnormal vibration characteristics, which may indicate potential mechanical failures or structural problems. Continuous monitoring and analysis of vibration patterns can help to promptly identify and address potential problems, thereby extending the service life of the line. Understanding the vibration characteristics of the conductors can help operators develop more refined maintenance strategies and improve maintenance efficiency. By integrating multiple data sources (abnormal condition data, meteorological analysis data, vibration analysis data), the system can provide a more comprehensive and accurate line status assessment. Finite element analysis helps identify key areas and factors that cause line failures, thereby conducting targeted prevention and maintenance. Through continuous monitoring and evaluation of line status, operators can manage the power grid more efficiently and reduce unnecessary downtime and maintenance costs.

[0090] In a preferred embodiment of the present invention, the above step 1, by constructing, training and tuning a convolutional neural network model, automatically identifies and classifies abnormal conditions of the line from the preprocessed image, and the abnormal conditions include equipment damage data, foreign body intrusion data and temperature abnormality data, which may include:

[0091] Step 11, collect real-time image data of the three-span line through video surveillance equipment, and perform image preprocessing to obtain preprocessed images, specifically including: using video surveillance equipment installed near the three-span line, such as cameras, to capture the image data of the line in real time. These devices can be set to continuous recording or timed recording to ensure that all important events are captured; preprocessing the collected original images to improve image quality and prepare for subsequent analysis. The preprocessing steps may include denoising, contrast enhancement, edge detection, etc. These operations help to highlight key features in the image, such as areas of equipment damage, foreign objects, or abnormal temperature.

[0092] Step 12, screening out samples containing equipment damage, foreign body intrusion, and temperature anomaly conditions from the preprocessed images, and annotating the samples to create a labeled dataset, specifically including: screening out samples containing equipment damage, foreign body intrusion, and temperature anomaly conditions from the preprocessed images, which can be done by observing specific patterns or anomalies in the images; annotating the screened samples. Annotation usually involves adding bounding boxes or masks to abnormal areas in the image and assigning corresponding labels to them (such as "equipment damage", "foreign body intrusion", or "temperature anomaly"), and combining all annotated samples into a labeled dataset.

[0093] Step 13, divide the labeled dataset into training set, validation set and test set, specifically including: randomly divide the labeled dataset into three subsets: training set, validation set and test set, the training set is used to train the model, the validation set is used to adjust the model parameters and hyperparameters, and the test set is used to evaluate the final performance of the model; according to the specific needs of the project and the total amount of available data, reasonably allocate the proportions of the three subsets, and the common division ratio can be 70% (training set), 15% (validation set) and 15% (test set).

[0094] Step 14: Build a convolutional neural network model and initialize the convolutional neural network model parameters, including weights and biases, including:

[0095] Determine the size of the input image. For example, if the preprocessed image size is 224×224 pixels, the input layer should accept images of this size. Add multiple convolutional layers to extract image features. Select the appropriate convolution kernel size (such as 3×3, 5×5) and stride for each convolutional layer. Set the number of output channels of each convolutional layer, that is, the number of convolution kernels. Apply activation functions such as ReLU (Rectified Linear Unit) after each convolutional layer to increase the nonlinearity of the network. Insert pooling layers (such as maximum pooling) between consecutive convolutional layers to reduce the dimension and computational complexity of the feature map, and set the size and stride of the pooling window. Add a fully connected layer to the last part of the network to integrate the previously extracted features and perform the final classification. Determine the number of neurons in the fully connected layer. Usually, the number of neurons in the last layer is equal to the number of categories of the classification task. Use the softmax function as the activation function of the output layer to obtain the predicted probability of each category. For the weights of convolutional layers and fully connected layers, you can choose a random initialization method, such as He initialization, to adjust the initial distribution of weights according to the number of neurons in the layer; the bias parameter is usually initialized to zero or a decimal close to zero, because the main function of the bias is to adjust the offset of the output value.

[0096] Step 15, using the training set to train the convolutional neural network model, calculating the predicted value through forward propagation, and optimizing the convolutional neural network model parameters through the back propagation algorithm to obtain the trained convolutional neural network model, specifically including: inputting the image in the training set into the convolutional neural network model, calculating through each layer of the model to obtain the predicted probability of each category; comparing the predicted result of the model with the true label of the image, and calculating the loss function (such as cross entropy loss); using the gradient descent algorithm, updating the weight and bias parameters of the model according to the gradient of the loss function, and repeating the above steps until the performance of the model on the validation set reaches the preset standard or the maximum number of training rounds is reached.

[0097] Step 16, using the test set to evaluate the trained convolutional neural network model and calculate the accuracy; adjusting the trained convolutional neural network model according to the accuracy to obtain the final convolutional neural network model, specifically including: using the test set to evaluate the trained convolutional neural network model and calculate the accuracy of the model; adjusting the model according to the accuracy, such as changing the network structure, increasing or decreasing the number of layers, adjusting hyperparameters such as the learning rate, etc.

[0098] Step 17, input the preprocessed new image into the final convolutional neural network model to obtain the abnormal condition of the line, which includes equipment damage, foreign body intrusion and temperature abnormality. Specifically, the following steps are performed: the newly collected three-span line image is preprocessed in the same manner as step 11; the preprocessed new image is input into the final convolutional neural network model to obtain the model's prediction results for the abnormal condition of the line, and the model's prediction results are displayed in a visual or other form for further analysis and processing by the operator.

[0099] In the embodiment of the present invention, image data is collected in real time by video monitoring equipment to ensure the timeliness and accuracy of information. Image preprocessing can remove noise and enhance image features, providing higher quality input for subsequent automatic recognition and classification. By screening samples of specific abnormal conditions, a data set focusing on line abnormal conditions can be created, improving the pertinence and efficiency of model training. Accurately labeling samples provides reliable labels for supervised learning, which helps the model accurately identify and classify abnormal conditions. By dividing the data set into training set, validation set and test set, the generalization ability of the model can be more effectively evaluated to prevent overfitting. An independent test set can provide an unbiased estimate of model performance. Constructing a convolutional neural network model specifically for identifying line abnormal conditions can better adapt to this specific task. Through training, the model can learn to extract useful features from images and accurately classify abnormal conditions. The back propagation algorithm can effectively adjust model parameters and improve the prediction performance of the model. Use an independent test set to evaluate model performance to ensure the reliability of the model in practical applications. Adjust the model according to the test results to further improve the accuracy and generalization ability of the model. Inputting new images into the trained model can automatically detect abnormal conditions of the line, greatly improving detection efficiency. By real-time monitoring of abnormal conditions of the line, problems can be discovered and handled in a timely manner to ensure the safe and stable operation of the power grid.

[0100] In a preferred embodiment of the present invention, the above step 16, using the test set to evaluate the trained convolutional neural network model and calculate the accuracy, may include:

[0101] Step 161, using the trained convolutional neural network model to predict each image in the test set, and recording the predicted label of each image; comparing the predicted label of each image in the test set with the real label, specifically including: loading the test set containing preprocessed images, these images have real labels; using the trained convolutional neural network model to predict each image in the test set, the prediction is completed through forward propagation, that is, the input image passes through each layer of the network to obtain the output; for each image, converting the output (category probability) predicted by the model into a specific category label, and recording these predicted labels; comparing the predicted label of each image with its corresponding real label to determine whether the prediction is correct.

[0102] Step 162, construct a confusion matrix to record the number of true classifications and predicted classifications of each category, the confusion matrix includes true positive examples TP, false positive examples FP, true negative examples TN and false negative examples FN, specifically including: creating a confusion matrix whose size is equal to the number of categories multiplied by the number of categories, each cell represents a combination of the true label and the predicted label of a category (such as TP, FP, TN, FN); according to the comparison result of the true label and the predicted label in step 161, update the corresponding cell in the confusion matrix. For example, if the true label of a sample is category A and the predicted label is also category A, then 1 is added to the corresponding position (A, A) of the confusion matrix, indicating that the number of true positive examples (TP) has increased.

[0103] Step 163, based on the confusion matrix, calculate the accuracy, that is, the proportion of all correctly predicted samples to the total samples, specifically including: traversing the diagonal elements of the confusion matrix, which represent the number of correctly predicted samples of each category (i.e., true positive samples TP), adding up all the values ​​on the diagonal to obtain the total number of correctly predicted samples; dividing the number of correctly predicted samples by the total number of samples in the test set to obtain the accuracy, which represents the proportion of the model's correct predictions in all test samples.

[0104] Step 164, initialize a list to save the accuracy of each iteration; for each of the K subsets, record it as a validation set, perform the following steps:

[0105] Merge the remaining K-1 subsets to form a temporary "test set"; use the trained convolutional neural network model to predict each image in the temporary "test set" and record the predicted label; compare the predicted label of each image in the temporary "test set" with the true label and update the confusion matrix; calculate the accuracy of the current iteration based on the confusion matrix, add it to the accuracy record, and reset the confusion matrix to prepare for the next iteration; calculate the average accuracy of K iterations as the final accuracy of the model, including:

[0106] Create an empty list to store the accuracy of each iteration; determine the value of K, that is, the number of data subsets, randomly divide the entire data set into K subsets of equal size, use a loop structure, such as a for loop, to traverse the K subsets, and in each iteration, select a subset as the validation set, and merge the remaining K-1 subsets into a temporary "test set". Create an empty data structure (such as a list or array) to store the images and labels of the temporary "test set", traverse all subsets except the current validation set, and add their images and labels to the temporary "test set". Use the trained convolutional neural network model to predict each image in the temporary "test set" and record the predicted label of each image. Compare the predicted label of each image in the temporary "test set" with its corresponding true label, update the confusion matrix based on the comparison results, and for each sample, add 1 to the corresponding position of the confusion matrix according to its true label and predicted label; calculate the accuracy based on the confusion matrix, the accuracy is the sum of the elements on the diagonal of the confusion matrix (true positive examples TP) divided by the total number of samples in the test set, and add the calculated accuracy to the previously initialized accuracy record list.

[0107] After each iteration, the confusion matrix is ​​reset to a zero matrix to prepare for the next iteration. This can be achieved by reinitializing the confusion matrix or using a zeroing operation. After all K iterations are completed, the average of all accuracy rates in the accuracy record list is calculated. This average represents the final performance evaluation result of the model under K-fold cross validation.

[0108] In an embodiment of the present invention, by predicting each image in the test set and comparing it with the true label, the performance of the model on unseen data can be comprehensively evaluated. The confusion matrix provides an intuitive way to visualize the classification performance of the model, which can clearly show the misclassification between categories; the accuracy is one of the important indicators for evaluating the performance of the classification model, which can intuitively reflect the correct proportion of the model prediction, and the accuracy can be used as a benchmark for performance comparison between different models or different iterations. By initializing a list to save the accuracy of each iteration, the performance changes of the model on different validation sets can be tracked, and the accuracy records of multiple iterations are helpful for analyzing the stability and generalization ability of the model; K-fold cross validation can ensure that each sample in the data set is used as a test set at least once, thereby making full use of limited data resources. Through multiple iterations and different data partitions, the performance robustness of the model under different data distributions can be evaluated. A single data partition may introduce accidental errors. K-fold cross validation can reduce the impact of this accidental error on model evaluation. By calculating the average accuracy of K iterations, a more reliable and stable model performance estimate can be obtained.

[0109] In a preferred embodiment of the present invention, in the above step 2, meteorological parameters are obtained in real time, and the meteorological parameters include temperature, humidity, wind speed, precipitation, air pressure and radiation; based on the meteorological parameters, the influence of the meteorological parameters on the line operation state is analyzed to obtain a meteorological analysis result, and the meteorological analysis result includes the influence of temperature, humidity and wind speed on the line operation state, which may include:

[0110] Identify a reliable source of meteorological data, which can be a local weather station, an online weather service API, or satellite data. According to the selected data source, set up the corresponding data acquisition system. If you are getting data from a weather station, you may need to install a data receiving device; if you are getting it from an online API, you need to write code to request data regularly. Ensure that the system can receive meteorological data in real time, which usually involves obtaining the latest meteorological parameters from the data source regularly (such as every minute, every hour); perform necessary preprocessing on the received raw data, such as format conversion, unit unification, outlier processing, etc., to ensure the accuracy and availability of the data.

[0111] Understand the type of line (such as overhead lines, cables, etc.) and its materials, design specifications, operating environment and other characteristics. On this basis, identify the meteorological parameters that may have a significant impact on the operating status of the line, focusing on temperature, humidity and wind speed; determine the purpose of the analysis, that is, to explore how these meteorological parameters affect the operating status of the line alone or together; ensure that the collected meteorological data is consistent with the line operating status data in time, so that the relationship between the two can be accurately analyzed, and compare the meteorological data and line status data at different time points. For example, whether the operating status of the line has changed significantly under high temperature, high humidity or strong wind weather. Analyze how temperature affects the conductivity of the line. For example, high temperature may increase the line resistance, thereby reducing the transmission efficiency. Investigate the effect of temperature on the thermal expansion of the line material, which may change the tension and sag of the line, thereby affecting the safety distance. Study how humidity affects the insulation material of the line. High humidity environment may cause a water film to form on the surface of the insulation layer, reducing the insulation resistance and increasing the risk of leakage. Analyze whether humidity accelerates the corrosion process of the line material, which may affect the structural strength and conductivity of the line in the long term. Explore the impact of wind speed on the mechanical stability of the line. Strong winds may cause the line to sway or even dance, increasing the risk of line contact or ground short circuit. Analyze the impact of wind speed changes on the wind load on the line, which is directly related to the line tension design and the stability of the supporting structure.

[0112] The findings, observations and conclusions from the above analysis process were recorded in detail, including the specific changes in the line operation status when various meteorological parameters changed. Visual tools such as charts and curves were used to display the relationship between meteorological parameters and line status, so as to facilitate a more intuitive understanding of the analysis results.

[0113] Based on the analysis results, targeted line maintenance and operation suggestions are put forward under different meteorological conditions. For example, in hot weather, it may be necessary to increase the inspection frequency of the line to ensure the normal operation of the heat dissipation equipment; in high humidity environment, it is necessary to pay attention to the condition of the insulation material and replace damaged parts in time; in windy weather, it may be necessary to adjust the line tension setting to enhance its wind resistance.

[0114] In a preferred embodiment of the present invention, the above step 3, real-time monitoring and identifying the vibration mode and characteristics of the three-span line conductors to evaluate the impact of the vibration on the line operation state, may include:

[0115] Step 31, define the initial temperature T of the simulated annealing algorithm o , termination temperature T f , cooling coefficient α and the number of iterations L at each temperature; set the initial solution S o , i.e., the initial position set of the vibration sensor;

[0116] Step 32, define an objective function for evaluating the quality of the position set S, wherein the calculation formula of the objective function f(S) is:

[0117]

[0118] Among them, λ1, λ2, λ3, λ4, λ5 and λ6 are weight coefficients; |X(f i )| 2 Indicates that each frequency f i The corresponding signal strength; N is the total number of frequency points, indicating the number of frequency components considered in the calculation; SNR (f i ) is each frequency point (f i ) signal-to-noise ratio; each sensor is circular with a radius of r i , then the coverage area of ​​each sensor is The total coverage area of ​​all sensors is obtained by summing up; K represents the total number of sensors, which represents the number of sensors in the layout; O(S i ,S j ) is the sensor S i and S j The area of ​​the overlap between them; i≠j represents all different sensor pairs; N t Indicates the number of iterations in the optimization process, indicating the number of steps the algorithm needs to execute; T maxIt represents the maximum allowable computing time. It ensures the best signal quality of the system by optimizing signal strength, signal-to-noise ratio and frequency response. It reduces sensor overlap, optimizes sensor layout, improves coverage efficiency, and reduces redundant signals and resource waste. It improves computing efficiency and reduces the computing burden of the system by controlling the number of calculation iterations and the maximum calculation time. These optimization effects help improve system performance, reduce resource waste, improve the efficiency of signal reception and processing, and ensure that the best network configuration or sensor layout solution can be obtained when time and resources are limited.

[0119] Step 33, for each iteration at the current temperature T, generate a new solution S new , specifically: at the current temperature, the algorithm will perform L iterations, and in each iteration, the algorithm will current Generate a new solution S new , which typically involves minor adjustments or reallocations of sensor locations.

[0120] Step 34, calculate the objective function value f(S new ), if f(S new ) is greater than the objective function value f(S current ), then accept the new solution and update S current =S new , where S current represents the current solution; if f(S new )≤f(S current ), then accept the new solution, and after completing the inner cycle, lower the temperature. If the current temperature T drops to the termination temperature T f Next, the simulated annealing process is stopped and the final solution, i.e., the final position set of the vibration sensor, is output. Specifically, the new solution S is calculated. new The objective function value f(S new ); if f(S new ) is greater than the objective function value f(S current ), then the new solution is accepted unconditionally and the current solution is updated to f(S new )≤f(S current ); if f(S new ) is less than or equal to f(S current ), then determine whether to accept the new solution, which allows the algorithm to avoid falling into the local optimum during the search process. After completing L iterations, the temperature T decreases according to the cooling coefficient α. When the temperature drops to the termination temperature T f When , the algorithm stops and outputs the final set of sensor positions.

[0121] Step 35, installing multiple vibration sensors at key positions of the three-span line to monitor the vibration data of the conductor in real time, and preprocessing the vibration data to obtain preprocessed vibration data, specifically including: installing vibration sensors at key positions of the three-span line according to the final sensor position set obtained by the simulated annealing algorithm, these sensors will monitor the vibration data of the conductor in real time, and transmit the data to the analysis system, and preprocessing the collected vibration data, such as filtering, denoising, etc., to obtain a clearer vibration signal.

[0122] Step 36, using Fourier transform to extract vibration features of the pre-processed vibration data, the vibration features include frequency, amplitude and phase, specifically including:

[0123] The vibration data in the time domain is converted to frequency domain data through the fast Fourier transform (FFT); FFT is used to calculate the Fourier transform of discrete signals, which can decompose the signal into a combination of sine waves and cosine waves of different frequencies; the frequency domain data obtained by FFT can determine the various frequency components present in the vibration signal, which correspond to the periodic components in the signal and are the key to identifying the vibration mode. For each frequency component, the corresponding amplitude value is extracted. The amplitude represents the intensity or amplitude of the corresponding frequency component in the vibration signal and is an important indicator for evaluating the distribution of vibration energy. In addition to frequency and amplitude, phase information can be extracted from the FFT results. The phase describes the time relationship between the various frequency components, which helps to understand the waveform structure of the vibration signal and the synchronization between them. The extracted frequency, amplitude, and phase features are organized into a structured data set, which can include creating a feature vector in which each element represents a specific feature value (such as the amplitude of a certain frequency).

[0124] Step 37, based on the vibration characteristics, a vibration analysis model is established to identify the vibration mode of the conductor, and the influence of the vibration on the line operation state is analyzed to obtain a vibration analysis result, which includes the vibration mode, frequency and amplitude of the conductor.

[0125] In an embodiment of the present invention, the optimal position of the vibration sensor is determined by the simulated annealing algorithm, which can ensure that the sensor layout can capture the key information of the conductor vibration and improve the efficiency and accuracy of data collection. By installing vibration sensors at key positions, the vibration of the conductor can be monitored in real time, providing data support for timely discovery of potential line problems; by preprocessing and Fourier transforming the vibration data, the key characteristics of the vibration, such as frequency, amplitude and phase, can be accurately extracted, and these characteristics provide an important basis for subsequent analysis. The vibration analysis model established based on the vibration characteristics can effectively identify different vibration modes of the conductor, which helps to deeply understand the dynamic behavior of the line under different environmental conditions. Through the vibration analysis results, the specific impact of vibration on the operation status of the line can be evaluated, including the risks of fatigue damage, looseness or breakage that may be caused, so as to take necessary maintenance measures in time. According to the vibration analysis results, a more accurate preventive maintenance plan can be formulated to extend the service life of the line and reduce the occurrence of unexpected power outages. Through continuous monitoring and analysis of conductor vibration, potential safety hazards can be discovered and resolved in a timely manner, thereby improving the stability and reliability of the entire power system.

[0126] In a preferred embodiment of the present invention, the above step 37 establishes a vibration analysis model based on the vibration characteristics, identifies the vibration mode of the conductor, and analyzes the impact of the vibration on the line operation state to obtain a vibration analysis result. The vibration analysis result includes the vibration mode, frequency and amplitude of the conductor, and may include:

[0127] Step 371, annotating the vibration data according to the characteristics of the vibration data, determining and annotating the corresponding vibration mode label to form an annotated data set;

[0128] Step 372, divide the labeled data set into a training set, a validation set and a test set, use the training set to train the recurrent neural network model, and during the training process, continuously adjust the weights and biases of the recurrent neural network model through the back propagation algorithm and the optimizer to obtain the trained recurrent neural network model, specifically including: dividing the labeled data set into a training set, a validation set and a test set according to a certain ratio (such as 70:15:15), selecting a suitable recurrent neural network (RNN) structure, such as a long short-term memory network (LSTM); initializing the weight and bias parameters of the RNN model, using the training set data to train the RNN model, the input is the vibration feature, and the output is the corresponding vibration mode label. During the training process, the gradient of the loss function with respect to the model parameters is calculated through the back propagation algorithm, and the weight and bias of the RNN model are updated according to the gradient information using the optimizer (such as Adam), and the steps are repeated until the performance of the model on the validation set reaches a preset standard or the number of training rounds reaches an upper limit.

[0129] Step 373, using the trained recurrent neural network model to classify the new vibration data to obtain the corresponding predicted vibration pattern label, specifically including: obtaining new unlabeled vibration data, using the trained RNN model to classify and predict the new data, and outputting the predicted vibration pattern label.

[0130] Step 374, by comparing the predicted vibration pattern label with the predefined vibration pattern label, different vibration patterns are identified, including breeze vibration and dancing, specifically including:

[0131] Predefine a set of vibration pattern labels, which are usually created based on historical data. Predefined vibration pattern labels include "breeze vibration" and "dancing" patterns. Each label corresponds to a specific vibration pattern, which differs in physical properties, frequency range, amplitude, etc. Use the previously trained recurrent neural network (RNN) model to classify and predict new, unlabeled vibration data. The RNN model will output the predicted labels corresponding to each data segment. These labels represent the vibration pattern to which the model believes the data segment belongs. Compare the predicted vibration pattern labels with the predefined vibration pattern labels. The comparison process can be string matching. If the predicted label completely matches a predefined label, the vibration pattern is considered to be successfully identified. For example, if the predicted label is "breeze vibration" and there is an identical label in the predefined label set, then it can be determined that the vibration pattern corresponding to the data segment is breeze vibration. For predicted labels that do not match or have a similarity below a threshold, they are further analyzed or classified as "unknown vibration patterns."

[0132] Step 375, analyzing and identifying the correlation between different vibration modes and line operation status, specifically including: analyzing the change trend of line operation status data such as current, voltage, temperature, etc. in different time periods, especially whether these data show significant abnormal fluctuations under the identified specific vibration mode; for each identified vibration mode, extracting the characteristic parameters of the corresponding line operation status data, such as current peak value, voltage valley value, average temperature and their fluctuation range, etc., these characteristic parameters will help to reveal the intrinsic connection between the vibration mode and the line state; applying the Pearson correlation coefficient formula to calculate the correlation coefficient between each vibration mode and the corresponding line operation status data, the Pearson correlation coefficient can provide a quantitative indicator to measure the strength and directionality of the linear relationship between two variables.

[0133] Step 376, based on the specific value of the correlation coefficient, assessing the risk level of the line under different vibration modes, specifically includes:

[0134] According to the specific value of the correlation coefficient, Calculate the risk index R, where C represents the correlation coefficient; D represents the duration; α, β, y, δ, and η are weight coefficients; A represents the amplitude; F represents the frequency; I represents the ideal frequency; M represents another reference frequency; S1 and S2 represent the standard deviation of the Gaussian function;

[0135] The risk index R is a comprehensive indicator. The higher the value of R, the greater the risk of the line in the current vibration mode. Set the risk threshold. The risk threshold is the critical value used to divide different risk levels. The risk threshold is set based on historical data. For example, three risk levels can be set: low risk, medium risk and high risk. Each level corresponds to a range of R values. Divide the risk level:

[0136] Low risk: When the R value is lower than a certain set low risk threshold, the line is considered to have a low risk in the current vibration mode, which means that the operating status of the line is relatively stable and the possibility of problems is small.

[0137] Medium risk: When the R value is between the low risk and high risk thresholds, the line is identified as medium risk. This means that although the line is still within the safe operation range, there are certain potential problems and it needs to be monitored and maintained more strictly.

[0138] High risk: if the R value exceeds the set high risk threshold, it means that the line is in a high risk state under the current vibration mode, which may mean that there are serious safety hazards on the line and immediate intervention measures are needed to prevent possible accidents.

[0139] For lines with different risk levels, corresponding response measures need to be formulated. For example, low-risk lines can be regularly inspected; medium-risk lines may need to increase the inspection frequency or perform certain preventive maintenance; high-risk lines may need to be shut down immediately for maintenance or take other emergency measures.

[0140] In an embodiment of the present invention, by marking the vibration data and training the recurrent neural network model, the vibration mode of the conductor can be more accurately identified. This method based on big data and machine learning has higher accuracy and efficiency than the traditional manual identification method; using the trained recurrent neural network model, the new vibration data can be automatically classified, reducing the need for manual intervention and improving the automation and intelligence level of processing. By real-time monitoring and identifying the vibration mode of the conductor, abnormal vibration conditions, such as breeze vibration and dancing, can be discovered in time, so as to timely carry out early warning and prevention to avoid the occurrence of line failures. Analyzing the correlation between different vibration modes and the operating status of the line can provide important decision support for the operation and maintenance of the power system. Based on this correlation, the risk level of the line under different vibration modes is evaluated, which helps to formulate a reasonable maintenance and repair plan. Through accurate identification and risk assessment of the vibration mode, potential safety hazards can be discovered and handled in time, thereby improving the reliability and stability of the entire power system. Through an intelligent monitoring and early warning system, unnecessary inspections and maintenance work can be reduced, thereby reducing the operation and maintenance costs of the power system.

[0141] In a preferred embodiment of the present invention, the above step 371, according to the characteristics of the vibration data, marks the vibration data, determines and marks the corresponding vibration mode label to form a marked data set, including:

[0142] Step 3711, determine the category of vibration modes, including breeze vibration and dancing, specifically including: conduct sufficient research and analysis on the vibration modes that may appear in the target field, including reading relevant literature, communicating with experts in the field, etc.; based on the research, determine two main vibration modes: breeze vibration and dancing, these two modes should be clearly distinguished in terms of physical characteristics, causes, etc.; record the determined vibration mode category and its definition in the document.

[0143] Step 3712, define a unique label for each vibration mode, convert the characteristics of the vibration data, i.e., frequency or amplitude, into genes in the genetic algorithm, and the genes are used to form the chromosomes of the individual, specifically including: designing a unique label for each vibration mode (breeze vibration and dancing). For example, numbers or specific strings can be used to represent different modes; select characteristics that can represent different vibration modes from the vibration data, such as frequency and amplitude; convert these characteristic values ​​into genes in the genetic algorithm, which usually involves mapping the data values ​​into a certain coding range, such as binary coding; the chromosomes of the individual are composed of multiple genes, and each chromosome represents a possible vibration mode label sequence.

[0144] Step 3713, randomly generate an initial population, in which each individual represents a vibration pattern label sequence, specifically including: determining the size of the initial population according to the complexity of the problem and the computing resources, using a random number generator to generate the initial population according to the structure of the chromosome, each individual is a randomly generated vibration pattern label sequence, and initializing the fitness and other record information for each individual in the population.

[0145] Step 3714, extracting key features for each vibration data sample, including time domain and frequency domain features, specifically including: performing necessary preprocessing on the original vibration data, such as denoising and normalization, calculating key features of the vibration data in the time domain, such as mean, variance, peak value, etc., converting the vibration data to the frequency domain through the Fourier transform method, and extracting key features, such as main frequency, spectrum energy distribution, etc., and integrating the extracted time domain and frequency domain features into a feature vector.

[0146] Step 3715, define a fitness function to evaluate the matching degree between each individual and the vibration data feature, wherein the calculation formula of the fitness function is:

[0147]

[0148] Among them, μ f (I j ) is the label I j The corresponding expected frequency value; σ f (I j ) is the standard deviation of the frequency feature; F(I) represents the fitness value of individual I, ranging from 0 to 1, and the higher the value, the better the matching degree; N is the total number of vibration data samples; I j represents the label corresponding to the jth data sample in individual I; and Respectively represent the frequency characteristics, amplitude characteristics and time domain characteristics of the jth data sample; μ a (I j ) is the label I j The corresponding expected amplitude value; σ a (I j ) is the standard deviation of the amplitude characteristic; μ t (I j ) is the label I j The corresponding expected time domain eigenvalue; σ t (I j ) is the standard deviation of the time domain feature; w f 、w a and w t are the weight coefficients of frequency, amplitude and time domain feature matching, satisfying w f +w a +w t =11.

[0149] Step 3716, according to the fitness function, select the corresponding individuals to enter the next generation, randomly select two individuals as parents, perform a crossover operation to generate new offspring, randomly mutate the genes of the offspring, that is, the labels of the vibration patterns, and repeat the selection, crossover and mutation operations until the termination condition is met. When the termination condition is met, stop the iteration and output the final individual as the labeling result of the vibration data, specifically including: according to the result of the fitness function, select individuals with high fitness to enter the next generation, which can be achieved through roulette selection, tournament selection and other methods, randomly select two individuals as parents, perform a crossover operation to generate new offspring, and the crossover point can be randomly selected; perform random gene mutation on the newly generated offspring to increase the diversity of the population, set the conditions for the termination of the iteration, such as reaching the maximum number of iterations, the fitness reaching a preset threshold, etc., stop the iteration when the termination condition is met, and after the iteration, output the individual with the highest fitness as the labeling result of the vibration data.

[0150] Step 3717, decoding the final individual into the corresponding vibration pattern label sequence, matching the vibration pattern label sequence with the original vibration data sample to form a labeled data set, specifically including: decoding the final individual into the corresponding vibration pattern label sequence, which involves converting the encoding in the genetic algorithm back to the original vibration pattern label; matching the decoded vibration pattern label sequence with the original vibration data sample to ensure that each data sample has its corresponding label; arranging all labeled data samples to form a complete labeled data set.

[0151] In an embodiment of the present invention, by optimizing the annotation process of the vibration mode through a genetic algorithm, different vibration modes (such as breeze vibration and dancing) can be more accurately identified and annotated. This method can more effectively utilize the characteristics of vibration data, such as frequency and amplitude, to improve the accuracy of annotation. The method realizes a certain degree of automation, reduces the need for manual intervention and manual annotation, and through the iterative optimization process, the algorithm can automatically find the label that best matches the vibration data characteristics, thereby improving the efficiency of annotation. Since the genetic algorithm is to find the optimal solution by simulating natural selection and genetics mechanisms, the method has strong robustness when dealing with complex and nonlinear problems. At the same time, through continuous iteration and optimization, the algorithm can learn the inherent laws and characteristics of the data, so that it has good generalization ability and can handle different types of vibration data. The method is not only suitable for the annotation of the two vibration modes of breeze vibration and dancing, but can also be extended to more types of vibration modes as needed. By defining new labels and fitness functions, new vibration modes can be easily incorporated into the annotation system. Through accurate vibration mode annotation, a high-quality annotation data set can be formed. Such a data set is crucial for subsequent tasks such as machine learning and pattern recognition, and helps to improve the performance and accuracy of related applications.

[0152] The fitness function provides a comprehensive evaluation method by combining the matching of frequency, amplitude and time domain features, ensuring the multi-dimensional matching between individuals and vibration data features, thereby improving the accuracy of annotation. f 、w a and w t , the importance of certain features can be flexibly emphasized or downplayed according to actual needs. This flexibility enables the fitness function to adapt to different application scenarios and data characteristics. Using a Gaussian function to calculate the similarity between features and labels makes the function tolerant to small changes in features, which enhances the robustness of the algorithm. This means that even if there is a certain degree of noise or fluctuation in the data, the algorithm can still effectively identify the vibration pattern label that best matches it.

[0153] In a preferred embodiment of the present invention, in the above step 4, the equipment damage data, foreign body intrusion data and temperature anomaly data are sorted to form an abnormal condition data set, which is used as the boundary condition of the finite element analysis, and may include:

[0154] The data is classified according to the type of abnormality (such as equipment damage, foreign body intrusion, and temperature anomaly), and corresponding labels are assigned to each type of data. The cleaned and labeled data are integrated into an abnormal condition data set, which will contain descriptions of various abnormal conditions, time of occurrence, location, severity, and other information; based on the abnormal condition data set, boundary conditions that affect finite element analysis are extracted, such as structural changes caused by equipment damage, external forces caused by foreign body intrusion, and changes in material properties caused by temperature anomalies.

[0155] In a preferred embodiment of the present invention, in the above step 4, the meteorological analysis results obtained based on the meteorological parameter analysis are summarized to form a meteorological analysis data set, and used as the environmental load for finite element analysis, which may include:

[0156] The meteorological parameters are analyzed to extract the features closely related to the conductor vibration, such as the average wind speed, maximum wind speed, and wind direction change frequency. The meteorological features obtained by the analysis are integrated into a meteorological analysis data set, which will contain the parameter values ​​under various meteorological conditions and their corresponding timestamps. Based on the meteorological analysis data set, the environmental loads required in the finite element analysis, such as wind loads and temperature loads, are determined. These loads will be used to simulate the dynamic behavior of the conductor under different meteorological conditions.

[0157] In a preferred embodiment of the present invention, in the above step 4, the vibration mode, frequency and amplitude of the conductor are sorted to form a vibration analysis data set for simulating the dynamic behavior of the conductor in the finite element method, which may include:

[0158] Key features are extracted from the preprocessed vibration data, such as the dominant vibration mode, main frequency components, maximum amplitude, etc. The extracted vibration features are integrated into a vibration analysis data set, which will be used to describe the vibration state of the conductor at different times and under different conditions. According to the vibration analysis data set, the corresponding dynamic behavior parameters are set in the finite element model to simulate the actual vibration of the conductor, which helps to more accurately evaluate the performance and safety of the conductor under various conditions.

[0159] In a preferred embodiment of the present invention, in the above step 4, a finite element model of the three-span line is established, and finite element analysis is performed to simulate the response of the three-span line under different meteorological conditions and vibration loads to identify the key areas and factors that cause line failures, including:

[0160] Determine the specific structure of the three-span line, including the geometry, materials and connection methods of the conductors, insulators and towers, including: obtaining detailed parameters of the conductors, insulators and towers from design drawings, technical specifications or on-site surveys; determining the three-dimensional geometry of the conductors, insulators and towers, including length, diameter, cross-sectional shape, etc.; determining the materials used for each component, such as the conductor may be aluminum or copper alloy, the insulator may be ceramic or glass, and the tower may be a steel structure; understanding the connection methods between the conductor and the insulator, and between the insulator and the tower, such as hanging ring connection, bolt connection, etc.

[0161] According to the specific structure of the three-span line, a three-span line geometric model is created, and corresponding material properties are specified for each component in the three-span line geometric model, including elastic modulus, density and Poisson's ratio; the connection relationship between each component is defined, including hinged or fixed connection, specifically including: based on the collected data, creating a geometric model of the conductor, insulator and tower in ANSYS software; assigning corresponding elastic modulus, density, Poisson's ratio and other material properties to each component, and setting the connection relationship between the components in the model to ensure that the model is consistent with the actual structure.

[0162] The geometric model of the three-span line is meshed, and the boundary conditions of the geometric model of the three-span line are set, including fixed constraints and load application positions. Specifically, the geometric model is meshed, a finite element mesh is generated for calculation, and the fixed constraints of the model are determined, such as the fixation of the bottom of the tower; and the load application position is set, such as the wind load application point on the conductor.

[0163] According to the meteorological analysis data set, different meteorological conditions are defined, including wind speed, wind direction and temperature, and applied as environmental loads to the three-span line geometry model. Specifically, the meteorological analysis data set collected previously is imported into the finite element analysis software; different wind speed, wind direction and temperature conditions are defined according to the data set, and these meteorological conditions are applied as environmental loads to the corresponding positions of the model, such as wind loads are applied to conductors and towers, and temperature loads affect the properties of the materials.

[0164] The vibration analysis data set is used to determine the vibration mode, frequency and amplitude of the conductor, and the vibration load is applied to the conductor part, specifically including: importing the vibration analysis data set into the software, extracting the vibration mode, frequency and amplitude of the conductor from the data set, applying these vibration loads to the conductor part of the model, and simulating the actual vibration of the conductor.

[0165] Set the solution parameters, including the time step and number of iterations, start the solver, and start running the simulation analysis to obtain the simulation results. The simulation results include stress distribution cloud diagrams, deformation diagrams, and vibration response curves, including:

[0166] Determine the needs and objectives of the analysis, which includes determining the physical phenomena to be simulated (such as stress distribution, deformation, vibration response, etc.), as well as the information expected to be obtained from the simulation results (such as maximum stress value, deformation at a specific location, vibration frequency and amplitude, etc.).

[0167] Set the time step. The time step is an important parameter in finite element analysis. It determines the precision of time advancement during the simulation process. Set the number of iterations. The number of iterations refers to the number of times the solver performs numerical calculations in each time step. After completing the setting of the time step and the number of iterations, the next step is to start the finite element analysis solver, which involves the following steps:

[0168] Select the corresponding solver type according to the nature of the problem (linear, nonlinear, static, dynamic, etc.); click "Start" or similar buttons to officially start the solver for simulation calculation. During the calculation process, monitor the progress and status of the calculation through the ANSYS software interface or log file, which includes viewing the current number of iterations, residual norm, calculation time and other information. If the calculation is found to be abnormal (such as non-convergence, slow calculation speed, etc.), you can adjust the parameters or interrupt the calculation in time. After the calculation is completed, the ANSYS software will generate a series of simulation results for users to view and analyze. These results include:

[0169] The stress distribution cloud diagram graphically displays the stress distribution of each part of the model to help identify areas of stress concentration.

[0170] Deformation diagram, showing the deformation of the model under the action of force, including displacement, strain, etc.

[0171] Vibration response curve, which displays the response of the model under vibration load in the form of a graph, such as amplitude-frequency curve, time-displacement curve, etc.

[0172] According to the simulation results, identify the key areas with stress concentration, large deformation or abnormal vibration response, including: carefully observing the stress distribution cloud map, deformation map and vibration response curve, identifying the areas with stress concentration, large deformation or abnormal vibration response, and marking these key areas on the model or result map to facilitate subsequent in-depth analysis.

[0173] Analyze the response characteristics of key areas under different meteorological conditions and vibration loads to determine the factors that lead to line failures. Specifically, it includes: Accurately extract detailed data of key areas from simulation results. These data usually include:

[0174] Specific stress values, locate key areas in the stress distribution cloud map, and record the maximum, minimum and average stress values ​​of these areas. These data help to understand the degree and scope of stress concentration.

[0175] Deformation, extract the deformation data of key areas from the deformation diagram, including displacement and strain values. These data can reflect the stability and deformation of the structure under force.

[0176] Vibration response parameters, extracting parameters such as vibration frequency and amplitude in key areas from the vibration response curve. These parameters are helpful in analyzing the dynamic characteristics and vibration stability of the structure.

[0177] After extracting the data, compare the response characteristics of key areas under different meteorological conditions and vibration loads, including:

[0178] Compare meteorological conditions, compare the stress values, deformations and vibration responses of key areas under different wind speeds, wind directions and temperatures, observe how these parameters change with changes in meteorological conditions, and whether there are obvious trends or patterns.

[0179] Comparison of vibration loads, analysis of the response of key areas at different vibration frequencies and amplitudes, paying special attention to the occurrence of resonance and its impact on structural stability.

[0180] During the comparison process, graphs, curves, or data tables can be used to clearly show the differences in responses under different conditions.

[0181] After comparing the response characteristics under different conditions, the specific impact of these responses on line faults is analyzed. This includes:

[0182] The impact of stress concentration. High stress areas may cause material fatigue, cracking or fracture, thereby causing line failure. Analyze the potential connection between stress concentration and failure.

[0183] Impact of deformation. Excessive deformation may lead to poor contact, misalignment or breakage between circuit components. Evaluate the impact of deformation on circuit performance and reliability.

[0184] Impact of vibration. Continuous vibration may cause the line to loosen, wear or break. Analyze the correlation between vibration response and line failure.

[0185] Finally, based on the results of the comparative analysis, determine the main factors that lead to line failures, which may include:

[0186] Certain meteorological conditions: such as strong winds, extreme temperatures, etc., may cause line stress concentration or excessive deformation.

[0187] Specific vibration loads: Certain vibration frequencies or amplitudes may cause resonance in the line, thereby accelerating line fatigue and damage.

[0188] Defects in structural design, such as insufficient rigidity in certain areas and unreasonable design of connection points, may also be the main factors leading to failure.

[0189] In an embodiment of the present invention, through simulation analysis, key areas where failures may occur in the line can be identified in advance, so as to strengthen the monitoring and maintenance of these areas in actual operation and effectively prevent the occurrence of line failures. Through finite element simulation, engineers can take into account the line response under various complex environments in the design stage, and then optimize the structural design of the line and improve the weather resistance and stability of the line. By accurately simulating the impact of different meteorological conditions and vibration loads on the line, a more accurate maintenance plan can be formulated, unnecessary maintenance work can be reduced, and maintenance costs can be reduced. Timely identification and handling of potential problems in key areas can significantly improve the overall safety of the line and reduce safety accidents caused by line failures. After having a deep understanding of the response of the line in different environments, once an emergency occurs, the problem can be located more quickly and effective countermeasures can be taken. Finite element simulation analysis is an important part of the intelligent management of modern power systems. Through accurate data analysis and simulation, it can provide strong support for the construction of smart grids.

[0190] In a preferred embodiment of the present invention, meshing of the three-span line geometric model includes:

[0191] Set the number of particles, speed range and position range for the particle swarm optimization algorithm; set the learning rate and number of iterations for the gradient descent algorithm, including: selecting an appropriate number of particles. Too many particles will increase the computational burden, while too few particles may affect the globality of the search; set a reasonable range for the speed of each particle to ensure that the movement of the particle in the search space is neither too fast nor too slow. Moving too fast may cause the particle to miss the optimal solution, while moving too slowly will affect the search efficiency; set a limit for the position of the particle based on the possible value range of the grid partition parameter, which ensures that the particle does not exceed the reasonable parameter space during the search process; the learning rate determines the step size of the parameter update during the gradient descent process. The appropriate learning rate can ensure the convergence speed while avoiding oscillation; set an appropriate number of iterations based on the complexity of the problem and the required optimization accuracy. Too many iterations may waste computing resources, while too few iterations may lead to incomplete optimization.

[0192] Generate a set of initial particles, each particle represents a grid division scheme, and the position of the particle represents the parameters of the grid division, including the grid size and distribution. Specifically, the initial position of each particle is randomly generated within the set position range. These positions represent different combinations of grid division parameters, and an initial speed is randomly assigned to each particle within the set speed range.

[0193] The grid division scheme represented by each particle is evaluated by the evaluation function, the speed and position of each particle are updated, and the fitness evaluation and particle update are repeated until the convergence condition is met to obtain the global final particle. Specifically, for the grid division scheme represented by each particle, the fitness value is calculated using the evaluation function, and the speed and position of the particle are updated according to the current position and speed of each particle, as well as its own historical best position and the global best position. The process of evaluating the fitness evaluation and particle update is repeated until the preset convergence condition is met (such as reaching the maximum number of iterations). The calculation formula of the evaluation function F(x) is:

[0194]

[0195] in, represents the summation symbol, which means that the sum is taken over all n elements (or units), and the contribution of each element is calculated and added up; represents the weight coefficient of the stress part; Represents stress calculation, which is the way stress is calculated, where F i is the external force applied to the i-th element, A i (x) is the cross-sectional area of ​​the ith element, which depends on the optimization parameter x. The stress is defined as the external force divided by the cross-sectional area; is the nonlinear correction term of the stress part, where α is the amplification factor and β is the nonlinear exponent controlling the stress-strain relationship; w i (x) is the weight coefficient related to the deformation of the i-th element; is the calculation method of deformation, where K i (x) is the stiffness coefficient of the i-th element; is the nonlinear correction term of the deformation part, where γ1, γ2 and γ are nonlinear exponents; x represents the optimization parameter. Each particle in the particle swarm optimization algorithm updates its velocity based on its current position and velocity, combined with its personal historical optimal position and the global optimal position. The velocity update formula is as follows:

[0196]

[0197] in, is the velocity of the ith particle after the k+1th iteration; w is the inertia weight, which is used to control the velocity of the particle; is the speed of the i-th particle at the k-th iteration; c1 and c2 are learning factors used to balance the influence of personal optimality and global optimality; r1 and r2 are random numbers, usually in the range of [0, 1]; is the personal optimal position of the i-th particle at the k-th iteration; g k is the position of the global optimal particle; is the position of the i-th particle at the k-th iteration; the particle position update is performed according to the velocity update, and the position update formula is as follows:

[0198]

[0199] in, is the new position of the ith particle after the k+1th iteration; is the position of the ith particle at the kth iteration; is the velocity of the ith particle after the k+1th iteration.

[0200] The meshing parameters are extracted from the global final particle as the initial scheme of gradient descent, which specifically includes: after the particle swarm optimization process is completed, the particle with the highest fitness value, that is, the global optimal particle, is found, and the meshing parameters, including the size and distribution of the mesh, are extracted from the global optimal particle.

[0201] The initial meshing scheme is disturbed, and the gradient of the optimization function with respect to the meshing parameters is calculated by the finite difference method. The meshing parameters are adjusted along the direction of gradient descent. The gradient calculation and meshing parameter adjustment are repeated until the preset number of iterations is reached to obtain the optimized meshing parameters, which specifically include:

[0202] Based on the meshing parameters provided by the global optimal particle, a small perturbation is performed to generate a series of similar meshing schemes;

[0203] Determine the set of grid partitioning parameters that need to be optimized, denoted as θ, where θ can be a multidimensional vector containing parameters such as grid size and distribution. Set a small perturbation ∈ to calculate the parameter changes when calculating the gradient. The calculation formula of the optimization function F(θ) is:

[0204]

[0205] Among them, V j is the volume of the jth grid cell; l max,j is the length of the longest side in the jth grid cell; A j is the surface area of ​​the jth grid cell; N(θ) is the number of grid cells; α and β are weight coefficients; c j is the center coordinate of the jth grid cell; μ is the center coordinate of the region of interest; σ is the standard deviation of the Gaussian function; ‖·‖ represents the Euclidean distance; prepare the vector to record the gradient Its dimension is the same as θ and is initialized to zero; for each parameter θ in the parameter vector θ i :

[0206] For the current parameter θ iPerform a forward perturbation to obtain a new parameter vector in Other parameters remain unchanged; calculate the optimization function value f(θ + );For the current parameter θ i Perform negative perturbation to obtain a new parameter vector θ - ,in Calculate the optimization function value f(θ) of the perturbation solution - ), use the central difference formula to calculate the parameter θ i The partial derivative at (i.e., a component of the gradient):

[0207]

[0208] According to the calculated gradient vector And the set learning rate η, update the grid partitioning parameters:

[0209]

[0210] Ensure that the updated θ new If the parameters are still within a reasonable range, truncate or adjust if necessary, repeat the steps until the preset number of iterations is reached, and output the final meshing parameter θ new .

[0211] The three-span line geometric model is meshed using the optimized meshing parameters, specifically including: using the meshing parameters optimized by gradient descent to perform actual meshing operations on the three-span line geometric model, and performing quality checks on the meshes after meshing to ensure that they meet the requirements of numerical simulation or analysis.

[0212] In an embodiment of the present invention, the particle swarm optimization algorithm can effectively explore the entire solution space through parallel search of multiple particles, avoid falling into the local optimal solution, and thus find a better grid division scheme in the global scope. The gradient descent algorithm makes further local fine adjustments based on the global optimal solution found by the particle swarm optimization algorithm. By calculating the gradient and updating the parameters along the gradient descent direction, the accuracy and quality of the grid division can be significantly improved. Combining the two algorithms, the strategy of global first and local later reduces unnecessary calculations. The particle swarm optimization algorithm quickly locates the potential optimal solution area, while the gradient descent algorithm performs efficient and fine search in the area, thereby improving the overall computing efficiency.

[0213] In a preferred embodiment of the present invention, in the above step 4, based on the key areas and factors causing the line fault, the safety status of the line is evaluated and the potential fault type is determined, which specifically includes:

[0214] Collect geological survey reports of key areas to understand soil types, groundwater levels, rock structure and other information; collect historical climate data, especially records of extreme weather events (such as heavy rain, strong winds, freezing, etc.); obtain traffic flow data in key areas through traffic monitoring systems, including peak hours, average vehicle speeds, etc.; analyze the impact of geological conditions on line stability, identify possible geological risks such as settlement and landslides, assess the potential impact of climate conditions on corrosion and aging of line facilities, and study the impact of traffic flow on track wear and signal system load. Based on the above analysis, determine the main factors leading to failures, such as geological instability, frequent extreme climate, and high traffic load.

[0215] Conduct field measurements of the flatness and wear of the track, check the operation log of the signal system, test the accuracy and response time of the signal lights, conduct visual inspections of electrical equipment, and measure key parameters such as voltage and current; record the results of each inspection, including numerical data, photos or video evidence, and mark the problems or potential hazards found; compare and analyze the field survey data with historical fault records, evaluate the overall safety status of each key area, identify areas with obvious safety hazards or performance degradation, and divide the key areas into different risk levels (such as low risk, medium risk, and high risk) based on the evaluation results; obtain the geological conditions, climate impacts, traffic flow and other characteristics of the key areas, and identify the key factors that may cause line failures in these characteristics; list the types of faults that have occurred in the area in history according to the characteristics of the key areas; summarize all possible types of faults, such as track deformation, signal failure, electrical short circuit, etc.; for each type of fault, study its possible causes; for example, track deformation may be caused by geological subsidence, frequent passage of heavy vehicles, etc.; determine the specific conditions that cause the fault to change from a potential state to an actual occurrence. For example, extreme climate events (water accumulation caused by heavy rain) may trigger an electrical short circuit. Depicting the complete development path from fault triggering to the final impact on line operation helps to understand how faults escalate step by step and the possibility of taking intervention measures at different stages.

[0216] Collect fault records of key areas over a period of time (such as 5 or 10 years), clean and classify the data, and ensure that the records of each fault type are accurate; count the number of times each fault type appears in historical data, and calculate the frequency of faults, such as the average number of occurrences per year; combine the fault frequency and the most recent safety status assessment results to assess the likelihood of each fault type occurring in the future.

[0217] Determine the dimensions of analysis, including line safety, operational efficiency, passenger experience, etc., and set specific evaluation indicators for each dimension, such as the delay time caused by the fault, the number of passenger complaints, etc.; based on historical data and simulation scenarios, estimate the impact of each fault type on each dimension; for example, signal failure may cause train delays and passenger dissatisfaction, and electrical short circuits may cause safety issues. Summarize the impact assessment results of each fault type on each dimension, use weighted scores, and rank the comprehensive impact of the fault to identify the fault type that needs the most attention.

[0218] like Figure 1 As shown, a three-span line status intelligent assessment system based on numerical weather forecasting and machine learning includes:

[0219] The recognition module is used to automatically identify and classify abnormal conditions of the line from the pre-processed images by building, training and tuning a convolutional neural network model. The abnormal conditions include equipment damage data, foreign object intrusion data and temperature abnormality data;

[0220] The meteorological analysis module is used to obtain meteorological parameters in real time, including temperature, humidity, wind speed, precipitation, air pressure and radiation; based on the meteorological parameters, the influence of the meteorological parameters on the line operation status is analyzed to obtain the meteorological analysis results, including the influence of temperature, humidity and wind speed on the line operation status;

[0221] An evaluation module is used to monitor and identify the vibration mode and characteristics of the three-span line conductors in real time to evaluate the impact of vibration on the line operation status;

[0222] The finite element analysis module is used to organize the equipment damage data, foreign body intrusion data and temperature anomaly data to form an abnormal condition data set, which is used as the boundary condition of the finite element analysis; the meteorological analysis results obtained based on the meteorological parameter analysis are summarized to form a meteorological analysis data set, which is used as the environmental load for the finite element analysis; the vibration mode, frequency and amplitude of the conductor are organized to form a vibration analysis data set, which is used to simulate the dynamic behavior of the conductor in the finite element; a finite element model of the three-span line is established, and the finite element analysis is run to simulate the response of the three-span line under different meteorological conditions and vibration loads to identify the key areas and factors that cause line failures; based on the key areas and factors that cause line failures, the safety status of the line is evaluated and the potential fault type is determined.

[0223] It should be noted that the system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0224] The embodiment of the present invention further provides a computing device, comprising: a processor, a memory storing a computer program, wherein when the computer program is executed by the processor, the method described above is executed. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0225] The embodiment of the present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the method described above. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0226] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A three-span line status intelligent assessment system based on numerical weather forecast and machine learning, characterized in that: include: The recognition module is used to automatically identify and classify abnormal conditions of the line from the pre-processed images by building, training and tuning a convolutional neural network model. The abnormal conditions include equipment damage data, foreign object intrusion data and temperature abnormality data; The meteorological analysis module is used to obtain meteorological parameters in real time, including temperature, humidity, wind speed, precipitation, air pressure and radiation; based on the meteorological parameters, the influence of the meteorological parameters on the line operation status is analyzed to obtain the meteorological analysis results, including the influence of temperature, humidity and wind speed on the line operation status; An evaluation module is used to monitor and identify the vibration mode and characteristics of the three-span line conductors in real time to evaluate the impact of vibration on the line operation status; The finite element analysis module is used to organize the equipment damage data, foreign body intrusion data and temperature anomaly data to form an abnormal condition data set, which is used as the boundary condition of the finite element analysis; the meteorological analysis results obtained based on the meteorological parameter analysis are summarized to form a meteorological analysis data set, which is used as the environmental load of the finite element analysis; The vibration mode, frequency and amplitude of the conductor are sorted to form a vibration analysis data set, which is used to simulate the dynamic behavior of the conductor in the finite element method; Establish a finite element model of the three-span line and run finite element analysis to simulate the response of the three-span line under different meteorological conditions and vibration loads to identify the key areas and factors that lead to line failures; Evaluate the safety status of the line and determine potential fault types based on the key areas and factors that cause line faults.

2. The three-span line status intelligent assessment system based on numerical weather forecast and machine learning according to claim 1 is characterized in that: By building, training, and tuning a convolutional neural network model, the system can automatically identify and classify abnormal line conditions from pre-processed images. The abnormal conditions include equipment damage data, foreign object intrusion data, and temperature abnormality data, including: The real-time image data of the three-span lines are collected by video monitoring equipment, and image preprocessing is performed to obtain preprocessed images; Screen out samples containing equipment damage, foreign matter intrusion, and temperature abnormalities from the preprocessed images, and annotate the samples to create a labeled dataset; Divide the labeled dataset into training, validation, and test sets; Build a convolutional neural network model and initialize the convolutional neural network model parameters, including weights and biases; Use the training set to train the convolutional neural network model, calculate the predicted value through forward propagation, and optimize the convolutional neural network model parameters through the back propagation algorithm to obtain the trained convolutional neural network model; Use the test set to evaluate the trained convolutional neural network model and calculate the accuracy; adjust the trained convolutional neural network model according to the accuracy to obtain the final convolutional neural network model; The preprocessed new image is input into the final convolutional neural network model to obtain the abnormal conditions of the line, which include equipment damage, foreign body intrusion and temperature abnormality.

3. The three-span line status intelligent assessment system based on numerical weather forecast and machine learning according to claim 2 is characterized in that: Use the test set to evaluate the trained convolutional neural network model and calculate the accuracy, including: Use the trained convolutional neural network model to predict each image in the test set and record the predicted label of each image; compare the predicted label of each image in the test set with the actual label; Construct a confusion matrix to record the number of true and predicted classifications of each category. The confusion matrix includes true positive examples TP, false positive examples FP, true negative examples TN, and false negative examples FN; According to the confusion matrix, calculate the accuracy, that is, the proportion of all correctly predicted samples to the total samples; Initialize a list to save the accuracy of each iteration; for each of the K subsets, record it as a validation set, and perform the following steps: Merge the remaining K-1 subsets to form a temporary "test set"; use the trained convolutional neural network model to predict each image in the temporary "test set" and record the predicted label; compare the predicted label of each image in the temporary "test set" with the true label, and update the confusion matrix; calculate the accuracy of the current iteration based on the confusion matrix, add it to the accuracy record, reset the confusion matrix to prepare for the next iteration; calculate the average accuracy of K iterations as the final accuracy of the model.

4. The three-span line status intelligent assessment system based on numerical weather forecast and machine learning according to claim 3 is characterized in that: Real-time monitoring and identification of the vibration modes and characteristics of three-span line conductors to assess the impact of vibration on line operation, including: Define the initial temperature T of the simulated annealing algorithm o , termination temperature T f , cooling coefficient α and the number of iterations L at each temperature; set the initial solution S o , i.e., the initial position set of the vibration sensor; Define an objective function to evaluate the quality of the position set S; For each iteration at the current temperature T, a new solution S is generated new ; Calculate the objective function value f(S new ), if f(S new ) is greater than the objective function value f(S current ), then accept the new solution and update S current =S new , where S current represents the current solution; if f(S new )≤f(S current ), then accept the new solution, and after completing the inner cycle, lower the temperature. If the current temperature T drops to the termination temperature T f Below, the simulated annealing process is stopped and the final solution, i.e., the final position set of the vibration sensor, is output; Install multiple vibration sensors at key locations of the three-span line to monitor the vibration data of the conductor in real time and pre-process the vibration data to obtain pre-processed vibration data; The vibration characteristics of the pre-processed vibration data are extracted by Fourier transform, and the vibration characteristics include frequency, amplitude and phase; Based on the vibration characteristics, a vibration analysis model is established to identify the vibration mode of the conductor and analyze the impact of vibration on the line operation status to obtain the vibration analysis results, which include the vibration mode, frequency and amplitude of the conductor.

5. The three-span line status intelligent assessment system based on numerical weather forecast and machine learning according to claim 4 is characterized in that: Based on the vibration characteristics, a vibration analysis model is established to identify the vibration mode of the conductor and analyze the impact of vibration on the line operation status to obtain the vibration analysis results. The vibration analysis results include the vibration mode, frequency and amplitude of the conductor, including: According to the characteristics of the vibration data, the vibration data is labeled, and the corresponding vibration mode labels are determined and labeled to form a labeled data set; The labeled data set is divided into a training set, a validation set, and a test set. The training set is used to train the recurrent neural network model. During the training process, the weights and biases of the recurrent neural network model are continuously adjusted through the back propagation algorithm and the optimizer to obtain the trained recurrent neural network model. Use the trained recurrent neural network model to classify new vibration data to obtain the corresponding predicted vibration pattern labels; By comparing the predicted vibration pattern labels with predefined vibration pattern labels, different vibration patterns, including breeze vibration and dancing, are identified; The analysis identifies the correlation between different vibration modes and the line operation status; Based on the correlation, the risk level of the line under different vibration modes is evaluated.

6. The three-span line status intelligent assessment system based on numerical weather forecast and machine learning according to claim 5 is characterized in that: According to the characteristics of the vibration data, the vibration data is labeled, and its corresponding vibration mode label is determined and labeled to form a labeled data set, including: Identify categories of vibration patterns, including breeze vibrations and dancing; Define a unique label for each vibration pattern, and convert the characteristics of the vibration data, namely frequency or amplitude, into genes in the genetic algorithm. Genes are used to constitute the chromosomes of individuals. An initial population is randomly generated, in which each individual represents a vibration pattern label sequence; Extract key features from each vibration data sample, including time domain and frequency domain features; Define a fitness function to evaluate how well each individual matches the vibration data features; According to the fitness function, the corresponding individuals are selected to enter the next generation, two individuals are randomly selected as parents, and a crossover operation is performed to produce new offspring. The genes of the offspring, i.e., the labels of the vibration patterns, are randomly mutated. The selection, crossover and mutation operations are repeated until the termination condition is met. When the termination condition is met, the iteration is stopped and the final individual is output as the labeling result of the vibration data. The final individuals are decoded into corresponding vibration pattern label sequences, and the vibration pattern label sequences are matched with the original vibration data samples to form a labeled dataset.

7. The three-span line status intelligent assessment system based on numerical weather forecast and machine learning according to claim 6 is characterized in that: A finite element model of the three-span line was built and finite element analysis was performed to simulate the response of the three-span line under different meteorological conditions and vibration loads to identify the key areas and factors that lead to line failures, including: Determine the detailed structure of the three-span line, including the geometry, materials and connection methods of conductors, insulators and towers; According to the specific structure of the three-span line, a three-span line geometric model is created, and corresponding material properties are specified for each component in the three-span line geometric model, including elastic modulus, density and Poisson's ratio; the connection relationship between each component is defined, including hinged or fixed connection; Meshing the geometric model of the three-span line and setting the boundary conditions of the geometric model of the three-span line, including fixed constraints and load application positions; Based on the meteorological analysis dataset, different meteorological conditions, including wind speed, wind direction and temperature, are defined and applied as environmental loads to the three-span line geometry model; Using vibration analysis data sets, determine the vibration mode, frequency and amplitude of the conductor and apply vibration loads to the conductor section; Set the solution parameters, including the time step and the number of iterations, start the solver, and start running the simulation analysis to obtain the simulation results, which include stress distribution cloud diagrams, deformation diagrams, and vibration response curves; Based on the simulation results, identify key areas with stress concentration, large deformation or abnormal vibration response; Analyze the response characteristics of key areas under different meteorological conditions and vibration loads to determine the factors that lead to line failures.

8. The three-span line status intelligent assessment system based on numerical weather forecast and machine learning according to claim 7 is characterized in that: Meshing of the three-span line geometry model, including: Set the number of particles, velocity range, and position range for the particle swarm optimization algorithm; set the learning rate and number of iterations for the gradient descent algorithm; Generate a set of initial particles, each particle represents a grid division scheme, and the position of the particle indicates the parameters of the grid division, including the grid size and distribution; The grid division scheme represented by each particle is evaluated through the evaluation function, the speed and position of each particle are updated, and the fitness is evaluated and the particles are updated repeatedly until the convergence condition is met to obtain the global final particle; Extract the meshing parameters from the global final particle as the initial scheme of gradient descent; The initial grid solution is disturbed, and the gradient of the optimization function with respect to the grid partitioning parameters is calculated by the finite difference method. The grid partitioning parameters are adjusted along the direction of gradient descent, and the gradient calculation and grid parameter adjustment are repeated until the preset number of iterations is reached to obtain the optimized grid partitioning parameters. The three-span line geometric model is meshed using the optimized meshing parameters.

9. A three-span line status intelligent assessment method based on numerical weather forecast and machine learning, characterized in that: The method is used to execute the system according to any one of claims 1 to 8, and the method comprises the following steps: By building, training, and tuning a convolutional neural network model, it can automatically identify and classify abnormal conditions of the line from pre-processed images, including equipment damage data, foreign object intrusion data, and temperature abnormality data; Acquire meteorological parameters in real time, including temperature, humidity, wind speed, precipitation, air pressure and radiation; analyze the impact of meteorological parameters on the line operation status based on the meteorological parameters to obtain meteorological analysis results, including the impact of temperature, humidity and wind speed on the line operation status; Real-time monitoring and identification of the vibration modes and characteristics of three-span line conductors to assess the impact of vibration on line operation status; The equipment damage data, foreign body intrusion data and temperature anomaly data are collated to form an abnormal condition data set, which is used as the boundary condition for finite element analysis. The meteorological analysis results obtained based on meteorological parameter analysis are summarized to form a meteorological analysis data set, which is used as the environmental load for finite element analysis. The vibration mode, frequency and amplitude of the conductor are collated to form a vibration analysis data set, which is used to simulate the dynamic behavior of the conductor in the finite element. A finite element model of the three-span line is established, and finite element analysis is run to simulate the response of the three-span line under different meteorological conditions and vibration loads to identify the key areas and factors that cause line failures. According to the key areas and factors that cause line failures, the safety status of the line is evaluated and the potential fault type is determined.

10. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in claim 9.

Images